You'll learn how to set the most you can pay for a lead and cut the spend that never becomes pipeline. Then you'll get Google bidding for closed deals instead of form fills. Every stage comes with the math and a calculator for your own data.
Google Ads was built around a transaction where the searcher is the purchaser, the pixel event is the money, and it resolves in days. B2B violates that shape in 3 places, and every technique here closes one of them.
The conversion is not the outcome
In ecommerce the conversion is the revenue. In B2B it is a request to talk, worth anywhere from zero to the full contract. A competitor's product team, a student, a job seeker, and a buyer with a signed budget all fire the same pixel.
Smart bidding produces more of whatever you report, so reporting every form fill as equal makes it find the cheapest. The symptom is conversions rising and cost per lead falling while sales reports a flat pipeline.
4 people, 1 pixel
A competitor's product team
A student
A job seeker
A buyer with a signed budget
04conversions Google counts
1buyer sales can close
The buyer is not one person
Audience targeting models one person making one decision. A B2B purchase involves several arriving at different times, and often only one ever clicks an ad.
Attribution therefore undercounts and overcounts at once. Last click gives paid search no credit for awareness it created months earlier, and multi-touch distributes credit across 4 captured touchpoints when 11 existed and 7 happened in meetings. There is also no segment for a company with approved budget and an incumbent contract expiring next quarter.
One purchase, several people
ChampionClicked the ad
User
Manager
Finance
IT and security
Arriving over months →
4 touchpoints attribution captured7 happened in meetings
The clock runs longer than the optimization loop
Every default was set for short cycles.
Google's clock against a B2B deal, in days
Learning phaseAssumes enough conversions to calibrate
2 weeks
2 weeks
Conversion windowAssumes the outcome lands within a month
30 days
30 days
Click ID expiryAssumes the outcome has happened
90 days
90 days
A B2B deal closesWhat B2B actually does
60 to 180 days
60 to 180 days
0306090120150180
The first 2 are settings you change. The third is not, and it does the most damage, because the largest deals take longest and drop out first. In the learning phase, a B2B account has a handful of leads and no pipeline data at all.
The human version is worse than the technical one. A rule of thumb I hold to is that a campaign needs one and a half sales cycles before you can read it on pipeline. At a 60-day cycle that is 90 days. Most are killed at 14, which is a verdict delivered before the evidence exists.
Find which gap is costing you most
Most accounts have all 3. The symptom names which one dominates.
Most Google Ads advice arrives as a list with no order, and the order decides whether any of it works. An account that rebuilt its landing pages, switched bidding strategy, and added 200 negatives in the same month has learned nothing about which of the 3 worked.
The 7 stages
1
Measure
What counts as a good outcome, and can the platform see it?
✓Done when CRM outcomes flow back and the 2 systems agree within 15%
Operate is the exception. Weekly search term review starts as soon as there is spend to review, but a formal cadence has to wait until the account stops changing shape every week.
The 4 inversions that cost the most
Prune before Measure
Measure→beforePrune
Spend falls, efficiency improves, pipeline quietly shrinks
A headline test runs 6 weeks, variant B wins, then loses over the next 6
✓The right order: Measure → Message
Scale before Automate
Automate→beforeScale
Budget triples, cost per lead rises more than proportionally, the channel gets declared unscalable
✓The right order: Automate → Scale
The first one leaves no evidence. A wrongly blocked query produces no impressions, no clicks, and no line in any export, so pruning against the wrong outcome is invisible and permanent. That is why Measure sits first rather than second.
03
Stage · Measure7 min
Chapter 3: The audit
The instinct on inheriting an account is to fix what you can see, and changing any of it in week one destroys the only clean baseline you will ever have.
The audit produces a frozen record of the starting state and one number that funds everything else. That number is the total cost of search terms that got clicks and produced nothing.
Week one, read only
Pull these 5 before forming any opinion.
1
Search terms report
Detail90 days, every Search campaign, with match type, cost, conversions, conversion value
What you are looking forPatterns, not individual queries. 30 days in a 60-day cycle contains almost no completed outcomes
2
Every conversion action
DetailPrimary or secondary, attribution model, conversion window, counting method
What you are looking forMore than one primary action, anything created over a year ago that nobody can explain
3
Platform vs CRM counts
DetailLast month, both systems side by side
What you are looking forA gap above 15%
4
Impression share
DetailLost to budget and lost to rank, by campaign
What you are looking forWhether the constraint is money or relevance
5
Campaign and account settings
DetailThe 2 tables below
What you are looking forDefaults and automations that quietly cost money
The 6 settings that default badly for B2B
Setting✕Default✓Correct for most B2B
Setting: Search Partner Network
✕Default: On
✓Correct for most B2B: Segment and evaluate before deciding
Setting: Display expansion on Search
✕Default: On
✓Correct for most B2B: Off
Setting: Location targeting
✕Default: Presence or interest
✓Correct for most B2B: Presence
Setting: Automatically created assets
✕Default: On for many campaign types
✓Correct for most B2B: Off until reviewed
Setting: Conversion window
✕Default: 30 days
✓Correct for most B2B: 60 or 90 days
Setting: Campaign-level broad match
✕Default: Varies
✓Correct for most B2B: Check it, because of the September 2026 AI Max migration in Chapter 9
Location targeting causes the most unexplained volume, because the default lets someone in another country searching about your market trigger your ad.
Check what Google is changing without asking you
Auto-apply lets Google alter your account while nobody is watching, and it is the first thing I check on an inherited account. Settings sit under Recommendations, then the Auto-apply tab. There is no master switch, and Google changes which recommendations are eligible, so this is a recurring check rather than a one-time setting.
Verdict for B2BAuto-apply categoryWhy
Verdict for B2B: OffAdd broad match keywordsWhy: You start serving on queries nobody approved, and spend rises while conversion rate falls
Verdict for B2B: OffChange bidding strategyWhy: Hands the algorithm control before you know the conversion signal is clean
Verdict for B2B: OffAdd or remove keywordsWhy: Keyword decisions depend on closed-won data the platform cannot see
Verdict for B2B: OffExpand to Display or partnersWhy: Adds inventory that reports blended with Search
Verdict for B2B: OffAdd automatically created assetsWhy: Generates copy you did not write, and triggers the September 2026 migration
Verdict for B2B: Review, usually applyUpgrade conversion tracking, fix broken trackingWhy: Hygiene rather than strategy
Verdict for B2B: Review, never autoAdd negative keywordsWhy: Google's suggestions ignore the 11 categories in Chapter 5
Optimization score is not a performance score
Optimization score measures whether you have reviewed recommendations, not whether the account works. Dismissing gives the same uplift as applying, so an account can dismiss every suggestion and sit at 100%.
That matters because it is the number a rep opens a quarterly review with.
0%100%
Optimization score
Every suggestion dismissed, none applied
Pipeline over the same monthsNo change
✓Verified 20 August 2026 against Google Ads Help documentation.
Why the platform and CRM disagree
CauseHow it shows✓Fix
A conversion action firing site-wide
How it shows: Platform count far exceeds CRM
Fix: ✓Remove the action or scope the trigger
Newsletter or footer form counted as a lead
How it shows: Platform exceeds CRM, no matching CRM records
Fix: ✓Reclassify as secondary
Duplicate firing on resubmission
How it shows: Platform inflated, same email repeated in CRM
Fix: ✓Change counting from every to one
Bot and spam form fills
How it shows: Platform count high, CRM records with no real company domain
How it shows: Both counts plausible, neither matches
Fix: ✓Align definitions before changing anything
The audit scorecard
12 calculations from the ad platform and the CRM. Each one either clears or names the next piece of work.
Funnel economics
All of it comes out of the CRM, which is what makes the audit runnable in an afternoon.
What to calculate
Formula
What it tells you
Lead to SQL rate
SQLs / leads, trailing 12 months
How much of what you buy survives first contact
SQL to win rate
Closed-won / SQLs, trailing 12 months
Whether the problem is lead quality or sales execution
Lead to won rate
Lead to SQL x SQL to won
The number that converts spend into customers
Allowable CAC
ACV x target CAC-to-ACV ratio
The ceiling the business will fund
Maximum CPL
Allowable CAC x lead to won rate
The only CPL figure that means anything for you
The CAC-to-ACV ratio is a business decision, not a benchmark. Whoever owns the P&L states it, and the audit records who did.
Both inputs are yours and both move the ceiling proportionally. On identical contract value, a conservative ratio on a weak funnel produces a ceiling several times smaller than an aggressive ratio on a strong one. That is why a borrowed CPL benchmark is worthless. Any audit reporting a ceiling without stating the funnel rate and the ratio behind it has reported an opinion.
Formula: Closed-won deals traceable to a click ID, out of 10
✓Healthy: 8 or more
✕Investigate: Under 8
What to calculate: Waste rate
Formula: Zero-conversion spend / non-brand spend
✓Healthy: Under 15%
✕Investigate: Over 25%
What to calculate: Actual CPL against ceiling
Formula: CPC / conversion rate, versus maximum CPL
✓Healthy: Under the ceiling
✕Investigate: At or above it
What to calculate: Campaign viability
Formula: (30 / conversion rate) x CPC
✓Healthy: Every campaign's budget clears it
✕Investigate: Any campaign below it
What to calculate: Budget headroom
Formula: Conversions x (IS lost to budget / current IS)
✓Healthy: Near zero
✕Investigate: Anything meaningful
What to calculate: Cost per SQL
Formula: Paid search spend / SQLs from paid search
✓Healthy: Close to cost per lead
✕Investigate: Far above cost per lead
Budget headroom is a ceiling rather than a forecast, because the auctions you are missing are the marginal ones. Expect to land below it.
The number that constrains every structural decision
The campaign floor is independent of your economics and comes from conversion volume alone.
=Campaign floor = (30 / conversion rate) x CPC
The answer is the monthly budget a single campaign needs to sustain independent optimization, and it constrains every structural decision in Chapter 6.
Campaign count is decided by your category and conversion rate, not by the size of your budget.
30conversions a month
÷÷ 4%conversion rate
== 750clicks
×× €12cost per click
== €9,000campaign floor a month
Example figures, the same ones the calculator in Chapter 6 starts with.
Week 2, quantify the leak
Filter the export to queries with at least one click and zero conversions, then sum the cost. That is your recovery ceiling rather than a promise, because some of those terms are legitimate and the false-positive check in Chapter 5 protects them.
Then n-gram the filtered pool, weighted by cost. A 14,000-row export usually collapses to 40 or 50 recurring problem words, and everything in Chapter 5 builds from that list.
14,000rows in the export
→n-gram, weighted by cost
40 to 50recurring problem words
The audit fits on one page
7 elements, and anything that does not fit on the page did not belong in the audit.
Freeze the baseline with the date range stated. Competitor analysis, keyword expansion, and creative recommendations do not belong in an audit. The first changes nothing about Monday and the other 2 depend on work that has not happened.
Put recoverable spend at the top of the page. It is the one number that funds everything else, and it is the line whoever approves the budget reads first.
Write the order of work as a sequence, not as a ranking by effort. Measurement fixes go first even when they look small, because every later change is measured against the data they make trustworthy.
The audit, on one page
1Recoverable spendA single euro figure, 90 days, with waste rate as a percentage
2Scorecard resultsAll 12 calculations, with the ones outside range flagged
3Measurement gapsPlatform versus CRM, with the cause named
4Automations foundAuto-apply categories currently enabled
5Settings to change nowThe short list from week one
6Order of work, next 60 daysSequenced, not ranked by effort
7Baseline metricsFrozen, dated, for later comparison
When the account already looks healthy
A clean account hides its problems behind good averages. Pull 10 closed-won deals from the last quarter, trace each back to a click identifier, and confirm it appears in the offline import.
Trace 10 closed-won deals
How many of the 10 trace back to a click ID? Pick the number.
Under 5 trace
The forms are not capturing the identifier, so fix that first.
5 to 7 trace
Capture is partial, so check multi-step forms and third-party booking tools.
8 to 10 trace
Measurement is operational, so move to Chapter 5.
04
Stage · Measure12 min
Chapter 4: Measurement
Bidding, budget pacing, audience expansion, and the learning phase all run downstream of how you define success. In B2B that definition is almost always a form submission, and everything between it and a signed contract happens in a CRM the platform cannot read.
What you are seeingWhat it actually means
More than one conversion action set to primary
What it actually means: →You have told Google 2 different events are interchangeable
Conversion window still at 30 days
What it actually means: →Your slower, larger deals are excluded from reporting by design
Form volume high, CRM full of records with no real company domain
What it actually means: →Bots are training your bidding
Largest deals never appear in the offline import
What it actually means: →Click identifiers expired before the deal closed
Remarketing lists shrank after 15 June 2026
What it actually means: →The account was relying on Google Signals rather than correct consent capture
Only one event should drive bidding
Google uses primary conversion actions for bidding. Secondary actions are recorded and reported but do not influence optimization.
New actions default to primary, so accounts accumulate them. Run one primary action per campaign objective and make everything else secondary.
PrimaryDrives bidding
Demo or trial request
Why: The action the campaign exists to produce
Qualified lead, from CRMonce flowing
Why: Replaces the form fill as the target
Phone call over 60 seconds
Why: A genuine buying action in most categories
SecondaryReported only
Content download
Why: Useful signal, not a buying action
Webinar registrationor primary on a dedicated campaign
Why: Depends what the campaign is for
Pricing page view
Why: Strong intent, not an outcome
Chat started
Why: High volume, inconsistent quality
Newsletter signupor removed
Why: Almost never belongs in bidding
3 settings sit alongside this and get set once at launch, then never revisited.
Setting✕Default✓Correct for B2BIf wrong
Setting: Counting method
✕Default: Every
✓Correct for B2B: One
If wrong: Trains bidding toward people who submit forms repeatedly
Setting: Conversion window
✕Default: 30 days
✓Correct for B2B: 60 to 90 days
If wrong: Excludes your slower, larger deals by design
Setting: Attribution model
✕Default: Data-driven where volume allows
✓Correct for B2B: Either, since it matters less than the definition
If wrong: Below roughly 300 conversions and 3,000 clicks monthly, data-driven behaves like last click regardless of the label
Stop bots before they train your bidding
Counting raw form fills teaches the algorithm to find more of whatever is filling your form. In an account with a bot problem, the bidding model is hunting bots.
Work in this order, because each step reduces the volume the next has to handle.
1Clean the traffic source.
Segment Search Partners, audit Performance Max placements, tighten match types, set location targeting to presence
What it fixes: Removes a common origin before it reaches the form
2Harden the form.
Bot management, submission timing checks, real-time email and phone validation, one open context question
What it fixes: Stops automated submissions at the point of entry
3Validate before the CRM.
Tag every lead valid or spam at intake
What it fixes: Keeps the CRM count honest, which keeps Chapter 3's variance check honest
4Report only validated leads back to Google
What it fixes: The algorithm learns from real people rather than from scripts
Steps 1 through 3 clean your data. Step 4 cleans the training signal, and until it happens the bidding model keeps optimizing toward the population you just filtered out of your CRM.
What each event is worth
Value mapping is what lets the platform prefer a good lead over a cheap one, and it is the prerequisite for every value-based bidding strategy in Chapter 8.
=Stage value = ACV x probability of reaching closed-won
Probability comes from the CRM. Count how many leads at each stage over the last 12 months eventually closed, and divide. The shape of the output matters more than the figures. In most B2B funnels the spread between a content download and an opportunity is 1 to 2 orders of magnitude, and that spread is the instruction that changes bidding behavior.
What each event is worth rises at every stage
1Content download
Where the probability comes from: Downloads that eventually closed, over 12 months
2Demo request
Where the probability comes from: Demo requests that eventually closed
3Sales accepted
Where the probability comes from: SALs that eventually closed
4Opportunity created
Where the probability comes from: Opportunities that eventually closed
5Closed-won
Where the probability comes from: Actual deal value, no calculation needed
Multiplying every row by gross margin lowers the absolute figures and leaves the ratios untouched, so it is optional rather than necessary.
2 ways this goes wrong. Values that flatter the funnel collapse the ratio. If your download value sits within one order of magnitude of your demo value, you have told the algorithm those events are close to interchangeable.
And values set once and never recalculated drift, because stage probabilities move every time the sales process or the ICP changes. Recalculate quarterly.
Adjust value without a CRM connection using value rules
Conversion value rules adjust a conversion's value as it fires, by location, device, or audience. They are the bridge for an account that cannot yet import CRM outcomes.
If closed-won data shows one segment closing at several times the rate of the rest, a value rule on that audience expresses the difference without waiting on an engineering queue.
2 conditions. The multiplier comes from CRM history rather than intuition, and the rule carries its source and a review date. A rule set on a since-changed close rate bids confidently toward the wrong segment.
Value rules are scaffolding. Once offline import is running, real outcome data outranks them and most rules get retired.
The data that constrain everything else
2 figures follow from the same inputs and belong in the account documentation before any bidding target is set.
=Allowable CAC = ACV x target CAC-to-ACV ratio
=Maximum CPL = allowable CAC x lead-to-won rate
Both figures come from Chapter 3 and neither has a default. The ratio is a decision the P&L owner makes and the lead-to-won rate is a fact your CRM already holds.
Compare the ceiling against what the account actually produces.
=Cost per lead = CPC / landing page conversion rate
Under the ceiling is headroom to bid harder or reach further. At or above it, the account is buying leads it cannot afford, and no campaign structure fixes that. The 2 levers are the click price and the page, in that order of difficulty.
Getting outcomes back to Google
When someone clicks an ad, Google appends a click identifier to the landing page URL. The form captures it as a hidden field and stores it against the CRM record. As the lead progresses, those stage changes go back to Google with values attached.
Capture, persistence, storage, and upload all have to hold. A break anywhere produces the same symptom, which is an import that runs successfully while containing almost nothing.
Check which import path you are on before anything else
2 2026 changes break accounts built before them, and nothing in the interface announces either.
Google now classifies GCLID-only import as the legacy path. Enhanced Conversions for Leads is the same import supplemented with hashed first-party data, so Google can match on email as well as on the click identifier.
The harder one is the API change. From 15 June 2026 these uploads moved to the Data Manager API and are blocked in the Google Ads API. Developer tokens that had not sent a request between January and June 2026 were not allowlisted, so UploadClickConversion requests from those tokens now fail.
If your setup is
Status
What to do
A custom integration using UploadClickConversion, untouched since before 2026
Possibly failing already
Check the import status page today. Move to the Data Manager API
GCLID-only import, working
Legacy but functional
Add hashed first-party data. Keep sending GCLIDs
HubSpot or Salesforce native connection
Current
Confirm it routes through Data Manager
Zapier, Stape, or a similar connector
Usually current
Confirm the connector has been updated for Data Manager
Nothing built yet
Start here
Build Enhanced Conversions for Leads through Data Manager. There is no reason to build the legacy version and then upgrade it
From April 2026, enhanced conversions for web and for leads became a single setting and existing users were migrated automatically.
Send both match keys rather than choosing. The click identifier attributes to a specific ad click and hashed first-party data recovers matches it missed, so accounts treating this as a choice give up half the match rate.
✓Verified 20 August 2026 against Google Ads Help and the Google Ads API documentation.
Capture all 3 identifiers
GCLID
CoversStandard clicks
If missedNothing works at all, so this one gets noticed
GBRAID
CoversiOS app-to-web under privacy restrictions
If missedMobile leads go missing from the import silently
WBRAID
CoversiOS web-to-web under privacy restrictions
If missedThe same loss, and nothing in reporting shows it
The second 2 are the dangerous ones. The import still runs and the data still look plausible.
Persistence is where implementations break
Redirect to a subdomain
Why it breaks: Parameter dropped in transit
Form posting to a different domain
Why it breaks: Hidden field never populated
Multi-step form across page loads
Why it breaks: Value lost between steps
Chat widget or third-party scheduler creating the record
Why it breaks: The form never sees the identifier
Visitor leaves and returns days later
Why it breaks: No first-party cookie holding it
Set the cookie duration to match your conversion window. A 90-day window with a 30-day cookie is a system built to lose its own data.
The 90-day expiry
Click identifiers expire after 90 days and uploads against an expired identifier are rejected. This fails in the most expensive direction available, because enterprise deals take longest and are worth most. An account uploading only closed-won will systematically train bidding on smaller, faster deals.
Under 60 days
Closed-won at actual value
When: On close
60 to 90 days
Sales accepted or opportunity, at stage value
When: On stage change
Over 90 days
Latest stage reliably occurring inside 60 days
When: Within 60 days of the click
Uploading an opportunity at its probability-weighted value rather than waiting for closed-won loses some precision and keeps the signal. That is the correct trade.
Enhanced Conversions for Leads softens this without solving it. Hashed first-party data gives Google a second way to match when the identifier has gone stale, but the conversion window still applies. Send the milestone inside the window regardless.
Set a consent property deliberately rather than accepting the default
Salesforce
Data Manager connection for opportunity stages and revenue
Map opportunity stage rather than lead status, or you import the wrong event
Pipedrive, Close, in-house
Zapier or Stape on stage change, through the Data Manager API
Confirm the connector moved off the Google Ads API
Anything else
Scheduled CSV through Data Manager
Somebody has to own checking the file arrived and the rows were accepted
Upload daily rather than weekly. Sync failures are silent and look identical to a bad month. Check the import status page weekly for rejection rate rather than acceptance count, and alert on upload volume dropping.
Verify before trusting it
Pull 10 closed-won deals from the last quarter. Trace each back through the CRM to a click identifier, then confirm that identifier appears in the offline import.
8 to 10 trace
The import is working, so move on to the automation chapters
5 to 7 trace
Capture is partial, so check multi-step forms and third-party booking tools
Under 5 trace
The forms are not capturing the identifier, so fix that before touching the import
Repeat quarterly and after any change to forms, landing pages, or the CRM.
What changes once it runs
Cost per conversion rises, often sharply, because the account now counts qualified leads instead of form fills. Expect a multiple rather than a percentage, and expect it to be roughly the inverse of your lead-to-SQL rate. The economics did not change, the definition did, and bidding targets have to be rebuilt against the new one.
Conversion volume falls. Explain that before it appears in a report. Chapter 11 covers how.
Consent and measurement in Europe
A broken consent configuration loses conversion tracking, stops building remarketing audiences, and degrades bidding data, with no error message. Consent Mode v2 has been mandatory since March 2024, enforced from 21 July 2025.
Parameter
Controls
Breaks without it
ad_storage
Advertising cookies and identifiers
Conversion tracking, remarketing, cross-device
ad_user_data
Sending user data to Google for measurement
Enhanced conversions, silently
ad_personalization
Personalized ads and remarketing
Audience building
analytics_storage
Analytics collection
GA4 measurement
A setup carrying only ad_storage and analytics_storage is a v1 configuration, and it is the most common defect in accounts that believe they are compliant. Use Advanced mode rather than Basic. Basic blocks tags until the visitor interacts with the banner, so declines produce nothing recoverable.
The June 2026 change most accounts have not audited
On 15 June 2026 Google retired Google Signals as a control over data flowing from Analytics into Google Ads. The ad_storage parameter is now the sole control, with no fallback and no warning if the banner fires it incorrectly.
3 checks on any account serving European traffic. Confirm all 4 parameters appear in the tag payload. Compare remarketing list sizes either side of 15 June 2026, since a sharp drop means the account was leaning on Google Signals. And confirm the consent platform fires before the Google tag, because a green status in Tag Assistant does not prove ordering.
✓Verified 20 August 2026.
Lawful basis for offline upload
Sending hashed customer identifiers to Google needs a lawful basis and a disclosure. 3 things have to exist:
A privacy policy disclosing the transfer in specific terms.
A record of processing with Google's data processing terms in place.
The ad_user_data signal reflecting the visitor's actual choice.
Hashing happens with SHA-256 before transmission, so the raw email never leaves your systems, which addresses the most common objection. This is a short privacy policy addition and one conversation with whoever owns the DPA. It is the most common reason a technically ready import sits unused.
05
Stage · Prune17 min
Chapter 5: Keywords and negatives
The category head term is an awareness keyword being asked to do a conversion job and priced as though it were doing one. What separates a buyer's query from everyone else's is rarely the noun, it is the qualifier attached to it.
Negatives do the other half. Blocking a wasteful query saves the click cost and stops a junk conversion becoming a targeting instruction, which is the half that costs more.
What you are seeingWhat it actually means
Nobody can name your zero-conversion spend for the last 90 days
What it actually means: →The account has never been sized for waste
The account carries under 50 negatives
What it actually means: →The first hour of this chapter is worth more than any bid change
A keyword was paused because its traffic looked bad
What it actually means: →You removed every search term it matched, including the converting ones
Keyword expansion runs off Keyword Planner volume
What it actually means: →You are expanding on what is searched rather than what closes
Broad match is on and the negative library is thin
What it actually means: →Every widening the platform applies lands in an account with no filter
Keywords and search terms are not the same thing
Half the keyword mistakes in B2B trace back to treating these as one object.
Keyword
Search term
What it is
What you bid on
What the user actually typed
Who wrote it
You
The user
Where you see it
Keywords tab
Search terms report
What negatives block
Not this
This
Reporting
Complete
Incomplete, because low-volume queries are suppressed for privacy
Count
Dozens
Thousands per keyword
4 consequences follow.
A keyword's performance is an average
What it means in practice: One keyword can match hundreds of search terms. Its cost per conversion tells you nothing about which of them worked
You optimize keywords and diagnose from search terms
What it means in practice: Pausing a keyword removes every search term it matched, including the ones converting
Adding a keyword does not guarantee you serve on it
What it means in practice: Adding a negative does guarantee you do not. Negatives are the only absolute control in the account
You cannot negate what you cannot see
What it means in practice: Suppressed low-volume queries never appear in the report, which is the argument for blocking patterns rather than individual terms
Qualifiers do the filtering that match types cannot
✕Raw term✓Qualified variantWhy it filters
✕Raw term: scheduling software
✓Qualified variant: employee scheduling software for multi-site
Why it filters: A structural condition only a business has
✕Raw term: rota planner
✓Qualified variant: rota planning software for retail chains
Why it filters: Industry plus product category
✕Raw term: shift management
✓Qualified variant: shift management system for care homes
Why it filters: Regulatory context implies a business buyer
✕Raw term: workforce app
✓Qualified variant: workforce scheduling platform for operations teams
Why it filters: Names the buying function
Qualifiers that work describe an operational reality. Multi-site, enterprise, for teams, API, SSO, and compliance language are words a consumer has no reason to type. Best, top, professional, and advanced appear in consumer queries at the same rate, so they cut volume without improving qualification.
Progress match types as evidence accumulates
Match type
Use it to
Prerequisite
What you give up
Exact
Establish which terms convert and at what cost
None
Volume ceiling, misses long-tail variants
Phrase
Expand around terms already proven at exact
A converting exact-match base
Some query control
Broad
Reach demand your keyword list cannot describe
Smart bidding, full negative library, 30+ monthly conversions
Query control almost entirely
Google keeps pushing accounts toward broad match through defaults and automatic upgrades, and the accounts that survive it have the negative library already built. Running all 3 match types for the same term in one ad group makes them compete, and the reporting stops attributing the result.
Where good keywords actually come from
Closed-won reverse lookup
What to doMatch closed deals to their originating keyword, rank by pipeline value rather than volume
What it findsVertical-qualified terms producing a fraction of traffic and a multiple of revenue
Search terms report
What to doWeekly, pull converting queries you are not explicitly bidding on
What it findsLong-tail commercial terms Keyword Planner never surfaces
Competitor keyword sets
What to doSort by how long a term has been bid on rather than by volume
What it findsTerms a competitor has run 6 months, which means they are profitable
When the CRM cannot track keywords
Work down this ladder and stop at the first tier you can actually build.
Tier
Method
Granularity
Effort
1
Click ID stored in the CRM, resolved by Google in the offline conversion report
Keyword
No CRM schema change. Google does the resolution
2
The utm_term parameter via ValueTrack into a hidden form field
Keyword, until a redirect drops the parameter
One form change, one URL template change
3
One landing page per ad group, attribute by page
Ad group
Page build work, no tracking work
4
A required "how did you hear about us" field, correlated to campaign timing
Campaign
Trivial, but the data is messy
5
No outcome attribution available
None
Read the next section
Build tier 1. Tier 3 is an acceptable stopping point, because knowing which ad group outproduces the others directs expansion as well as knowing which keyword did.
Operating on platform data alone
Some accounts cannot get outcomes back at all. A locked CRM, no engineering time, a sales process that runs on email. The answer is to change what you trust and add a manual loop rather than to guess or to stop optimizing.
Platform signal✓Trust it for✕Do not trust it for
Platform signal: Search terms report
✓Trust it for: Finding waste and finding expansion terms
✕Do not trust it for: This one is complete, so nothing else is needed
Platform signal: Impression share lost to budget and rank
✓Trust it for: Deciding where growth is available
✕Do not trust it for: This one is complete too
Platform signal: CPC and CTR
✓Trust it for: Auction diagnosis, ad relevance
✕Do not trust it for: Lead quality
Platform signal: Conversion volume
✓Trust it for: Relative comparison between keywords
✕Do not trust it for: Whether those conversions were worth anything
Platform signal: Cost per conversion
✓Trust it for: Directional trend over time
✕Do not trust it for: Setting a target, because the denominator is unverified
Platform signal: Value-based bidding
✓Trust it for: Do not run it at all
✕Do not trust it for: The values would be fictional
3 adjustments make platform-only operation defensible.
Set the CPL target below your ceiling
What to doRun at 60 to 70% of the maximum CPL from Chapter 3
WhyThe margin absorbs the quality you cannot verify
Add a manual sales feedback loop
What to doMonthly, review 20 recent leads with sales, tag good or bad, map back to campaign and ad group
Why20 leads a month is 240 a year, which is enough to see a pattern
Substitute holdout tests for attribution
What to doPause a keyword group for one full sales cycle, compare pipeline created in the following cycle against the prior one
WhyCrude, slow, and real evidence rather than an assumed model
Attribution answers what caused a result and holdouts answer what a thing is worth, so when attribution is unavailable the second question is the one that matters. Run one at a time, on a group large enough to move the number and small enough that pausing it is survivable.
The 3 calculations that settle keyword arguments
What a keyword is allowed to cost
=Maximum CPC = maximum CPL x that keyword's conversion rate
=Break-even conversion rate = CPC / maximum CPL
The same relationship reads in both directions, and between them they turn a keyword list into a pass and fail column.
Divide each keyword's cost per click by your maximum CPL to get the conversion rate it must hit. Sort the list by actual conversion rate, draw the line at that figure, and everything below it loses money at any volume.
Read the other direction for what a keyword is allowed to cost. A keyword converting at half your account average supports half the click price.
This also settles the head term question. Divide a head term's click price by your demo ceiling and you get a rate it will never reach. Divide it by the acceptable cost of a benchmark download and you get one it reaches. The keyword was fine, the offer attached to it was wrong.
What the negative keyword work is worth
=Recovered leads = recoverable spend / actual CPL
=Recovered pipeline = recovered leads x lead-to-won rate x ACV
Run it with your recoverable spend figure from Chapter 3.
Recovered spend is not a cost saving, it is budget moving from queries that cannot convert to queries that can. In a strong funnel an afternoon of negative keyword work converts into pipeline worth several times the spend recovered, and stating it that way is what gets the work approved.
Whether a qualifier earns its narrowing
Metric
Head term
Qualified variant
Which wins
Monthly clicks
Higher
Lower, often by 70%
Neither, since volume is not the test
CPC
Higher
Usually lower
Neither
Cost per lead
Often lower
Often higher
Neither, and using it reverses the answer
Cost per SQL
Read both
Read both
The only one that decides it
A qualified term producing a third of the leads at twice the cost per lead is winning if its cost per SQL is lower. Judging it on cost per lead is how a working qualifier gets paused.
How negatives match
Match type
Syntax
Blocks when
Use for
Broad
job
Every word appears anywhere, in any order
Words that disqualify on their own
Phrase
"shift planning jobs"
The words appear together in that order
Multi-word intent where individual words are legitimate
Exact
[free rota software]
The query matches exactly
Protecting a valid term while blocking one bad variant
Default to phrase when unsure, because broad negatives are order-independent and easy to misfire.
Google handles misspellings and capitalization, not plurals, stems, synonyms, or accented characters, so rota will not block rotas. Negatives beyond the sixteenth word of a query do not apply, and a negative always overrides a positive keyword for the same term, which explains most unexplained keyword outages.
Where each list belongs
Level
Reach
Limit
What belongs here
Account
All Search and Shopping, including future campaigns
1,000
Universal disqualifiers only
Shared list
Any campaign you attach it to, edits propagate
20 lists, 5,000 each
The bulk of the library, one per category
Campaign
One campaign
Up to 10,000
Navigational terms, brand exclusions
Ad group
One ad group
Shares the campaign pool
Traffic sculpting only
✓Verified August 2026. Negatives govern Search and Shopping only. Display and YouTube need placement exclusions.
The 11 categories
The deployment rules matter more than the terms, because the terms are account-specific and the rules are not.
Category
Match
Deploy at
Skip or adjust if
Informational and research
Broad
Shared list, MOFU and BOFU only
You gate lead magnets. Remove template, checklist, worksheet
Free and low budget
Broad and phrase
Account
You have a free tier. Never add free as a standalone broad negative
Jobs and careers
Broad
Account
This has the lowest false-positive risk of any category
Education and certification
Broad
Account
You sell training. Move training and certification to phrase
Consumer and personal use
Broad and phrase
Account
Check nonprofit, church, and school against closed-won first
Developer and self-hosted
Phrase and broad
Shared list
You offer self-hosting. Keep spreadsheet, excel, and sheets out
Competitor navigational
Broad
Competitor campaigns only
Never at account level. These appear in your own brand queries
Wrong industry and adjacent
Phrase and exact
Shared list
No template exists. Build it from your n-gram output
Agency, reseller, affiliate
Phrase
Shared list
Agencies are in your ICP. Also check rfp and tender
Investor, analyst, press
Phrase
Shared list
Never negate gartner, forrester, g2, or capterra
Your own brand navigational
Phrase
Brand campaign only
Protect pricing, demo, and free trial
The last 3 are the ones most published lists miss. Own brand navigational is the one almost nobody blocks, where existing customers searching your brand plus a support term click a paid ad above your own organic result.
Grouping the 11 categories into a deployable structure
The 11 categories answer one question: what kind of query is this. They don't answer a different question that matters just as much: how many lists should exist, and what someone will call each one in a review 6 months from now. Collapsing both into one step is how an account ends up with 11 files nobody can name from memory.
Classify against all 11. Deploy against far fewer, grouped by how the terms actually attach and get talked about, not by which category produced them.
Merge by deployment behavior, not by topic
Group
Built from
Deploy at
Account-level (not a list)
Jobs and careers + Consumer and personal use + Education and certification + Free and low budget
Account
Informational Intent
Informational and research
Shared list, MOFU/BOFU only
Off-ICP
Wrong industry and adjacent + Developer and self-hosted
Own brand navigational, minus pricing, demo, and free-trial terms
One list, attached to both the brand campaign and non-brand campaigns
4 categories carry no per-campaign nuance. Every query in them disqualifies, whichever campaign serves it. Those go straight into the account's own negative keywords, not a named list.
2 pairs of categories share nothing topically but share how they get handled. Wrong-industry and developer/self-hosted both describe a real evaluator who isn't a fit, so they take the same list. Agency/reseller/affiliate and investor/analyst/press don't describe the same buyer at all, but both carry the same warning: some of these terms sit inside the ICP. So they share a list that gets reviewed manually, never deployed on sight.
Naming convention
NEG | \name uses a pipe separator and a name a team would say out loud in a meeting. "Off-ICP" beats "Wrong-Industry-And-Adjacent." If a name needs a parenthetical to explain what it means, it's the wrong name.
Split one category in 2 when campaign structure demands it
Informational intent behaves differently depending on how the account is built. Where TOFU is its own campaign, exclude the category in full from MOFU/BOFU. Where TOFU, MOFU, and BOFU share one campaign (usually a volume constraint, not a choice), full exclusion costs too much. It strips the only top-of-funnel entry point the account has.
Version
Contains
Use when
Full
Every informational/research term
TOFU runs as its own campaign
Blended
Same list, minus terms carrying tool or template intent (contains words like template, builder, dashboard, tracker, software)
TOFU, MOFU, and BOFU share one campaign
A query like "kpi dashboard template" sits closer to evaluation than "what is a kpi" does. The blended version exists so a volume-constrained account doesn't lose the former on its only funnel entry point.
Run the core-category check before any of this ships
One check catches the mistake that costs the most: negating a term that sits inside the account's own product category. A gamification company blocking "gamification." A KPI-tracking tool blocking "kpi software."
These pass every other guardrail, with real volume and a plausible non-buyer reading in isolation. They still shouldn't be blocked, because the category name and the buyer's search term are the same word.
Every flagged term gets one of 2 outcomes, not an indefinite hold:
Outcome
When
Example
Remove permanently
The term names the product category or a core feature directly
A flagged term still sitting unresolved after one review cycle means the check ran but the decision didn't. Fix that before it becomes a permanent third bucket.
Reserve names for signals that don't exist yet
Not every useful list comes from a search-intent category. A list of queries with real impression volume and zero clicks is a performance signal, not a classification tag. It needs a live feed to populate, not a static export.
Name it and document what it will hold before the feed exists. The account then has a defined home for that data on day one, rather than a rebuild under a different name later.
NEG | Non-Engaged Impressions holds queries serving with meaningful impressions and no clicks. It stays empty until a recurring pull feeds it.
Once it is populated, a zero-click query still isn't automatically a negative. It can mean the query is irrelevant, or that the ad isn't landing for a relevant query, which is a message problem rather than a targeting one. Route every candidate through the plausible-buyer check before it becomes a negative.
Build the rest from your own data
1
Export 90 days of search terms, every Search campaign
Output: Raw pool, often 10,000 rows or more
2
Filter to clicks with zero conversions, sum the cost
Output: Your recovery ceiling, in euros
3
N-gram pass weighted by cost, 1 to 3-word phrases
Output: 40 to 50 recurring problem words
4
Classify each n-gram against the 11 categories
Output: Categorized candidate list
5
Choose the narrowest match type that works
Output: Match type assigned
6
Run the false-positive check against every converting term
Output: Candidates downgraded or dropped
7
Upload through Google Ads Editor
Output: A reviewable diff before anything goes live
8
Log each addition with date, level, source query, reason
Output: An audit trail you can reverse
Anything fitting no category is either wrong-industry waste or a targeting gap, and the second is often worth more.
Avoid blocking your own buyers
Over-negating leaves no evidence, because a blocked buyer produces no impressions, no clicks, and no line in any export.
Converting term
The questionDoes any converting term contain this word?
RuleIf yes, never add it as a broad negative
Plausible buyer
The questionCould your ICP type this while looking to buy?
RuleJob seekers never type demo. Buyers sometimes type template
Sample size
The questionIs there enough data to justify a block?
RuleBelow 20 impressions or 5 clicks, wait. Never negate a term with a conversion
Scope
The questionCategory or single query?
RuleOne query gets exact match. Leave the category alone
A broad negative on free blocks every free trial query including branded ones, and informational negatives at account level gut whichever campaign was built to capture research intent.
What goes wrong, and what it looks like
MistakeHow it shows up✓Fix
Using free as a broad negative
How it shows up: Branded free trial queries stop serving
Fix: ✓Phrase negatives on the bad combinations only
Informational negatives at account level
How it shows up: The TOFU campaign loses most of its volume
Fix: ✓Shared list, MOFU and BOFU only
All 3 match types in one ad group
How it shows up: Reporting stops attributing results
Fix: ✓One match type per ad group
Head term measured by demo conversions
How it shows up: A useful awareness term gets paused
Fix: ✓Judge it against the offer its stage can convert
Negatives added without logging
How it shows up: Nobody can safely reverse anything
Fix: ✓Log date, level, source query, reason
Qualifier measured by cost per lead
How it shows up: A working qualifier looks like a failure
Fix: ✓Compare on cost per SQL
What moves after you deploy
↓DownShare of spend on zero-conversion termsThe primary metric. Everything follows it
↓DownCost per sales-accepted leadRecovered budget reaching real buyers
↑UpConversion rateThe same offer, a cleaner audience
↓DownClick-through rate and clicksFewer curiosity clicks, which is the point
Flag the last row before anyone reads the dashboard, and expect 2 to 3 weeks of noise while smart bidding recalibrates. If lead volume falls and sales-accepted leads fall with it, you have over-negated.
06
Stage · Structure9 min
Chapter 6: Structure
Splitting by persona, region, funnel stage, product line, and match type each has a defensible rationale. Together they divide conversion volume until no campaign has enough data to optimize against.
The account that looks most organized is usually the one performing the worst.
What you are seeingWhat it actually means
Campaigns running under 30 conversions a month
What it actually means: →Smart bidding never leaves learning. Merge until each clears its floor
Brand sitting inside a non-brand campaign
What it actually means: →Your blended cost per conversion is hiding a failing non-brand campaign
Competitor terms measured by cost per lead
What it actually means: →You are measuring that campaign type on the one metric it cannot win
Ad groups drawn around keywords rather than messages
What it actually means: →Every ad in them hedges across 2 promises
A restructure planned because the account looks untidy
What it actually means: →You are about to pay a learning period for tidiness
Brand campaign running because it always has
What it actually means: →Nobody has tested whether it is buying anything you would not get free
The constraint that governs every decision
=Campaign floor = (30 / conversion rate) x CPC
=Campaigns you can run = non-brand budget / campaign floor
That second number is a ceiling rather than a target. When a campaign sits below its floor there are 4 moves, in order of preference.
Merge into an adjacent campaign
When it appliesThe 2 would share a bidding target
EffectOne learning campaign instead of 2 starving ones
Move down the conversion action
When it appliesThe campaign has a viable mid-funnel offer
EffectMore conversions at the same spend, and the model learns
Switch to Maximize Conversions without a target
When it appliesThe campaign has to stay separate for a real reason
EffectRemoves the target the campaign cannot hit
Run manual CPC
When it appliesVolume will not recover this quarter
EffectStops pretending automation has enough to work with
The fourth is a legitimate answer. A campaign producing single-digit monthly conversions on manual CPC beats the same campaign on Target CPA permanently recalibrating.
The 2 splits that always earn their place
Brand and non-brand
Why it holds at any spendBrand converts at multiples of non-brand and costs a fraction per click
What happens without itA blended cost per conversion sits comfortably within target while the non-brand half runs several times above the ceiling
Competitor
Why it holds at any spendDifferent CPCs, Quality Scores, and conversion rates
What happens without itOne CPA target applied across 2 populations that behave differently, corrupting both
Brand is the one exception to the 30-conversion rule, because separating it is about measurement integrity rather than optimization.
Whether to bid on your own brand at all
Separating brand is not the same question as running it. Most B2B accounts run brand because they always have, and almost none have tested what the spend buys. It earns its place under 3 conditions and is close to pure waste outside them.
Competitors appear on your brand terms
Because: You are defending a term someone else is buying, and the click was going to leave either way
You need to control the message
Because: Promotions, positioning changes, and pricing pages that organic does not surface
Your organic result for your own brand is weak or ambiguous
Because: Review sites, marketplaces, or an old subdomain outrank you
Outside those 3 you are paying for a click that arrives free, because the organic result sits directly beneath the ad and the searcher has already decided who they want.
Test it rather than argue. Pause brand in one geography for a full sales cycle, leave it running elsewhere, then compare total branded traffic and conversions across both. Choose a geography large enough to read and run a full cycle, because branded search lags awareness activity.
Every other split is conditional
Split✓Justified when✕Not justified when
Split: Funnel stage
✓Justified when: The offers differ and the acceptable CPA differs
✕Not justified when: You would run the same offer at the same target
Split: Region
✓Justified when: Sales is regionalized, or CPCs differ by more than 40% between markets
✕Not justified when: One team covers all markets and costs are similar
Split: Language
✓Justified when: Always, because ad delivery and Quality Score both degrade on a language mismatch
✕Not justified when: There is no case for mixing languages in one campaign
Split: Persona
✓Justified when: Ad copy and landing page differ in substance
✕Not justified when: You would write the same ad for both
Split: Product line
✓Justified when: Separate buyers and separate pages
✕Not justified when: The products share a buyer and a demo form
Split: Device
✓Justified when: Almost never
✕Not justified when: Use device exclusions instead, per Chapter 8
If the ad copy would be identical for both personas, the split only halves your conversion volume.
Ad groups are message containers
An ad group is the smallest set of keywords one ad and one landing page can serve honestly, and that definition decides every split.
Message
RuleOne ad and one page must serve every keyword in the group
Signal you got it wrongYou are writing an ad that hedges across 2 different promises
Impression volume
RuleThe group needs enough impressions for Google to rate its assets
Signal you got it wrongAssets still showing no rating after 30 days
Match type
RuleOne match type per group
Signal you got it wrongReporting stops attributing results between them
Landing page
RuleOne page per group, matched to the query theme
Signal you got it wrong2 groups pointing at the same page should be one group
The unrated asset signal is the volume test. Open asset details after 30 days, and if most assets carry no label the group is too small to learn from.
Single keyword ad groups fragment volume badly under current matching, and the message match benefit is available through responsive search ads and landing page alignment instead.
Settings that restructure the account for you
Setting✕Default✓Correct for B2BWhat the default does
Setting: Location targeting
✕Default: Presence or interest
✓Correct for B2B: Presence
What the default does: Serves to people outside your market who search about it
Setting: Display expansion on Search
✕Default: On for some types
✓Correct for B2B: Off
What the default does: Adds display inventory to a search-intent campaign
Setting: Search Partner Network
✕Default: On
✓Correct for B2B: Segment, then decide
What the default does: Adds third-party inventory that reports blended
Setting: Automatically created assets
✕Default: On for many campaigns
✓Correct for B2B: Off until reviewed
What the default does: Generates copy you did not write, and triggers the September 2026 migration in Chapter 9
Competitor terms have different economics
Higher CPCs, lower Quality Scores, and lower conversion rates are structural here, because relevance is scored against a query about someone else's product and the visitor already has a solution.
I set the competitor target CPA 40 to 60% above non-brand, then adjust against the opportunity rate. Measured by CPL against non-brand, it gets paused, and the pipeline disappears without anyone connecting the 2 events.
What to target and what to block
✓Target✕BlockWhy
✓Target: [competitor] alternative
✕Block: [competitor] alone
Why: Alternative signals active evaluation
✓Target: [competitor] pricing
✕Block: [competitor] login
Why: Pricing signals a buying committee doing diligence
✓Target: [competitor] vs [you]
✕Block: [competitor] support
Why: The highest intent query available
✓Target: [competitor] vs [other]
✕Block: [competitor] status
Why: Shortlist stage, and you are not on it yet
✓Target: [competitor] problems
✕Block: [competitor] careers
Why: Frustration is the cheapest switching signal
✓Target: switch from [competitor]
✕Block: [competitor] tutorial
Why: Intent stated explicitly
Deploy competitor navigational negatives at campaign level before launch, because most people searching a competitor's name alone are their customers trying to log in. The reviews query splits between evaluators and existing customers, so test it separately.
Legal boundaries
Bidding on competitor trademarks as keywords
Detail: Permitted in most jurisdictions and by Google policy
Using a competitor trademark in ad text
Detail: Restricted, and Google will disapprove in many cases
Dynamic Keyword Insertion on competitor campaigns
Detail: Never, because it can insert their trademark into your headline
Multi-market accounts
Detail: Check per market, because trademark rules and complaint handling vary
The page decides whether this works
The visitor already has a solution, so the page leads with migration rather than capability.
Comparison table including rows where they win
Why it earns space: A table where you win everything destroys the page's credibility
Migration timeline
Why it earns space: The fear is downtime and data loss, not price
One named customer who switched, with the time cost stated
Why it earns space: "Moved 40 sites across a weekend" is evidence. "Transitioned smoothly" is not
Pricing, or pricing structure
Why it earns space: Comparison shoppers who cannot find pricing assume bad news
An honest statement of where you lose
Why it earns space: Earns the right to be believed about where you win, and filters deals you would lose anyway
Budget for brand defense first, because competitors monitor auction insights and some respond by bidding on your terms.
When not to run competitor campaigns
Your product is worse for the segment they serve
Why: Produces demos that waste sales time and negative word of mouth
The account cannot support a separate campaign
Why: Merging it into non-brand corrupts both targets
No comparison page exists and none is planned
Why: The campaign converts badly and the keyword strategy gets blamed
When to restructure
Restructuring restarts learning across everything it touches.
✓Restructure when
Campaigns cannot reach conversion volume
The structure no longer matches how the business sells
2 starving campaigns could become one working one
✕Leave it alone when
The structure is untidy but functioning
A new person prefers a different convention
You want it to match a template
Do it in one move rather than incrementally. A single learning period across the whole change costs less than 4 sequential ones.
What goes wrong, and what it looks like
MistakeHow it shows up✓Fix
Segmenting below the volume threshold
How it shows up: Performance looks random, campaigns never exit learning
Fix: ✓Merge until each clears the floor
Competitor inside non-brand
How it shows up: One CPA target across 2 populations, both underperform
Fix: ✓Separate, target 40 to 60% higher
Persona split with identical ad copy
How it shows up: Conversion volume halves for no gain
Fix: ✓Merge unless the message differs
Brand campaign never tested for incrementality
How it shows up: Paying for clicks organic was already winning
Fix: ✓Run the geo holdout
Restructuring and changing bidding in the same month
How it shows up: Neither effect is attributable
Fix: ✓One change, one learning period
Name campaigns so the next person understands them
=Stage | Intent | Geo | Language
Campaign name
Reads as
BOFU / Non-Brand / NL / NL
Bottom of funnel, non-brand, Netherlands, Dutch
BOFU / Competitor / DACH / DE
Bottom of funnel, competitor, DACH, German
MOFU / Non-Brand / UK / EN
Middle of funnel, non-brand, UK, English
Brand / Defense / ALL / EN
Brand defense, all geographies, English
NEG / Jobs
Shared negative list, jobs category
A name is the only documentation that appears automatically in every report. In 12 months, someone who did not build the account has to understand why a campaign exists before changing it.
07
Stage · Message9 min
Chapter 7: Message
Weak B2B ads are rarely badly written, they are aimed at the wrong stage. 6 weeks of copy testing on a demo ad served to problem-aware traffic measures 2 wrong options against each other.
The ad and the page are one unit, and the query decides what that unit has to do.
What you are seeingWhat it actually means
Paid traffic lands on the homepage or one shared page
What it actually means: →You are paying for a message match and not delivering it. Build one page per ad group
Every ad ends in the same demo CTA regardless of query
What it actually means: →The offer is fixed and the intent is not. Match the offer to the stage
Lead volume rose after shortening a form, SQLs did not
What it actually means: →Form length was treated as a volume lever. It is a routing decision
Assets sit at Low for months
What it actually means: →Nobody reads asset reporting. Replace anything at Low for 30 days
Enterprise logos on pages serving mid-market queries
What it actually means: →Proof is disqualifying the visitor before they reach pricing
No pricing anywhere on the page
What it actually means: →Comparison shoppers assume bad news and leave without converting
The ad's job changes with awareness
1Problem aware
What the ad doesNames the problem more precisely than the searcher can
The offer that fitsBenchmark, calculator, diagnostic
Judge it onCost per download, not cost per demo
2Solution aware
What the ad doesExplains why this category solves it
The offer that fitsComparison, buyer's guide, webinar
Judge it onCost per registration
3Vendor aware
What the ad doesRemoves the final objection
The offer that fitsDemo, trial, pricing conversation
Judge it onCost per SQL
The test is whether the ad reads as a reasonable next sentence. "How do I reduce overtime across sites" followed by "Book a 30-minute demo" is not one.
The fourth column makes the first 3 usable. Problem-aware traffic converts to demos at a fraction of vendor-aware rates, and judging it on cost per SQL kills a working campaign.
Structure that survives the auction
Responsive search ads assemble combinations, so each asset works in isolation rather than as a sequence.
Position
Carries
Why
Headline 1
The keyword theme
Google bolds matched terms, which earns the scan before any argument does
Headline 2
Proof or the qualifier
Tells the wrong searcher to keep scrolling
Headline 3
The action
One CTA, never 2
Description
A moment the buyer recognizes
Operational specificity proves familiarity
A worked set for the query "employee scheduling software for retail".
Asset
Copy
Doing the work
Headline 1
Employee Scheduling for Retail Chains
Echoes the query, gets bolded
Headline 2
Built for 4 to 400 Locations
Filters single-site and enterprise out
Headline 3
See a 20-Minute Walkthrough
Lower commitment than book a demo
Description
Publish rotas across every site in one pass. Track overtime before it happens, not after payroll closes.
Names a monthly operational moment
"Before payroll closes" is a moment an operations director lives through. "Improve workforce efficiency" could belong to any product in any category, and that distinction is most of B2B ad copy.
Proof has to match the segment the query implies
Enterprise logos shown to mid-market intent suppress conversion. The visitor reads 3 enterprise names, concludes the product is priced above them, and leaves before reaching pricing.
Query implies✓Proof that works✕Proof that backfires
Query implies: Small or mid-market
✓Proof that works: A customer of similar size, named
✕Proof that backfires: Enterprise logo bar
Query implies: A specific vertical
✓Proof that works: A customer in that vertical
✕Proof that backfires: Generic cross-industry logos
Query implies: Competitor comparison
✓Proof that works: Migration count, switching timeline
✕Proof that backfires: Any customer list
Query implies: Unrecognized brand names
✓Proof that works: No logo bar at all
✕Proof that backfires: A logo bar of names they do not know
The last row costs the most, because unfamiliar logos occupy the best space on the page and transfer no credibility.
Working with responsive search ads
Element
Rule
Why
Headlines
8 to 12
Under 8 limits combinations, over 12 and assets never accumulate impressions
Descriptions
3 to 4
The same reason
Pinning
Position 1 only, for the keyword headline
Guarantees message match, keeps the rest testable
Asset review
Monthly, replace anything at Low for 30 days
Low assets dilute the pool
Unrated assets
Check impressions before removing
An unrated asset is not a failing asset
Assets that do real work in B2B
Sitelinks
Use it forPricing, comparison, integrations, case study
The one that surprises peoplePricing, because it pre-qualifies people out before you pay for their form fill
Callouts
Use it forObjections from sales calls, not features
The one that surprises people"No per-location fees" beats "Trusted by thousands"
Structured snippets
Use it forIntegration lists
The one that surprises peopleIn operations and finance software this is the first disqualifying question
Lead form assets
Use it forLow-commitment offers only
The one that surprises peopleRemoving friction on a considered purchase produces leads that were not considering anything
Message match is the first 2 seconds
A visitor arriving from an ad checks whether they are in the right place, and that check resolves on the headline. Say the ad promises "employee scheduling software for retail chains" and the page says "smarter workforce operations for modern businesses." The visitor is not certain they are in the wrong place. That is worse than being certain.
Use the same words rather than similar ones. Matching query language also supports Quality Score.
6 jobs the page has to do
A fixed element order is the wrong frame. B2B buyers arrive having already read reviews, comparisons, community threads, and whatever an AI assistant summarized for them. That puts them further along than a page built to convince them assumes, and further from talking to you than a demo form assumes.
1
Confirm they are in the right place in 2 seconds
Why it matters nowThe check resolves on the headline, before any argument
How it shows up on the pageHeadline uses the query's own words
2
Let them see the product without talking to anyone
Why it matters nowMost of the evaluation happens before contact. Gating the product behind a call loses everyone not ready to have it
How it shows up on the pageProduct video, interactive tour, real screenshots rather than a stock photo
3
Answer the pricing question
Why it matters nowComparison shoppers who cannot find pricing assume bad news, and they are usually right
How it shows up on the pageA price, a range, or the structure and what drives it
4
Survive being forwarded
Why it matters nowThe person who found you is often not the person who signs. The page gets pasted into Slack with no context
How it shows up on the pageSelf-contained, with no session-dependent state
5
Offer 2 commitment levels
Why it matters nowA demo is the highest-friction action on the page. Some are ready, most are not
How it shows up on the pageDemo plus one lower-friction path that still identifies them
6
Handle the objections that stall deals
Why it matters nowImplementation time, integrations, security, and per-seat versus per-site pricing are what a buying committee argues about
How it shows up on the pageAn FAQ answering the 2 your sales team hears most
The second job is where most B2B pages fail. A page describing the product in prose with a demo form next to it asks for a meeting before the buyer has seen the thing. A 90-second product recording moves more pipeline than another paragraph of benefit copy.
The fourth is rarely designed for. Buying committees mean pages travel, and a page that only makes sense to whoever clicked the ad loses the person approving the spend. 2 constraints apply regardless of layout. Evidence needs a number, and navigation counts as competing calls to action.
The page is read by machines too
Pricing pages, integration lists, and security pages are increasingly parsed by AI assistants answering buyer questions before the buyer ever searches. Clear headings, factual claims, and readable pricing improve how you are described in those answers. Chapter 9 covers the paid side of the same shift.
Form length is a routing decision
More fields means fewer leads and better ones. The deciding input is what a sales conversation costs.
VolumeQuality
2, name and email
Volume: Highest. Quality: Lowest
Right when: Content downloads, top of funnel only
4, plus company and role
Volume: Moderate. Quality: Good
Right when: Most B2B demo requests
6 or more, plus size, budget, timeline
Volume: Lowest. Quality: Highest
Right when: Sales time is the binding constraint
A qualifying field like number of locations pays for itself by routing inquiries away from a team that cannot serve them. Budget and timeline fields do not, because both attract optimistic answers, and a form collecting unreliable qualification while suppressing volume has cost twice.
Use multi-step forms above 5 fields, so the visitor commits before seeing the full ask and partial completions stay recoverable.
What the page is worth fixing
=Monthly pipeline gap = monthly clicks x (achievable CVR - current CVR) x lead-to-won rate x ACV
Run it before commissioning a page rebuild. Source the achievable rate rather than guess it, because your own best-performing page is the honest ceiling.
Then count ad groups against landing pages. One page per campaign is too few and one per keyword is too many, so the 2 counts should match. An account with 14 ad groups and 3 pages is paying for message match on 11 of them and delivering it on 3. That gap is usually larger than any conversion rate optimization available on the pages that exist.
Testing without producing unreadable results
One variable at a time
Why: A test that swaps headline, description, and page has no diagnostic value
Full sales cycle before judging downstream metrics
Why: CTR and conversion rate read in 2 weeks. Lead quality does not
Never test while measurement is unstable
Why: A winner produced on a broken conversion action reverses when it is fixed
One test per campaign at a time
Why: 2 concurrent tests produce a result neither can explain
Marketing automation scripts firing before content renders.
Test on a mid-range Android on a normal connection rather than a recent iPhone on office wifi.
What goes wrong, and what it looks like
MistakeHow it shows up✓Fix
Demo CTA on problem-aware queries
How it shows up: High CTR, near-zero conversion rate
Fix: ✓Match the offer to the stage
Pinning all 3 headline positions
How it shows up: Asset reporting shows almost no variation
Fix: ✓Pin position 1 only
Form shortened to lift volume
How it shows up: Lead count rises, SQL count flat or down
Fix: ✓Length is a routing decision, not a volume lever
Copy test run during a tracking fix
How it shows up: The winner reverses 6 weeks later
Fix: ✓Sequence per Chapter 2
Unsubstantiated superlatives
How it shows up: Disapprovals, and skepticism when they do serve
Fix: ✓Replace with a specific, checkable claim
08
Stage · Automate9 min
Chapter 8: Bidding
Smart bidding executes the definition of success it was given, and in most B2B accounts that definition is a form submission. That is why this chapter sits after measurement.
What you are seeingWhat it actually means
Target CPA set at launch and never revisited
What it actually means: →Since August 2026 that target pulls costs toward it rather than being quietly beaten
Conversion volume up, pipeline flat, after moving to smart bidding
What it actually means: →The algorithm found cheaper form fills, which is what it was asked to do
Target ROAS running without validated CRM values
What it actually means: →The campaign optimizes against figures somebody assigned rather than observed
Campaign stopped serving after a target change
What it actually means: →The target sits below what any auction is available at
Device and audience bid adjustments carefully maintained
What it actually means: →Smart bidding has been ignoring every one of them
A tracking outage last month, no data exclusion set
What it actually means: →The algorithm learned from days you know were wrong
A campaign called at 14 days on a 60-day cycle
What it actually means: →The verdict arrived before the evidence
Which strategy this account should be on
Read down. The first row that matches is your answer.
Condition
Strategy
No history you trust, or a new account
Manual CPC until you have a baseline
Under 30 conversions per campaign per month
Maximize Conversions, no target
Permanently limited by budget
Maximize Conversions, no target
30 or more conversions, offline import not yet validated
Target CPA
Offline import validated, value map live
Maximize Conversion Value
Values validated against real closed deals for 60 days
Target ROAS
Stopping at Target CPA is a legitimate outcome, and stopping at Maximize Conversions is correct for a campaign that will never clear 30 conversions.
Setting and moving a target
Parameter
Value
Opening target
Trailing 30-day actual cost per conversion
Tightening step
Current target x 0.85, once per 7 days at most
Hard floor
Maximum CPL from Chapter 3
Serving risk threshold
Any target below trailing actual x 0.7
A target below current performance does not make the algorithm try harder. It makes the campaign stop serving, because no auctions exist at that price.
Actual is 20% under target consistently for 30 days
Tighten by 15%, wait a week, re-read
Campaign stopped serving after a tightening
Revert to the previous target rather than to something between
Target already at your maximum CPL
Stop tightening. The constraint is now conversion rate or CPC
Actual and target both above maximum CPL
The campaign is unprofitable at any volume. Fix the page or the keywords
Target ROAS
Target ROAS beats Target CPA for B2B, but only after offline conversions have been validated for 60 days. Before that it is worse, because it bids confidently against values that were assigned rather than observed.
A tROAS campaign on estimated values still produces a ROAS figure, it still appears in reporting, and nothing signals the denominator is fiction.
Most bid adjustments are ignored under smart bidding
Device, location, audience, demographic, and ad schedule bid adjustments are silently ignored under any smart bidding strategy. They sit in the interface, they look maintained, and the algorithm does not read them.
Control
Status under smart bidding
Device bid adjustment as a percentage
Ignored, except that a device adjustment on Target CPA modifies the CPA target rather than the bid
Device exclusion at -100%
Respected
Location bid adjustment
Ignored, though location targeting itself still applies
Audience bid adjustment
Ignored, though audience exclusions still apply
Demographic bid adjustment
Ignored, though demographic exclusions still apply
Ad schedule bid adjustment
Ignored, though the schedule itself, meaning the hours ads are eligible, still applies
Seasonality adjustments
Respected
Data exclusions
Respected
Under smart bidding, exclusions and targeting still work and percentage modifiers do not. An account that spent 2 years tuning a mobile adjustment has been tuning nothing.
✓Verified 20 August 2026 against Google Ads Help documentation.
The 2 advanced controls that do work
Both live under Tools, Shared library, Bid strategies, Advanced controls. Between them they are most of the real influence you have over smart bidding.
Data exclusions
A data exclusion tells smart bidding to ignore conversion data from a past date range because it is unreliable. A tag removed during a site release, a broken form, a consent banner misconfiguration. It is a retrospective correction, and almost no B2B account uses it despite tracking outages being common.
✓Use it when
Conversion tracking was broken or removed for a known date range
A form failed and submissions were lost
A consent or tag change suppressed recorded conversions
✕Do not use it when
Performance was poor
You dislike a period's results
The outage lasted weeks rather than days
Google's guidance is that using them often or for long periods degrades smart bidding, so treat them as a rare correction with a documented reason. Scope to affected campaigns rather than the whole channel, and set them the day the outage is found.
Seasonality adjustments
A seasonality adjustment tells smart bidding to expect a conversion rate change for a short known event. It is forward-looking and built for 1 to 7 days.
Most B2B accounts have 3 or 4 legitimate uses a year. A conference where your category buys, a fiscal year-end procurement window, a product launch, a webinar with a known registration spike.
Never exceed 7 days, because smart bidding already models longer patterns and a long adjustment fights it. Calibrate conservatively, because a large adjustment moves bids directly and can exhaust a daily budget within hours. Data collected during the period is not fully excluded from learning, so a wrong adjustment leaves a mark.
Test with Campaign Experiments rather than by switching
Campaign Experiments clones a campaign, splits traffic by a percentage you set, and reports both arms separately. It is the only way to compare a change against a control that ran in the same auctions at the same time.
✓Use an experiment for
Bidding strategy changes
AI Max feature enablement
Match type expansion
Landing page tests where the URL differs
✕Not for
Anything you can judge from existing reporting
Changes with no meaningful traffic to split
A campaign under 30 conversions a month, where neither arm will learn
More than one variable at a time
Split 50/50 and run one full sales cycle. A 2-week experiment on a 60-day cycle measures leads, which is what the change was most likely to inflate.
Learning phase rules
Daily budget changes under 20%
Consequence of breaking it: Larger moves restart learning rather than scaling within it
7 days before judging any change
Consequence of breaking it: The first days after a restart are the least representative data available
One change at a time
Consequence of breaking it: A bid adjustment, new creative, and a geo expansion in one week produce a result attributable to none of them
The judgment window
A campaign needs roughly one and a half sales cycles before you can read it on pipeline. A rule of thumb rather than a platform rule, and it exists because the alternative is a verdict delivered before the evidence.
Reads within 2 weeks
Search term relevance
Conversion rate
Cost per lead
Impression share
Needs a full cycle
Cost per opportunity
Deal size by campaign
Win rate by campaign
Pipeline created
A campaign killed at day 21 for expensive leads may have been producing the right ones, and the pipeline that would have arrived at day 75 gets credited elsewhere.
The August 2026 target change
On 17 August 2026, budget-limited campaigns on target-based bidding began optimizing toward the stated target rather than overdelivering against it. A target that has been quietly overdelivering will drift up toward the number in the field.
Maximize Conversions or Conversion Value with no target set
Campaigns that have never been budget-limited
Target CPC on Demand Gen is in scope per Google's documentation and most coverage omitted it. The flag also covers any campaign budget-limited at any point in the previous 12 months, so dormant campaigns are included.
B2B is more exposed than ecommerce, because lead gen campaigns sit budget-limited routinely and targets go stale once an account moves from form-fill to qualified-lead measurement.
Run this against every target-based campaign
Test
Condition
Action
Exposure
Trailing 90-day actual is under 80% of target
Costs will drift up to target. Reset it
Affordability
Target is above maximum CPL from Chapter 3
Never affordable. Reset to the ceiling
Staleness
Target untouched for 6 months
Recheck regardless of the 2 above
Fit
Permanently budget-limited and chasing volume
Move to Maximize Conversions with no target
Google's Bid Target Adjustment Tool, released 6 July 2026, flags affected campaigns and suggests targets from recent performance rather than your economics. It changes nothing automatically.
Running a deliberately loose target to give bidding room to explore no longer works on a budget-limited campaign.
✓Verified 20 August 2026 against Google's Help Center documentation.
What stays in your hands
Negative keywords
Why automation does not cover it: The algorithm optimizes within your targeting, it does not question it
Device exclusions at -100%
Why automation does not cover it: The only device control smart bidding respects
Ad schedule, meaning eligible hours
Why automation does not cover it: The schedule applies even though hourly bid adjustments do not
Geographic targeting and exclusions
Why automation does not cover it: Presence targeting and market exclusions are strategy decisions
Customer and audience exclusions
Why automation does not cover it: Stops paying to re-reach people who already pay you
Data exclusions and seasonality adjustments
Why automation does not cover it: The 2 advanced controls the algorithm reads
Budget allocation between campaigns
Why automation does not cover it: Smart bidding spends a campaign's budget efficiently. It never tells you the budget belongs elsewhere
Portfolio strategies optimize within their member campaigns rather than across your account, so reallocation stays a human decision.
What goes wrong, and what it looks like
MistakeHow it shows up✓Fix
Target set from a spreadsheet rather than trailing actual
How it shows up: Campaign stops serving, impression share collapses
Fix: ✓Reset to trailing 30-day actual
tROAS before the offline import is validated
How it shows up: Confident bidding, a ROAS figure describing nothing
Fix: ✓Run the 10-deal trace first
Smart bidding before pruning
How it shows up: Conversion volume rises, pipeline does not
Fix: ✓Chapter 5 before this chapter
Seasonality adjustment set too high
How it shows up: Daily budget gone before noon on day one
Fix: ✓Start smaller, 7 days maximum
Judging on cost per conversion after the offline import
How it shows up: The number appears to have quadrupled overnight
Fix: ✓The definition changed, not the performance
Tightening more than 15% at a time
How it shows up: Serving drops, then recovers, then drops again
Fix: ✓1 step, 7 days, re-read
09
Stage · Automate10 min
Chapter 9: The automation layer
AI Max, Demand Gen, and Performance Max each amplify whatever conversion signal you give them. In an account still counting form fills, all 3 find cheaper form fills faster than anything else. They also overlap, and running them against the same query space starves every model.
What you are seeingWhat it actually means
Campaign-level broad match or automatically created assets enabled anywhere
What it actually means: →That campaign migrates to AI Max during September 2026
Search, Performance Max, and AI Max all live
What it actually means: →3 models competing for the same queries on a fraction of the signal each
Demand Gen optimizing toward demo requests
What it actually means: →The algorithm is hunting a pocket too small to fund
Performance Max running without a validated offline import
What it actually means: →It is finding the cheapest form fills on the internet
Conversion volume up sharply since enabling any of these
What it actually means: →Check cost per SQL before celebrating
Standalone Display campaigns still running
What it actually means: →They migrate to Demand Gen, and several controls do not survive
Where each one belongs
Search, brandDefends the term, controls the message
Share of B2B budget: 5 to 8%
Search, non-brand and competitorCaptures existing demand
Share of B2B budget: 55 to 65%
Demand GenCreates demand, fills the retargeting pool
Share of B2B budget: 20 to 30%
AI MaxReaches demand your keyword list cannot describe
Share of B2B budget: Experiment only
Performance MaxHarvests, across surfaces
Share of B2B budget: 0 to 15%The usual recommendation: 30 to 40%
0%10%20%30%40%50%60%70%80%
One layer per intent tier. The usual recommendation puts Performance Max at 30 to 40% of a B2B budget and I put it at 0 to 15%.
Performance Max harvests people close to a purchase across every surface at once, which works when the conversion event carries its own value. A demo request does not. Until CRM data is flowing the campaign has nothing to harvest, so it finds the cheapest available form fill with less query visibility than Search gives you.
The 30 to 40% recommendation holds for high lead volume, short cycles, or a product-led funnel. On a 60-day considered purchase it is a large bet on a system you have not fed.
Demand Gen at 20 to 30% invites the same challenge and gets a different answer. An account that only harvests existing demand is capped by how many people are already looking, and in most B2B categories that number is a fraction of the addressable market. The allocation buys future search volume. Hold it accountable with the mid-funnel conversion goal and the offline import rather than with a smaller budget.
Before adding a layer, divide your monthly account conversions by the number already running. Smart bidding needs roughly 30 a month per model, so 3 layers on 90 account conversions sit at the floor.
AI Max
AI Max has 3 opt-in features inside an existing Search campaign rather than a campaign type, which makes the decision reversible and scoped.
Search term matching
What it doesExpands beyond your keywords using query understanding
The B2B riskPulls research and job-seeker intent your negatives were built to block
Text customization
What it doesGenerates headline and description variants from your assets and pages
The B2B riskFlattens positioning that took months to build
Final URL expansion
What it doesRoutes traffic to whichever page Google judges most relevant
The B2B riskSends buyers to pages that educate rather than convert
Final URL expansion depends on text customization being active, so partial adoption is structurally incomplete rather than cautious. Expect cheaper clicks, more of them, and worse pipeline, which a cost-per-lead report will not show.
The September 2026 migration
3 Aug 2026
Creation of new campaign-level broad match and legacy automatically created assets blocked across UI, Editor, and API
5 Aug 2026
Direct email to affected advertisers
12 Aug 2026
Technical mechanics published on the Google Ads Developer Blog
Sept 2026
Automatic migration to AI Max, running across the month
15 Jan 2027
Reminder notifications for advertisers still on Dynamic Search Ads
Feb 2027
Automatic DSA migration, new DSA ad group creation removed
The June 2026 announcement delayed Dynamic Search Ads to February 2027. It did not move this deadline.
Campaign was using
After migration it has
Final URL expansion
Automatically created assets
Search term matching and text customization
Off
Campaign-level broad match
Search term matching only
Off
Read the first row twice. A campaign that opted into automated headline writing inherits automated query expansion, which converts consent for creative generation into consent for query targeting. Brand exclusions carry over automatically, so a campaign with none configured migrates with none.
Google Ads API v25.1, released 19 August 2026, exposes the scheduled date per campaign through the fields Campaign.aca_migration_date_time and Campaign.broad_match_migration_date_time.
✓Verified 20 August 2026 against the Google Ads Developer Blog post of 12 August 2026.
Before the end of September
1
Audit every Search campaign for both legacy settings
Why: Most accounts are unaffected and stop here
2
Configure brand exclusions at campaign level on affected campaigns
Why: They carry over, so nothing configured means nothing protected
3
Freeze a dated performance baseline per affected campaign
Why: After migration there is no clean before to compare against
4
Decide whether to migrate deliberately or let it happen
Why: There is no opt-out, only a choice of timing and configuration
Testing AI Max deliberately
1
One non-brand campaign with 50 or more conversions in the trailing 90 days
2
Confirm offline conversions are importing and validated first
3
Configure brand exclusions at campaign level before enabling anything
4
Split 50/50 using Campaign Experiments, against an unmodified control
5
Enable all 3 features rather than one
6
Exclude or segment Search Partners
7
Run one full sales cycle minimum, 90 days at a 60-day cycle
8
Judge on cost per opportunity, never CPC, CTR, or CPL
9
Read results at account level, since AI Max shifts traffic between campaigns
2 conditions beyond the prerequisites. Impression share lost to rank is near zero, and positioning is not what you compete on. If either is false, I would not run the test.
AI Max and Performance Max are the only routes to serving in AI Overviews, and reporting shows little about when an ad served or what triggered it. Most B2B clients answering to a CFO should not take that trade yet.
Demand Gen
Demand Gen is the only place on Google inventory where you reach the buyer before they search, so judging it against Search on the same conversion goal fails every time.
March 2024
Discovery campaigns force-migrated to Demand Gen
July 2025
Video Action Campaigns absorbed, completing April 2026
Sept 2025
Primary bidding moved from Target CPA to Target ROAS
Jan 2026
Google Maps promoted pins added as a surface
May 2026
Standalone Display announced as moving into Demand Gen
June 2026
Display migration tool begins phased rollout
Jan 2027
Reported date for standalone Display creation ending
Demand Gen never serves on Google Search results pages. The confusion comes from YouTube Search, a Demand Gen placement that looks similar in reporting.
Surface
Video
Image
YouTube in-stream
Yes, but 6-second bumpers are excluded
No
YouTube in-feed, Shorts, Home
Yes
Yes
Discover
No
Yes
Gmail Promotions and Social
No
Yes
Google Maps
No
Yes
Google Display Network
No
Yes
7 decisions that decide whether it works
Channel controls
The rule: All channels on, 2 to 3 weeks, 50 to 100 conversions per channel before disabling the weakest
Conversion goal
The rule: A mid-funnel action with real monthly volume, never demo requests
Feedback loop
The rule: Import the down-funnel outcome per Chapter 4, or this is an awareness buy with no evidence
Creative refresh
The rule: Every 4 to 6 weeks
Creative format
The rule: Screen recording with captions over produced film, because most inventory autoplays muted
Landing page
The rule: Education-first. A demo wall is how this channel gets written off
Invalid traffic
The rule: Check monthly. Look for CRM records with no real company domain
The Display migration
The migration tool carries 42 days of performance history, cutting learning to 1 or 2 days, which is the reason to migrate deliberately rather than wait.
Google Display Network opted in by default
Detail: Cannot be unselected during migration. Other channels can be added after
The migration is one-way
Detail: No revert
Content suitability exclusions
Detail: Move to account level before migrating, or they are lost
For an account running Display on manual CPC with careful bid adjustments, this removes the entire control model.
Performance Max
B2B lead generation has no purchase to harvest. The signal separating a valuable conversation from a worthless one lives in a CRM the campaign cannot see unless you connect it. The black box criticism is out of date.
Control
State
Why it matters in B2B
Campaign-level negatives
100 to 10,000, self-serve
Most accounts still run under 50
Channel-level reporting
Budget split across every surface
Lets you correlate channels against CRM outcomes
First-party audience exclusions
Available
Stops paying to reach existing customers
Search themes
50 per asset group
The only steering input on the Search portion
Smart Bidding Exploration without a feed
Generally available since 15 June 2026
The first credible reason to reconsider PMax for lead gen
Performance Max is steerable now, and steering still requires something to steer toward, which is why I put it last.
Why: The difference between finding buyers and finding form-fillers
A lower-funnel conversion action with real monthly volume
Why: 12 qualified leads a month gives the model nothing
1,000 or more Customer Match records
Why: 100 provides almost no pattern to model from
Negative library from Chapter 5, applied at campaign level
Why: The 10,000 limit exists to be used
Account-level placement exclusions reviewed
Why: Negatives govern Search and Shopping only
All of them rather than most.
2 things once it runs. Correlate channel reporting against CRM outcomes rather than platform conversions, and reduce a weak channel by supplying fewer assets in its format, since there is no clean channel exclusion. And block negative categories rather than individual queries, because a negative for every non-converting term removes the campaign's ability to find anyone new.
What goes wrong, and what it looks like
MistakeHow it shows up✓Fix
AI Max without brand exclusions
How it shows up: Budget lands on competitor brand terms
Fix: ✓Configure at campaign level before enabling
Judging AI Max on CPL
How it shows up: Looks like a win, pipeline says otherwise
Fix: ✓Cost per opportunity, at account level
Demand Gen channels disabled in week one
How it shows up: A YouTube-only campaign that never learns
Fix: ✓50 to 100 conversions per channel first
Display migrated without moving content exclusions
How it shows up: Years of exclusions lost
Fix: ✓Move to account level first
PMax launched to hit a quarterly target
How it shows up: Cheap conversions, no pipeline, no way to diagnose
Fix: ✓Prerequisites first
PMax negatives added one query at a time
How it shows up: Exploration collapses, volume falls
Fix: ✓Block categories
New campaign type read as incremental
How it shows up: Campaign-level wins, account-level flat
Fix: ✓Compare account totals rather than campaign totals
10
Stage · Scale7 min
Chapter 10: Audiences and scaling
Google's audience system models one consumer doing one thing, so what works in B2B is your own data applied to 2 jobs most accounts collapse into one. Reaching people who showed intent, and excluding people you should not be paying to reach. The second is where the recoverable money sits.
Scaling has the same shape. The budget is the last lever, and the cheapest growth in most accounts needs no new strategy.
What you are seeingWhat it actually means
Existing customers appearing in remarketing pools
What it actually means: →You are paying to reach people who already pay you
Retargeting window set at 30 days
What it actually means: →It expires before a B2B buying committee finishes its first discussion
Lookalikes seeded from all leads
What it actually means: →You are modeling people who fill in forms rather than people who buy
Budget raised on a campaign already winning its auctions
What it actually means: →You are buying the next tier down, which is less qualified by definition
Lead volume tripled, sales says quality dropped
What it actually means: →Check follow-up time before changing anything in the ad account
Cost per lead flat while cost per opportunity rises
What it actually means: →Expansion has moved past the qualified audience
Exclusions come before targeting
In subscription businesses your most engaged visitors are often the people least worth acquiring. Build these 3 before any remarketing list.
Existing customers, as a Customer Match list
Where it appliesNon-brand, competitor, Demand Gen, Performance Max
WhyPure loss. They search your category while evaluating adjacent products
Recent converters, for the length of your sales cycle
Where it appliesAll acquisition campaigns
WhySomeone who requested a demo 11 days ago does not need an acquisition ad
Disqualified leads, where the CRM records a reason
Performance Max gained first-party audience exclusions in 2026, so this is now possible across every campaign type where it previously was not.
Segment by behavior rather than by visit
The default list of all visitors over 30 days contains job applicants, existing customers, competitors, and people who read one blog post.
SegmentSignal strengthTreatment
Started a form, did not submitSignal strength: HighestTreatment: Direct offer, very short window
Viewed pricing, did not convertSignal strength: HighTreatment: Direct offer, short window
Viewed a comparison or competitor pageSignal strength: HighTreatment: Migration or switching message
3 or more pages in one sessionSignal strength: MediumTreatment: Case study or proof content
Viewed one blog postSignal strength: LowTreatment: Exclude, or treat as cold
Visited the careers pageSignal strength: NegativeTreatment: Exclude entirely
Form abandonment is the strongest signal available and the least commonly built, because it needs an event rather than a pageview.
The careers page exclusion looks small and is not. Job applicants are engaged repeat visitors who generate strong behavioral signals and never buy.
Window length follows the sales cycle
Set the window to at least the length of your sales cycle, because a shorter one stops reaching people exactly when the evaluation becomes serious. Then segment by recency inside it. Recent visitors get the direct offer, and older ones get proof, because the question has moved from whether the product fits to whether the company is a safe choice.
Customer Match and lookalikes
Upload 3 lists rather than one.
All customers
Used for: Exclusion
Closed-won, last 2 years
Used for: Lookalike seed and value assignment
Closed-lost at late stage
Used for: Re-approach at renewal timing, and a second lookalike seed
The third list is rarely built and it is the one worth building. These companies qualified, evaluated, and chose otherwise, which makes them a closer match to your buyer than any lead list.
Seed lookalikes from closed-won rather than from all leads. A lookalike built on your lead list models people who fill in forms. Lists in the low hundreds produce weak modeling regardless of whether they clear the technical minimum, so treat 1,000 records as the practical floor.
Audience signals are not targeting
In Performance Max and Demand Gen, audiences guide the model rather than constrain it, so the campaign serves outside the audience you provided. Exclusions are respected. In automated campaign types your exclusions carry more weight than your targeting, which is why the first section of this chapter comes first.
Cap frequency before the pool gets small
B2B retargeting over-serves more often than it under-serves, because the pool is small and the budget will find a way to spend itself. Cap frequency where the campaign type allows it, control it through window length and budget where it does not, and check impressions per user rather than assuming.
Scaling starts where growth is already available
Auction depth is finite. Past a certain share, additional impressions come from less relevant queries, and cost per opportunity rises while cost per lead stays flat.
1
Close impression share lost to budget
Why it sits here: You are already winning these auctions and choosing not to
2
Close impression share lost to rank
Why it sits here: Improving relevance is cheaper than buying worse traffic
3
Expand keywords from converting search terms
Why it sits here: Demand you have already proven exists
4
Add adjacent verticals or use cases
Why it sits here: New demand, same buyer profile
5
Add geographies
Why it sits here: New demand, new operational requirements
6
Add campaign types
Why it sits here: New surfaces, new learning periods
7
Raise budget on saturated campaigns
Why it sits here: Last, because this is where efficiency degrades
Most accounts start at 7 because it takes 10 seconds and needs no analysis. The first 3 are almost always available and almost always cheaper.
The budget headroom calculation from Chapter 3 sizes priority one, as a ceiling rather than a forecast.
For impression shares lost to rank, diagnose the component before acting. Google reports expected click-through rate, ad relevance, and landing page experience separately, and each has a different fix. Bidding higher works and is the expensive answer.
Scaling within a campaign
Increment size
Detail: Under 20% of daily budget at a time
Interval
Detail: 7 days between increases
Never combine
Detail: Budget change and bidding target change in the same week
The metric to watch
Detail: Cost per opportunity rather than cost per lead
Where accounts break at 3 times the spend
ConstraintHow it shows up✓Where the fix lives
Auction depth
How it shows up: Cost per opportunity rises, cost per lead flat
Where the fix lives: ✓Stop expanding, return to priority 1
Landing page capacity
How it shows up: Conversion rate falls as volume rises
Where the fix lives: ✓The page did not change, the audience mix did
Sales capacity
How it shows up: Follow-up slows, downstream conversion drops
Where the fix lives: ✓Operational, not in the ad account
Creative fatigue
How it shows up: Demand Gen and Display performance decays within weeks
Where the fix lives: ✓Refresh cadence from Chapter 9
Sales capacity produces a confident wrong diagnosis, because marketing sees lead quality declining, sales reports the same, and both are describing follow-up time. Before any scale-up, ask what happens to speed to first contact at 3 times current volume.
Geographic expansion
A new market is a new account wearing the clothes of an existing one. CPCs differ, competitor sets differ, and query language differs even where the language is nominally the same.
Never mix languages in one campaign
Why: Ad delivery and Quality Score both degrade on a mismatch
Build the negative library separately per market
Why: The wrong-industry category is market-specific by nature
Do not assume a translated page performs like the original
Why: The qualifiers that filter buyers in one language often have no equivalent
Treat the first 90 days as measurement
Why: The purpose is learning what a qualified query looks like there
What goes wrong, and what it looks like
MistakeHow it shows up✓Fix
No customer exclusion list
How it shows up: Recurring spend reaching people who already pay you
Fix: ✓Upload Customer Match, exclude everywhere
30-day retargeting window
How it shows up: Lists expire mid-evaluation
Fix: ✓Set to the sales cycle, segment by recency inside it
Audience signals treated as targeting
How it shows up: Confusion when traffic arrives from outside the list
Fix: ✓Signals guide, exclusions constrain
Budget raised before closing impression share lost to budget
How it shows up: Efficiency degrades on the way to growth already available
Fix: ✓Work the priority order
Scaling and bidding changes in the same week
How it shows up: Neither effect is attributable
Fix: ✓One change, one learning period
A new campaign type read as incremental
How it shows up: Campaign-level wins, account-level flat
Fix: ✓Compare account totals
11
Stage · Operate6 min
Chapter 11: Operating and reporting
Google Ads reports on activity that completes in days for a business that completes in months. Last month's conversions sit next to a fully incurred cost and read as money going out.
What you are seeingWhat it actually means
Reporting leads with conversions and cost per conversion
What it actually means: →You are reporting activity rather than outcomes
Cost per conversion quadrupled after the offline import went live
What it actually means: →The definition changed, not the performance
No change log
What it actually means: →Every performance investigation starts from memory, and memory is usually wrong
A 20-tab spreadsheet goes to the CFO
What it actually means: →It communicates thoroughness and prevents a decision
Weekly reporting on a 60-day cycle
What it actually means: →Noise that invites intervention
Measurement owned by nobody in particular
What it actually means: →The offline import sits unbuilt between marketing and engineering
Report outcomes rather than activity
✕Stop reporting
Impressions
Clicks
Click-through rate
Cost per conversion
Conversions
Cost
✓Report instead
Impression share on target terms
Qualified leads
Conversion rate to qualified lead
Cost per sales-accepted lead
Opportunities created
Cost per opportunity, and pipeline created
The right column needs the CRM connection from Chapter 4. Where it does not exist, say so, because that turns a measurement gap into a resourcing conversation.
The 4 metrics a CFO needs
What it answers: Is this getting better or worse?
Number: Cost per opportunity, trended over 2 quarters
What it answers: What did the spend produce?
Number: Pipeline created, with the offset stated
What it answers: Does this work for our economics?
Number: Cost per opportunity against allowable CAC
What it answers: Would more spend produce more?
Number: Headroom from impression share lost to budget
One page. 4 metrics, a trend line, and a written interpretation. The interpretation is the only part that says what should happen next.
Monthly for leadership. Quarterly for strategy, because that is the first interval at which a full cohort has matured.
Handling the lag
This month's spend produces next quarter's pipeline, so putting both in one table implies a relationship that does not exist.
Offset periods
How it works: This month's opportunities against spend from 2 months ago. Label the offset
Cohorts
How it works: Leads acquired in June, tracked forward, shown at current maturity
State maturity
How it works: A cohort at 30 days is incomplete and never sits next to a mature one
Without it, a campaign launched 8 weeks ago shows high cost and few opportunities, gets cut, and the pipeline that would have arrived in week 11 never does.
Explaining the metrics that get worse on purpose
Negative keyword deployment
What gets worseCTR and clicks fall
What to sayStopped paying for people who cannot buy
Offline conversion import
What gets worseConversion count falls, CPA rises sharply
What to sayNow counting qualified leads instead of form fills
Longer forms
What gets worseLead volume falls
What to sayRouted unqualified inquiries away from sales
Competitor campaign
What gets worseCPL well above account average
What to sayDifferent economics, measured by opportunity rate
Removing secondary conversions from bidding
What gets worseReported conversions fall
What to sayThose events were never revenue
Turning off auto-apply
What gets worseOptimization score drops
What to sayThe score measures review, not performance
Row 2 causes trouble. Cost per conversion rises by a multiple rather than a percentage, roughly the inverse of your lead-to-SQL rate, in a number leadership already watches. Restate last quarter under the new definition so there is something to compare against.
Attribution honesty
Last click gives paid search no credit for awareness created months earlier, and multi-touch distributes credit across 4 captured touchpoints when 11 existed and 7 happened in meetings.
I use platform attribution for optimization decisions and a required "how did you hear about us" field for directional truth. Report both, name the gap, and do not resolve it artificially.
The operating cadence
The running checklist at the back sets out the weekly, monthly, quarterly, and annual work. Book the weekly 30 minutes as a recurring block with a name on it.
The change log
One row per change: date, what changed, which campaign, why, and what you expected to happen.
The expectation field turns every change into a test, and it means that when cost per opportunity moves in March the answer takes 4 minutes rather than a week.
Handling the quarterly Google review
Your rep is measured on your spend and your optimization score. Neither is bad faith, and neither is your objective. A director who cannot hold that conversation calmly ends up either implementing recommendations they do not believe in or losing the useful parts of the relationship.
Your optimization score is 62%
The position to hold: The score measures whether recommendations have been reviewed, not whether the account works. Dismissing gives the same uplift as applying
You should test Performance Max
The position to hold: Name the prerequisites from Chapter 9. Offer to run it once offline conversions have been validated for 60 days
Broad match will find incremental demand
The position to hold: Agree in principle, then show the negative library and ask which categories they would remove
Increase budget on this campaign
The position to hold: Ask for impression share lost to budget by campaign. If it is near zero, more budget buys worse auctions
Enable auto-apply for these categories
The position to hold: Ask which ones change keywords, bids, or assets. Those stay off
This campaign has low conversion volume
The position to hold: Correct, and the conversion is a qualified lead rather than a form fill. Show cost per opportunity instead
2 things in the relationship are worth having. Reps see platform changes before the trade press does, and they escalate policy disapprovals and billing faster than any support queue.
Bring cost per opportunity against allowable CAC to every review. It moves the discussion from Google's metrics to yours.
Name who owns each of the 3 roles
Weekly routine
Accountable for: That it happens, not for performance
Accountable for: Authority to restructure, kill campaigns, and reallocate budget
Measurement integrity is the one that ends up owned by nobody, between marketing and engineering.
When performance drops
Check these 4 before changing anything, because roughly half the time one explains it.
1
The change log, 7 to 14 days back
What it catches: Something you did
2
Measurement
What it catches: A failed import, broken tag, or consent change looks exactly like a bad month
3
Auction insights
What it catches: A new entrant changes your costs with nothing changing in your account
4
Platform changes in the window
What it catches: Migrations alter campaign behavior without any advertiser action
What has to survive a handover
Naming convention and what each element means
Why: The only documentation that appears in every report
The change log
Why: The account's history
Value map, with source data and last recalculation date
Why: Otherwise it gets treated as permanent
Negative library, with categories and deployment levels
Why: Otherwise it gets rebuilt or ignored
Allowable CAC and maximum CPL, with inputs
Why: Otherwise targets get set against assumptions
Settings deliberately changed from default, and why
Why: A new manager will turn Search Partners back on and rediscover the reason at full cost
Data exclusions and seasonality adjustments applied, with dates and reasons
Why: Otherwise they look arbitrary and get deleted
00
4 min
Chapter 0: The first 90 days
Every chapter ends with one action. These are those actions in sequence, gated on a scorecard number rather than a date, so the plan paces itself against the account.
Steps 1 to 3, about a week
1
Export 90 days of search terms, filter to clicks with zero conversions, sum the cost
Move on whenThe routine has an owner and a calendar slot
Steps 1 through 3 take a week and steps 4 through 8 take 4 to 6. Step 9 depends on engineering availability and is where the sequence usually stalls, which is why every step needing nobody else sits ahead of it. Steps 10 through 13 take a day between them and get skipped most often, because by then the account is performing better and the urgency has gone.
None of it requires a new campaign type, a new budget, or a new agency. It requires the account to be measured before it is optimized.
What a healthy account looks like at steady state
This is the target, so you know when to stop fixing and start operating.
Measurement✓
CRM outcomes flow back daily and both systems agree
The number that proves itMeasurement variance under 15%, trace rate 8 of 10 or better
Signal quality✓
Bidding trains on qualified leads rather than form fills
The number that proves itCost per SQL within a predictable multiple of cost per lead
Waste✓
The negative library is maintained rather than built once
The number that proves itShare of spend on zero-conversion terms under 15% and flat
Structure✓
Every campaign has enough volume to optimize
The number that proves itNo campaign under 30 conversions a month
Message✓
Every ad group has a page written for its query
The number that proves itAd group count equals landing page count
Bidding✓
Targets reflect what the business can afford
The number that proves itNo target above maximum CPL, none untouched for 6 months
Automation✓
One layer per intent tier, each with something to steer toward
The number that proves itAccount totals rise when a layer is added, not just campaign totals
Governance✓
Changes are logged and the weekly routine has an owner
The number that proves itA performance question gets answered in minutes rather than days
An account in that state needs the weekly 30 minutes, the monthly reconciliation, and someone paying attention when the platform changes.
The running optimization checklist
The recurring work after the 90 days. Every row states why it is on the list, because a checklist without reasons gets abandoned the first week something urgent happens.
Check
Why it earns the slot
What good looks like
New search terms, sorted by cost
The expensive queries get caught in week one rather than week 10
5 to 15 new negatives, fewer over time
No keyword stopped serving from last week's negatives
A negative always overrides a positive keyword, and this is the most common self-inflicted outage
Zero unexplained keyword outages
Spend pacing against budget
A campaign limited by budget mid-month needs a decision rather than a discovery on the last day
Pacing within 10% of plan
Change log updated
90 seconds now saves a week of reconstruction in March
Every change has a row with an expectation
Check
Why it earns the slot
What good looks like
N-gram pass on the last 30 days
Finds the word doing damage across hundreds of queries rather than one at a time
20 to 50 new negatives, trending down
Platform against CRM reconciliation
Catches a broken tag or import before it costs a full quarter of bidding data
Variance under 15%
Offline import rejection rate
Sync failures are silent and look identical to a bad month
Rejections under 5%, upload volume stable
Bidding targets against trailing actual
Since August 2026 a stale target pulls costs toward it rather than being quietly beaten
No target more than 20% off actual
Asset-level reporting
Low assets dilute the pool and unrated assets mean the ad group is too small
Nothing at Low for more than 30 days
Impression share by campaign
Tells you whether the constraint is money or relevance before you spend either
Lost to budget near zero on your best campaigns
Cost per SQL by campaign
The gap against cost per lead is the earliest read on quality drift
The gap stable or narrowing
Check
Why it earns the slot
What good looks like
Prune the negative library
Lists accumulate, and a term blocked 10 months ago may now be one you want
Documented removals, no valid traffic blocked
Recalculate the value map
Stage probabilities drift every time the sales process or the ICP changes
Values within a quarter of current close rates
Re-derive allowable CAC and maximum CPL
ACV and close rates move more often than anyone updates the target
Both figures dated within 90 days
Recheck auto-apply settings
Google changes which recommendations are eligible, so a category you switched off can be replaced
Only tracking hygiene enabled
Review landing pages against current ads
Positioning drifts and pages fall out of sync with the copy
Message match holds on your top 5 campaigns
Run one structured experiment
One, because 3 simultaneous tests produce a result none of them can explain
A readable result on a full sales cycle
Read the quarter's platform changes
Migrations alter campaign behavior with no advertiser action
You knew before the change landed
Re-run the full audit from Chapter 3, including the 12 calculations and the 10-deal trace. Then rebuild structure and negative categories against the current ICP, because the account that fit the business 2 years ago describes a company that no longer exists.
Want us to run this play with you?
Book 30 minutes. We'll look at where you are and whether this play fits your quarter.