Find out why AI names your competitors and not you, which of 5 layers is holding your brand back, and how to measure your share of AI answers every month. You leave with a scorecard, the metrics to track and a 90-day plan.
By Sabih Ahmed, FounderUpdated Oct 2026
Which scheduling software should a multi-site operator use?
✦
3 options come up most for teams running several sites:
Marketing to run the prompts and the outreach, with engineering and security for crawler access and rendering. Under a day to take the baseline, then a monthly prompt run and a quarterly rebuild.
!Skip it if
Buyers in your category don't use AI assistants to research yet.
Score your account
How strong is each link in your chain?
Each line is what good enough looks like at one layer, in this playbook's words. Answer for your brand today and see where your chain stands.
AI visibility is not a ranking position. It is a chain.
Before an AI system recommends you, it has to find information about you and understand what you do. Then it has to judge you relevant, see evidence you are credible, and trust that recommending you is safe. The chain breaks at its weakest link rather than averaging out.
There is no 30-day promise attached to this, because the work depends on where your chain is weak. Chapter 2 tells you how to find that before you read a single tactic.
Discoverability
Clarity
Relevance
Authority
✕Trust
Recommendation
When Authority is the weak link, nothing after it counts.
Information reaches an answer through 2 routes, and which one dominates your category decides whether your program takes quarters or years.
Route 1, what the model already knows
Training creates associations between brands, categories, and problems before any search happens, so some brands are already in the candidate set when the question is asked.
You cannot edit a trained model, and nobody can promise inclusion in the next training set. What you influence is the public footprint those datasets are built from, which is a multi-year position rather than a campaign.
Route 2, live retrieval
The system interprets the need, expands the question into searches, retrieves and corroborates sources, then synthesizes.
A brand with no presence in memory can still be recommended if the right sources say the right things this quarter. That is why the test below matters more than any tactic in this playbook.
Test which route your category runs on
It takes 40 minutes and 5 prompts, with no tooling.
1
Pick 5 prompts a buyer would ask, weighted to category and comparison questions
2
Open a clean session with memory off and search disabled. Log out or use a temporary chat
3
Run all 5, record every brand named. This is the parametric candidate set
4
Repeat with search enabled in a fresh clean session. This is the retrieved set
5
Compare the 2 brand lists
Nearly identical lists
The model holds an opinion. Multi-quarter build, short-term content moves little
Substantially different lists
Retrieval dominates. Movement possible in weeks
Incumbents identical, challengers different
The common B2B pattern. Incumbents defend memory, challengers compete for retrieval slots
The third row applies to most B2B categories. Competing for memory slots takes years, competing for retrieval slots takes quarters, and this test tells you which one you are doing.
Search offWhat the model remembers
Incumbents, named either way
Search onWhat it finds liveChallengers, named only when it searches
The common B2B pattern
The query you see is not the query AI runs
A complex question gets decomposed into several smaller searches, which Google describes as query fan-out. Take a buyer asking which CRM suits a 50-person SaaS company that is moving off Salesforce, uses Slack and has a set budget. That one question can trigger separate investigations into category software, alternatives, vendor pricing, integration depth, implementation reviews, support sentiment, and migration guidance.
You can be excluded at any one of them. Your pricing page, integration page, and migration docs are all part of an answer none of them was written for.
Which CRM suits a 50-person SaaS company moving off Salesforce?
Category software
Alternatives
✕Vendor pricing: Missing from one search, missing from the answer
Integration depth
Implementation reviews
Support sentiment
Migration guidance
Missing from one search, missing from the answer
Fetched, cited, mentioned, and recommended are different things
1Fetched
What happenedRetrieved behind the scenes. The user never knows
What it is worthAccessibility, nothing more
2Cited
What happenedVisible attribution for a claim
What it is worthUseful, though the claim often supports someone else's recommendation
3Mentioned
What happenedYour company appears, possibly without your site being cited
What it is worthEntry to the consideration set
4Recommended
What happenedPresented as a suitable choice for the buyer described
What it is worthThe commercial outcome
Measure all 4 separately. Reporting citations as visibility measures the wrong end of the chain, and it is the most common reporting error in this discipline.
02
Stage · Days 1 to 102 min
Chapter 2: The brand AI visibility model
The model has 5 layers. Each answers one business question and produces one working document you keep.
1
Brand Discoverability
The question it answersCan AI find and retrieve us?
Working output10-URL Discoverability Audit
2
Brand Clarity
The question it answersDoes AI understand what we are and when we fit?
Working outputBrand Truth Sheet
3
Brand Relevance
The question it answersDo we supply what AI needs to answer buyers?
Working outputPrompt-to-Evidence Map
4
Brand Authority
The question it answersDoes the ecosystem treat us as credible?
Working outputRetrieval Map
5
Brand Trust
The question it answersIs there independent evidence to recommend us?
Working outputClaim-to-Evidence Matrix
How long each layer takes, and what it can win
DiscoverabilityA gate. Passing wins nothing alone
Days to weeks
Days to weeks
ClarityHigh, and cheap for the return
2 to 4 weeks
2 to 4 weeks
RelevanceHigh
1 to 2 quarters
1 to 2 quarters
AuthorityThe highest, and the slowest
Quarters to years
Quarters to years
TrustGates whether recommendation is safe
Continuous
Continuous
DaysWeeksMonthsQuartersYears
Discoverability and Clarity are gates. Pass them, maintain them, and further investment returns nothing. Authority is the lever you pull for years, and Chapter 6 carries the evidence for why.
The outcome, which is brand recommendation
When the right buyer asks the right question, does your brand make the consideration set?
An AI answer names a handful of brands. In a category with 30 credible players that is a shortlist, not a ranking, and there is no position 11 collecting traffic.
=Share of consideration set = times your brand is named / times the prompt is run
Run each prompt 5 times across 2 platforms in clean sessions. Record the share, never the single answer.
Best [category] for [audience]ChatGPT
Mentioned2 of 5
Recommended0 of 5
Position and context: Third in a list of 6, no detailSources cited: 2 review platforms, a vendor blog, a community thread
Best [category] for [audience]Perplexity
Mentioned4 of 5
Recommended1 of 5
Position and context: Recommended for mid-market, our page cited onceSources cited: 2 review platforms, our site, a trade publication
[Competitor] alternativesChatGPT
Mentioned0 of 5
Recommended0 of 5
Position and context: Absent from every runSources cited: Competitor site, 2 third-party listicles
[Use case] for [audience]Perplexity
Mentioned1 of 5
Recommended0 of 5
Position and context: Named as an option, no fit criteria attachedSources cited: One review platform, 2 vendor blogs
Those rows show the pattern most brands find. Presence where retrieval is live, absence where the answer comes from memory, and nothing at all on competitor-alternative prompts, which is usually the highest-intent question in the set.
Find your constraint
Work down. Stop at the first row that describes your brand. When several apply the earliest wins, because Authority work spends months producing sources that point at a site the system cannot read.
If none applies and you are still absent, you are in a parametric category with an incumbent set the model already holds. Go to Chapter 6 and read the cold start section.
03
Stage · Days 1 to 108 min
Chapter 3: Brand discoverability
Can AI systems find, access, and retrieve your brand's information?
Discoverability is a gate. Passing it wins nothing on its own, and failing it removes you from consideration however strong the rest of your chain is. It is also the cheapest layer to fix and the one most often broken by a change nobody flagged as a marketing decision.
What you are seeingWhat it actually means
Traffic is healthy and AI never cites you
What it actually means: →Pages render for humans and return nothing to a crawler
Robots.txt allows everything and bots still get blocked
What it actually means: →The security layer answers before robots.txt is read
Someone blocked Google-Extended to protect the brand
What it actually means: →That does not remove you from AI Overviews and never did
Your best evidence sits behind a form
What it actually means: →It cannot be retrieved, so it cannot be cited
Nobody has checked which AI crawlers reach the site
What it actually means: →Server logs hold the answer and nobody has looked
Which crawler does what
Blocking decisions get made without knowing which bot does which job, and the 2 most expensive mistakes both come from that confusion.
Google
Googlebot
What it feeds: Search index, AI Overviews, AI Mode
Blocking it means: You disappear from Google entirely
Google-Extended
What it feeds: Gemini training, Vertex grounding
Blocking it means: You opt out of training. AI Overviews are unaffected
OpenAI
GPTBot
What it feeds: Model training
Blocking it means: You opt out of training. ChatGPT search is unaffected
OAI-SearchBot
What it feeds: ChatGPT search index
Blocking it means: You disappear from ChatGPT's live results
ChatGPT-User
What it feeds: User-triggered fetches mid-conversation
Blocking it means: Real-time retrieval when a user asks about you
Anthropic
ClaudeBot
What it feeds: Model training
Blocking it means: You opt out of training
Claude-SearchBot, Claude-User
What it feeds: Search index and user-triggered fetches
Blocking it means: Live retrieval in Claude
Perplexity
PerplexityBot
What it feeds: Perplexity index
Blocking it means: You disappear from the most citation-heavy surface
Microsoft
Bingbot
What it feeds: Bing index, Copilot
Blocking it means: You disappear from Copilot
Common Crawl
CCBot
What it feeds: Open dataset used widely in research and training
Blocking it means: Long-term footprint only
Blocking Google-Extended does not protect you from AI
It has no effect on AI Overviews or AI Mode, which are served from the standard index through Googlebot. Teams block it, feel protected, remain fully exposed, and give up Gemini grounding for nothing.
Training crawlers and search crawlers are different decisions
Blocking GPTBot is a defensible content-licensing position. Blocking OAI-SearchBot is a visibility decision dressed as a legal one.
If your policy is to opt out of training while staying retrievable, block the training bots and allow the search and user-triggered ones. Write the policy down, because the next person to inherit robots.txt will not be able to tell which choice was deliberate.
Robots.txt does not reliably control user-triggered fetchers
Fetchers that fire when a user asks about you behave differently from indexing crawlers. The firewall is the enforcement point, not the text file.
The 9 tests, in the order that finds problems fastest
Run these in sequence and stop when you find the break. Tests 1, 2, and 4 account for most real failures. Test 9 is worth knowing and almost never worth acting on this quarter.
1
Security layer
How to run itCurl your top 10 URLs with an AI bot user agent from an external IP
Fails whenAnything other than a 200
2
JavaScript rendering
How to run itDisable JavaScript and load the same 10 pages
Fails whenContent, headings, or prices disappear
3
Crawler policy
How to run itRead robots.txt against the table above
Fails whenA search or user crawler is disallowed by accident
4
Retrievability of key facts
How to run itView source and search for pricing, integrations, positioning
Fails whenThe fact exists only in an image, PDF, tab, or accordion
5
Indexing status
How to run itSite search and Search Console coverage
Fails whenPriority pages excluded or not indexed
6
Structured data
How to run itValidate Organization, Product, FAQ, Article markup
Fails whenSchema contradicts the visible page
7
Server log evidence
How to run itFilter 30 days of logs by AI user agent
Fails whenNo AI crawler appears at all
8
Gated content audit
How to run itList everything behind a form or login
Test 1, the most common failure and the least visible
Robots.txt permits the crawler, but the WAF, bot manager, or CDN rejects it before the rule is consulted. Marketing sees a permissive robots.txt and assumes access while the bot receives a 403 or a challenge page.
Run this from outside your network, once per crawler you care about.
curl -A "OAI-SearchBot" -o /dev/null -s -w "%{http_code}\n" https://yourdomain.com/pricing
ResponseDiagnosisOwner
Response: 200 with full HTMLAccessibleOwner: None
Response: 403 or 401Security layer blocking. Robots.txt is irrelevant hereOwner: Security or infrastructure
Response: 429Rate limiting. The crawler gives up before retrievingOwner: Infrastructure
Response: 200 with a challenge or interstitialBot management returning a shellOwner: Security
Response: 200 with an empty shellRendering problem, go to test 2Owner: Engineering
Bot management is usually configured by a security team with no visibility into the commercial consequence, and no ticket exists connecting the 2. That is why this test comes first.
Test 2, because rendering failures look like success
Disable JavaScript and load your 10 priority pages. What remains is close to what a non-rendering retrieval system receives.
Pricing tables
Consequence: Absent from every pricing and budget prompt
Integration lists
Consequence: Lost from every integration query in the fan-out
Product descriptions
Consequence: Clarity fails, not only discoverability
Reviews and testimonials
Consequence: Trust evidence never reaches the system
Navigation
Consequence: Internal discovery of your other pages fails
Not every AI system renders JavaScript, and those that do may not wait for slow scripts. Server-side rendering or static generation for priority pages is the fix, and it is an engineering decision with a marketing consequence.
The llms.txt question, answered
Every prospect asks. The evidence is now strong enough to answer plainly.
Google AI optimization guidance, May and June 2026
Finding: Not needed for AI Overviews, AI Mode, or any generative search feature
Google Search team, on record
Finding: Gary Illyes confirmed no support and no plan to add it. John Mueller compared it to the keywords meta tag
SE Ranking, 300,000 domains
Finding: 10.13% adoption. Among the 50 most AI-cited domains, one had the file
SE Ranking predictive model
Finding: Removing the llms.txt variable improved accuracy at predicting citation frequency
Limy, 500 million AI bot events over 90 days
Finding: Hundreds of requests targeted the file directly
Originality.ai
Finding: 97 percent of published files received zero AI requests
The model result is the one to remember. The variable did not fail to help, it added noise, which is what a signal nothing reads looks like in a dataset.
One narrow use is legitimate. Anthropic recommends the file for agent-facing documentation, OpenAI references it for the Agents SDK, and Chrome's Lighthouse added an agentic browsing audit in 2026 that checks for it. If you publish API documentation or expect agent traffic, half a day is a reasonable bet.
For a B2B brand trying to enter category answers it is not an AI visibility tactic. Google's same guidance names 3 others in that category: content chunking for AI, AI-specific content rewriting, and special AI schema.
What schema does, and what it will not do
Schema does not make AI cite you. It reduces ambiguity about what you are, which helps entity resolution and cannot hurt.
TypePriorityWhat it clarifies
OrganizationPriority: Highest, because this is entity resolutionWhat it clarifies: Legal name, alternate names, logo, sameAs links to your profiles
Product or SoftwareApplicationPriority: HighWhat it clarifies: What you sell, category, pricing model
FAQPriority: MediumWhat it clarifies: Question and answer pairs already visible on the page
ArticlePriority: Medium, and update date matters for freshnessWhat it clarifies: Author, publish date, update date
One rule outranks any type choice. Schema must match the visible page, because markup claiming a price the page does not show is a contradiction, and contradictions cost you more.
The 10-URL Discoverability Audit
Choose the 10 URLs an AI system would most need to answer a buying question about you. Usually homepage, pricing, 2 or 3 product pages, integrations, a comparison page, security, documentation, and your strongest case study.
Check
Pricing
Integrations
Security
Comparison
URL
/pricing
/integrations
/security
/compare/[competitor]
Bot status
403 to OAI-SearchBot
200
200
200
Renders without JS
No, table is JS
Partial (lazy-load)
Yes
Yes
Key facts in HTML
No, price in script
Partial
Yes
Yes
Indexed
Yes
Yes
No
Yes
Schema valid
Organization only
None
None
Article
AI bot hits, 30 days
0
4
0
31
Action
Allow the bot, render the table server-side
Render the list server-side, add Product schema
Remove no-index from staging
None
Owner
Infra, then Engineering
Engineering
Engineering
None
Those 4 rows show the common pattern. The pages carrying your most decision-relevant facts are the ones most likely to be JavaScript-dependent, because they were built as interactive components rather than as documents.
04
Stage · Days 11 to 304 min
Chapter 4: Brand clarity
Does AI understand what you are, what you offer, and when you are the right choice?
A system that cannot classify you confidently will not recommend you, because recommending the wrong thing is a worse failure for an assistant than omitting an option. Ambiguity is not neutral. It removes you.
What you are seeingWhat it actually means
AI describes you as a category you left 2 years ago
What it actually means: →The old association is better corroborated than the new one
Your positioning line appears nowhere in AI descriptions
What it actually means: →It is a slogan rather than a classifiable fact
AI recommends you to the wrong buyer size
What it actually means: →Nothing on the public record states who you are not for
2 review platforms list you under different categories
What it actually means: →The system has conflicting evidence and picks the majority
Everyone in your category sounds identical in AI answers
What it actually means: →The sameness problem below
Give the model a reason to prefer you
If 12 brands in a category describe themselves as an AI-powered platform for modern teams, no model has a basis to prefer any of them. This failure gets mistaken for a visibility problem when it is a legibility problem.
Differentiation here is not brand voice but machine-readable difference, and 3 things create it.
A named constraint
Example of the shapeBuilt for organizations running 4 or more physical sites
Why it worksA structural fact a model can match against a query
A named exclusion
Example of the shapeNot suitable for single-location businesses
Why it worksReduces wrong recommendations, which increases confidence in the right ones
A named comparison position
Example of the shapeChosen over [competitor] when implementation speed matters more than customization depth
Why it worksGives the system the fit criterion it needs to justify naming you
The exclusion is the one teams resist and the one that pays. A system deciding whether to recommend you is looking for reasons to be confident. A brand that states who it is not for is easier to place than a brand that claims to serve everyone.
Entity consistency across the public record
AI systems resolve your brand by corroborating facts across sources. Contradictions between them cost you more than any single weak source.
Audit these and record what each currently says.
Your brandYour websiteOrganization schema✕Review platformsLinkedIn and CrunchbaseWikipedia and WikidataPartner listings
Your brand
Your website
Organization schema
Review platforms
LinkedIn and Crunchbase
Wikipedia and Wikidata
Partner listings
One review platform files you under a category you left. The system sees conflicting evidence and picks the majority.
Your website
What to checkLegal name, category descriptor, what you do in one sentence
Common failureMarketing language where a classifiable fact belongs
Organization schema
What to checkName, alternateName, sameAs links to every profile
Common failureMissing sameAs, so profiles are never connected to the entity
Review platforms
What to checkCategory assignment, description, company size served
Common failureCategory chosen years ago by whoever created the listing
LinkedIn and Crunchbase
What to checkIndustry, size, description, founding facts
Common failureStale after a repositioning
Wikipedia and Wikidata
What to checkExistence, accuracy, category
Common failureAbsent, or inaccurate and nobody watching
Partner and marketplace listings
What to checkProduct description, integration claims
Common failureWritten by the partner, never reviewed by you
Wikipedia and Wikidata deserve particular attention because they are weighted heavily in entity resolution across many systems. 2 rules apply. Notability thresholds are real and being a paying customer of a PR firm does not meet them, so a page created without independent coverage will be removed. And you should not edit your own entry, because conflict-of-interest edits are reverted and the reversion is public.
The Brand Truth Sheet
This is the phase output, and the source document every other asset inherits from. Fill it once, then make it the reference for your website, your review profiles, your partner listings, and your sales deck.
The last 2 rows are what separate a Truth Sheet from a positioning document. A claim with no publicly retrievable evidence behind it is a claim the system cannot corroborate, which returns you to Chapter 7.
Brand Truth Sheet
1Legal name and alternate namesEvery name you are known by, including former namesWorked example shape: Registered entity, trading name, common abbreviation
2Category, in the buyer's wordsThe term buyers search, not the one you inventedWorked example shape: Workforce scheduling software, not workforce experience platform
3What we do, one sentenceA classifiable fact with no adjectivesWorked example shape: Publishes and manages staff rotas across multiple physical sites
4Who we serveSize, sector, structural conditionWorked example shape: Multi-site operators, 200 to 2,000 staff, 4 or more locations
5Who we are not forThe exclusion, stated plainlyWorked example shape: Single-site businesses and organizations under 50 staff
6When we are the right choiceThe fit criterionWorked example shape: When scheduling spans sites and overtime is the cost being managed
7When a competitor is the better choiceNamed, with the reasonWorked example shape: When deep HR suite integration matters more than scheduling depth
8Pricing modelStructure and what drives itWorked example shape: Per site, not per seat. Tiered by number of locations
9Proof of each claimWhere the evidence lives publiclyWorked example shape: Customer case study, review platform, trade publication
When Clarity is good enough
AI describes your category, your buyer, and your fit criteria correctly in a clean session. The same facts appear consistently across your site, your schema, and your top 5 external profiles. Your exclusions are public. Then stop, and maintain it whenever the product or positioning moves.
05
Stage · Days 31 to 605 min
Chapter 5: Brand relevance
Do you provide the information AI needs to answer your buyers' questions?
Relevance is not about being mentioned but about supplying the evidence a system needs while it works through the fan-out. Most B2B sites answer the question the buyer asked while missing the 6 the system actually ran.
What you are seeingWhat it actually means
You appear for brand queries and nothing else
What it actually means: →You have answered questions about you, not questions about the category
Your comparison page exists and never gets cited
What it actually means: →It compares features rather than deciding fit
Pricing is available on request
What it actually means: →You are absent from every budget-qualified prompt
Your best content is a gated PDF
What it actually means: →It cannot be retrieved, so it cannot be used
Content volume is up and citations are flat
What it actually means: →Volume is the weakest signal in the evidence in Chapter 6
The 6 prompt families
Buyers do not ask one kind of question. Map coverage across all 6, because a gap in any one removes you from that branch of the fan-out.
Category
The shape of the questionBest tools for [job]
What it needs from youA page that defines the category and states your fit within it
Use case
The shape of the questionHow to [outcome] for [audience]
What it needs from youEvidence that you solve that specific job, with a named result
Problem
The shape of the questionWhy is [pain] happening and what fixes it
What it needs from youDiagnostic content that names causes before naming products
Comparison
The shape of the question[Competitor] alternatives, X versus Y
What it needs from youFit criteria, migration facts, and where you lose
Integration
The shape of the questionDoes it work with [system]
What it needs from youA retrievable list, not a logo grid image
Budget
The shape of the questionPricing, cost, cheapest option for [size]
What it needs from youReal figures or a real structure, publicly readable
Work one fan-out end to end
Take one high-value prompt and write down every sub-question a system would need to answer it. Then check whether you supply each one in retrievable form. This is the exercise that turns Relevance from a content calendar into a gap list.
Sub-question the system runsDo we answer itWhereRetrievable
What are the leading tools in this categoryDo we answer it: PartiallyWhere: Blog post, not category-definingRetrievable: Yes
Which ones suit companies of this sizeDo we answer it: NoWhere: NowhereRetrievable: No
What does each costDo we answer it: NoWhere: Pricing on requestRetrievable: No
Does it integrate with [system]Do we answer it: YesWhere: Integrations pageRetrievable: No, list is JavaScript
How long does implementation takeDo we answer it: NoWhere: Sales conversation onlyRetrievable: No
What do customers say about supportDo we answer it: PartiallyWhere: Review platform, not our siteRetrievable: Yes, third party
How hard is migration from [competitor]Do we answer it: NoWhere: NowhereRetrievable: No
4 of 7 sub-questions unanswered is a normal first result, and it explains absence better than any ranking report. The gaps become your content brief, ordered by how often each sub-question appears across your priority prompts.
What the citation evidence says about format
Page format changes citation odds independent of quality, and 3 findings are directly actionable.
Listicles earn 21.9% of citations, against 16.7% for standard articles and 13.7% for product pages
SourceWix research
What to do with itStructure category content as evaluative lists rather than essays
Adding statistics lifts AI visibility by roughly 30 to 40 percent
SourcePrinceton generative engine optimization research
What to do with itEvery priority page needs sourced data, not adjectives
44 percent of ChatGPT citations come from the first third of the source
SourceSearch Engine Land, February 2026
What to do with itPut the answer near the top. Long preambles push your best passage out of the cited zone
The listicle finding has a second implication larger than the first. If evaluative lists are the most-cited format, the strongest asset in your category is a best-tools page you do not own. Getting into it is a placement problem rather than a content problem. Chapter 6 covers that play.
Make facts extractable
A fact that exists only inside an image, a PDF, a video, a tab, or an accordion is a fact the system cannot use. 3 rules cover most of it.
State facts as sentences in the visible HTML, not as design elements. Answer the question in the first paragraph under its heading rather than building to it. And give every claim a number where a number exists, because a specific figure is corroborable and an adjective is not.
The Prompt-to-Evidence Map
This is the phase output, with one row per priority prompt showing what the system needs and whether you supply it.
Why switch, migration effort, fit difference, pricing
Comparison page
No migration timeline, no honest loss case
High
How to reduce overtime across sites
Problem
Causes, approaches, tools, evidence
Nowhere
Entire family absent
High
Does [category] integrate with [payroll system]
Integration
Integration existence, depth, setup effort
Integrations page
Not retrievable, list is JavaScript
Medium
4 rows show a finished map. The highest-intent families are usually the least covered, because comparison and problem questions were never treated as content.
When Relevance is good enough
Every one of the 6 families has at least one retrievable asset. A worked fan-out on your top 3 prompts shows no unanswered sub-question. Pricing and integrations are readable without JavaScript. Then move on, and revisit when the product changes.
06
Stage · Days 31 to 609 min
Chapter 6: Brand authority
Does the wider information ecosystem treat your company as a credible option?
This is the chapter the playbook is actually about. It has the highest ceiling, the longest timeline, and the strongest evidence behind it, and it is the layer most B2B teams fund last.
What you are seeingWhat it actually means
The AI visibility plan is a website project
What it actually means: →Your own domain is a small fraction of what gets cited about you
Success is reported as backlinks and domain rating
What it actually means: →You are tracking 2 of the weakest signals in the published research
Competitors appear in category answers and you do not
What it actually means: →They are present in the sources being retrieved and you are not
The content team publishes more to compete
What it actually means: →Content volume is the weakest correlate measured
You have no idea which sources AI keeps citing
What it actually means: →The Retrieval Map below is a 2-hour exercise nobody has run
Your website is a minority of the evidence
A brand's own domain accounts for a small share of the sources AI systems reference when answering category questions. Analyses aggregating this question put earned media at the large majority of citations.
Read that against how most B2B teams allocate. The website gets the redesign, the content calendar, the schema project, and the technical audit. Everything else gets a quarterly PR retainer, if anything.
This does not make the site unimportant. It makes it a qualifying condition, which is what Chapters 3 and 4 exist to satisfy. Passing that gate is a weeks-long project. This chapter is a permanent program.
What the evidence says correlates with visibility
Ahrefs analyzed 75,000 brands across ChatGPT, Google AI Mode, and AI Overviews. The sample was restricted to domains with a Domain Rating above 40 and at least one keyword drawing 800 or more monthly searches.
The study uses Spearman correlation, which measures how consistently 2 things rise and fall together. A value of 1 means they move in perfect step. It is not a percentage and it is not a causal claim.
How strongly each signal moves with AI visibilitySpearman correlation, 0 to 1
The coefficients are the technical answer. The same study produces a plainer one. Brands in the top quartile for web mentions earn up to 10 times more AI Overview mentions than brands in the next quartile down.
Backlinks and content volume, the 2 things most B2B marketing budgets are built around, sit at the bottom of an 8-row table.
Why links still matter, just not the way you think
Taken alone the 0.218 figure invites a wrong conclusion. Seer Interactive found a 0.70 correlation between a brand being mentioned on highly linked pages and its AI Overview visibility.
That reconciles the 2. Links do not earn citations directly. They determine which pages accumulate the authority that makes AI systems willing to cite them, and your brand needs to be mentioned on those pages.
Stop building links to your own domain as the objective and start earning placement on the pages that already have them. A mention inside a well-linked comparison article outperforms a link from a page nothing cites, and the 2 require different work.
Ahrefs also found that 38 percent of AI Overview citations come from pages already ranking in Google's top 10. Classic search performance remains a live input rather than a superseded discipline.
Where these figures stop being reliable
Correlation, not causation
Why it matters: Mentions are partly a proxy for being a real company people discuss. The marker and the mechanism are entangled
Sample is established brands
Why it matters: Domain Rating above 40 and keywords with 800 or more monthly searches. It says little about early-stage brands
English-language skew
Why it matters: The datasets and surfaces are English-dominated
Visibility is not revenue
Why it matters: Being named and being bought are different outcomes
A 2026 snapshot
Why it matters: The ordering is more durable than the figures, because these systems reason over entities rather than link graphs
Ahrefs describes every factor in their own study as showing moderate to very weak correlations, and say so in the write-up. Every competing guide quotes 0.664 as if it were a law. The researchers do not.
Find the sources that actually get retrieved
Before choosing targets, find out which sources your category's answers are built from. Run your priority prompts with search enabled and log every cited domain.
=Source concentration = citations from the top 5 domains / all citations logged
In most B2B categories that figure is high, which is the good news. You do not have 500 targets. You have 5 to 15, and they repeat.
How long presence takes, by source
Partner and marketplace listingsRelationships you already have, usually unclaimed
Days to weeks
Days to weeks
Review platformsVolume, recency, and depth of reviews
Weeks to months
Weeks to months
Comparison and best-of listsOutreach, or meeting the author's criteria
Weeks to a quarter
Weeks to a quarter
Trade publicationsOriginal data, commentary, or a news hook
1 to 2 quarters
1 to 2 quarters
Communities and forumsNamed people, not brand accounts
Quarters, cannot be rushed
Quarters, cannot be rushed
YouTube and podcastsAppearances, demos, and transcripts
Quarters, highest correlated signal
Quarters, highest correlated signal
DaysWeeksMonthsQuartersYears
Partner listings are the row worth checking first. Most B2B companies already have integration partners with directory pages they have never optimized, and those pages sit on domains with existing authority.
Where B2B citations actually come from
The most-quoted statistic in this field is wrong for your category, and repeating it will send your program in the wrong direction.
Aggregate studies synthesizing hundreds of millions of citations put Reddit at roughly 40 percent of AI citations and the top 15 domains at around 68 percent. Those datasets are weighted toward consumer queries. In a study restricted to B2B prompts covering 115,843 citations, Reddit accounted for roughly 0.7 percent. The top 10 domains covered only 11 to 13 percent across more than 7,000 unique cited domains.
The same distinction changes what "concentrated" means. Consumer AI citations are concentrated. B2B AI citations are a long tail, which is why the Retrieval Map below is built from your own prompts rather than from anyone's published list.
Software directories
What it looks like in B2BG2, Capterra, Clutch, TrustRadius
Why it earns citationsLead B2B verticals in citation share, often by a wide margin. They carry structured category, pricing, and review data in one place
Professional networks
What it looks like in B2BLinkedIn
Why it earns citationsRanks as the most-cited domain for professional and career-adjacent queries in several studies
Reference
What it looks like in B2BWikipedia
Why it earns citationsHeavily weighted, and disproportionately so in ChatGPT
Video
What it looks like in B2BYouTube
Why it earns citationsThe strongest correlated signal in the Ahrefs data, and transcripts are indexable text
Trade publications and newsletters
What it looks like in B2BCategory-specific titles with structured archives
Why it earns citationsSmall audiences, disproportionate citation weight, because archives are well-structured and authors are named
Communities
What it looks like in B2BReddit, Stack Overflow, specialist forums
Why it earns citationsDominant in consumer categories, marginal in most B2B ones. Verify before investing
Software directories are the row to act on first. They lead B2B citation share, and they carry the category and pricing data a fan-out needs. Your presence on them is a review generation problem rather than an outreach problem, which makes it the one authority signal you can create on demand.
Reddit is the row to verify before funding. Run your own prompts, count how often it appears, and decide from that rather than from a headline figure drawn from consumer queries.
The Retrieval Map
This is the phase output, with one row per source that appeared in your logged citations and your presence and your competitor's side by side.
SourceTimes cited in our promptsAre we presentIs [competitor] presentWhat presence requiresOwner
Review platform ATimes cited in our prompts: 14Are we present: Yes, 9 reviews, oldest 2023Is [competitor] present: Yes, 60 reviewsWhat presence requires: A review generation programOwner: Marketing
Best-tools listicle, publication BTimes cited in our prompts: 11Are we present: NoIs [competitor] present: Yes, listed secondWhat presence requires: Outreach to the author, with a fit criterion they lackOwner: Marketing
Community thread, forum CTimes cited in our prompts: 8Are we present: Mentioned once, negativelyIs [competitor] present: Mentioned favorably, 4 timesWhat presence requires: Named people participating, not a brand accountOwner: Product and support
Partner directory DTimes cited in our prompts: 6Are we present: Listed, description from 2022Is [competitor] present: Listed, currentWhat presence requires: Update the listing. Nobody owns itOwner: Partnerships
Trade publication ETimes cited in our prompts: 5Are we present: NoIs [competitor] present: Quoted twiceWhat presence requires: Original data they cannot get elsewhereOwner: Marketing
5 rows show a finished map. Those rows show the 4 shapes a target takes. One is free, one is a program, one cannot be bought, and one requires creating something before anyone will publish you.
=Authority gap = sources where a competitor appears and you do not / total sources logged
That figure is your target list, ordered by citation frequency, and it converts an abstract authority problem into a finite list of named places.
The listicle placement play
If evaluative lists are the most-cited format and you do not own them, getting into them is the single most direct authority move available in B2B.
1
Log every best-tools and comparison page cited in your prompts. These are the targets
2
Read each one and find the inclusion criteria the author used, which are usually stated or inferable
3
Identify the segment the list underserves. Most lists are written for a generic buyer and miss a structural condition
4
Approach the author with the gap, not with your product. Offer the criterion they are missing and the evidence for it
5
Supply retrievable proof, since an author who cannot verify a claim will not publish it
6
Track re-crawl. A listicle updated today changes answers within weeks on live-retrieval surfaces
Step 3 is the one that works. An author has no reason to add another generic vendor and every reason to add the one option that serves a buyer their list currently fails.
Cold start, when you have no coverage and no reviews
Early-stage brands are excluded from the evidence above, and the honest answer is that the fastest paths are the ones requiring no third-party permission.
1
Claim and complete every profile you are entitled to, including partner directories and marketplaces
Why it comes here: Free, immediate, and usually unclaimed
2
Generate reviews from existing customers on the platforms your prompts cite
Why it comes here: The only authority signal you can create on demand
3
Publish original data nobody else has
Why it comes here: The one reason a publication covers a company nobody knows
4
Put named people into communities and on podcasts
Why it comes here: Slow, uncopyable, and the highest correlated signal
5
Approach listicle authors with a segment gap
Why it comes here: Requires steps 2 and 3 as evidence first
Do not start with digital PR outreach. A publication has no reason to cover a company with no reviews, no data, and no named experts, and the outreach burns the relationship you will need in step 5.
When Authority is good enough
You appear in the majority of the top 5 sources your prompts retrieve. Your authority gap figure is falling quarter on quarter. New third-party mentions appear without you commissioning each one.
This layer never finishes. It compounds, and it decays if abandoned.
07
Stage · Days 31 to 604 min
Chapter 7: Brand trust
Is there enough independent evidence for AI to recommend you confidently?
A system deciding whether to recommend you is managing risk, and mentioning you costs it nothing. Recommending you and being wrong costs it the user's confidence, which is why the gap between mentioned and recommended is almost always a corroboration gap rather than a visibility one.
What you are seeingWhat it actually means
You are mentioned but never recommended
What it actually means: →Your claims are not corroborated anywhere the system can check
You are named with a caveat attached
What it actually means: →The system found a contradiction or a gap and hedged
Your case studies are all logos and adjectives
What it actually means: →Nothing in them is verifiable, so nothing in them is usable
Reviews are strong but 3 years old
What it actually means: →Recency decays, and stale proof reads as a company that stopped growing
Competitors get recommended with less product
What it actually means: →They have more evidence, not more capability
The proof ladder
Evidence has 5 levels and only the top 2 survive corroboration.
Only the top 2 levels survive corroboration
1A customer logo
Whether a system can use it: No, because it asserts a relationship and nothing about outcome
2A testimonial quote
Whether a system can use it: Barely, since it is attributable to a person but the claim is unverifiable
3A specific result stated as a number
Whether a system can use it: Yes, if the number is attached to a named customer and context
4A contextual case study naming the situation, action, and outcome
Whether a system can use it: Yes, and it supplies fit criteria as a byproduct
5Independent confirmation on a source you do not control
Whether a system can use it: Strongest, because corroboration requires a second party
B2B proof libraries cluster at levels 1 and 2, which is why a company with good outcomes gets described as an option rather than a recommendation. Moving from level 2 to level 4 is a writing exercise. Moving from 4 to 5 is Chapter 6.
Claims need evidence that exists in public
The test is not whether a claim is true but whether a system can verify it without asking you.
=Corroboration rate = material claims with public independent evidence / all material claims
Count only claims that would influence a buying decision. Positioning, outcomes, integrations, security posture, implementation time, and pricing structure all count. Adjectives do not.
This is the phase output, with one row per material claim, where the evidence lives and whether anyone outside your company confirms it.
Claim we makeWhere we state itPublic evidenceIndependentGapOwner
Implementation takes 2 weeksWhere we state it: Homepage, sales deckPublic evidence: One case study, oursIndependent: NoGap: Needs a customer saying it on a review platformOwner: Customer success
Works with [payroll system]Where we state it: Integrations pagePublic evidence: Partner directory listingIndependent: YesGap: NoneOwner: None
Cuts overtime for multi-site operatorsWhere we state it: Homepage headlinePublic evidence: NothingIndependent: NoGap: Needs a named customer with a numberOwner: Marketing
Rated well for supportWhere we state it: Website testimonialsPublic evidence: Review platform averageIndependent: YesGap: Recency, newest review is 14 months oldOwner: Customer success
SOC 2 compliantWhere we state it: Security pagePublic evidence: Auditor listingIndependent: YesGap: NoneOwner: None
5 rows show a finished matrix. The pattern is consistent. Compliance and integration claims are well corroborated because a third party issued them, and the commercial claims that drive the buying decision are corroborated by nobody.
Fit and limitations increase recommendation odds
Stating who you are not for, and where a competitor is stronger, appears to reduce recommendations and does the opposite. A system looking for confidence gains it from a source that demonstrates it is not overselling. Your fit criteria give it the justification it needs to name you for a specific buyer.
This connects directly to the exclusion row in your Brand Truth Sheet. Clarity states the limitation, Trust corroborates it, and the 2 together are what turn a mention into a recommendation.
Evidence has a shelf life
Trust decays, because reviews age, case studies date, and a proof library that stopped growing 2 years ago describes a company that stopped growing 2 years ago.
Set a floor rather than a target, so every priority claim has at least one piece of public evidence from the last 12 months. The review platforms your prompts cite should show a steady flow rather than a burst followed by silence.
When Trust is good enough
Your corroboration rate on material claims is high and rising. Every commercial claim has at least one independent source. Recent evidence exists on the platforms your prompts actually retrieve. Your limitations are public.
08
Stage · Days 61 to 905 min
Chapter 8: Operating the system
Visibility work decays. Models retrain, sources get republished, competitors publish, and your own site changes without anyone connecting the release to the outcome.
The platforms are not one audience
Google AI Overviews and AI Mode
What the evidence showsStrongest correlation with branded authority signals. 38 percent of citations from pages ranking in the top 10
Tactical consequenceClassic SEO transfers most directly. Rankings still matter
ChatGPT
What the evidence showsWeakest correlation with classic authority metrics, with Domain Rating around 0.27
Tactical consequenceDomain authority buys less. Smaller brands often find this the most accessible surface
Perplexity
What the evidence showsCites in the large majority of responses
Tactical consequenceThe most measurable surface, and the best place to start prompt tracking
Claude
What the evidence showsNot covered by the major correlation studies. Retrieves live and attributes sources when it does
Tactical consequenceTrack it separately rather than assuming ChatGPT findings transfer. Weight it by whether your buyers are technical
Claude is the surface most often left out of tracking, and in technical and developer-adjacent B2B categories that omission removes a meaningful share of your buying audience from measurement. Add it to the prompt set even if you cannot yet explain its behavior, because you cannot manage a surface you never sample.
If your buyers research in ChatGPT and your agency reports domain rating improvements, those 2 facts are barely related.
Measurement that survives volatility
A single AI answer is a sample rather than a ranking, and 4 rules make the data usable.
5 runs minimum per prompt, per platform
Why: One answer tells you nothing about the distribution
Clean sessions, memory off, logged out
Why: Assistant memory contaminates results and makes them unrepeatable
At least 2 platforms
Why: The surfaces weight signals differently enough to disagree
Same prompt wording every time
Why: Rewording changes the fan-out, so you are measuring a different question
30%Share of consideration setFormula: Times named / times runWorked example: Named in 3 of 10 runs, so 30%What it tells youThe headline number. Everything else explains it
33%Recommendation rateFormula: Times recommended / times namedWorked example: Named 3 times, recommended once, so 33%What it tells youWhether your problem is visibility or trust
71%Source concentrationFormula: Citations from top 5 domains / all citationsWorked example: 34 of 48 citations from 5 domains, so 71%What it tells youHow short your target list actually is
60%Authority gapFormula: Sources where a competitor appears and you do not / sources loggedWorked example: Competitor present on 9 of 15 sources you are absent from, so 60%What it tells youYour prioritized target list
40%Corroboration rateFormula: Claims with public independent evidence / material claimsWorked example: 4 of 10 claims independently evidenced, so 40%What it tells youWhether Chapter 7 is done
Recommendation rate is the diagnostic that separates 2 opposite problems hiding behind the same headline number.
Visible and not trusted
Named8 of 10
Recommended1 of 8
Where to work: Chapter 7
Trusted and not visible
Named1 of 10
Recommended1 of 1
Where to work: Chapter 6
Neither
Named1 of 10
Recommended0 of 1
Where to work: Chapter 4 first, then 6
A brand at 80 percent share and 12 percent recommendation rate does not have a visibility problem, and buying more coverage will not fix it.
Run these on a schedule rather than when someone remembers
Weekly
No scheduled work. This layer does not move weekly and checking it produces noise
Monthly
Run the priority prompt set, record share of consideration set and recommendation rate, log newly cited sources
Quarterly
Rebuild the Retrieval Map, recalculate authority gap and corroboration rate, re-run the 10-URL audit, review the Brand Truth Sheet against current positioning
On every site release
Re-run the curl test on the 10 priority URLs, because a release is the most common cause of a new 403 or a new rendering failure
On repositioning or launch
Rebuild the Brand Truth Sheet first, then propagate it to every external profile before publishing anything
The release trigger is the one teams skip and the one that causes the most avoidable damage. A deploy that moves pricing into a client-side component removes you from every budget prompt, and nobody finds out for a quarter.
Unowned work is why programs lapse in month 3
Marketing
What they ownPrompt tracking, Retrieval Map, content gaps, outreach
First sign it is unownedThe last prompt run is more than 6 weeks old
Engineering
What they ownCrawler access, rendering, schema, release checks
First sign it is unownedA release shipped without the curl test running
Security or infrastructure
What they ownBot management policy
First sign it is unownedNobody can name who approved the current WAF rules
Product
What they ownFactual accuracy of product claims
First sign it is unownedThe Truth Sheet still describes a deprecated feature
Customer success
What they ownReview generation, case study evidence
First sign it is unownedNewest public review is more than 6 months old
Leadership
What they ownWhich layer gets funded
First sign it is unownedBudget sits on Discoverability while Authority is unfunded
=Program survival rate = recurring items with a named owner / total recurring items
Anything below 1.0 means the program depends on the person who started it. Count the recurring items in the cadence table above, count how many have a name written next to them, and divide.
Security or infrastructure is the row left unassigned most often, and it is the one that removes you from every AI surface with a configuration change nobody reviews.
2 functions here sit outside marketing, which is the structural reason this work stalls. Neither engineering nor security has AI visibility in their objectives, so the request arrives as a favor rather than a ticket. Frame both as revenue items with a named commercial consequence, because a 403 on your pricing page is not an SEO issue.
09
6 min
Chapter 9: The 90-day plan
The action closing each chapter appears here in sequence, gated on a finding rather than a date, so the plan paces itself against your brand rather than against a calendar.
Before you start, set the target the route test in Chapter 1 says is achievable.
Parametric category
Little to none
What success looks like instead: →A working system, a target list, and a corrected description of your brand
Retrieval dominant
Measurable on live-retrieval surfaces
What success looks like instead: →Share of consideration set moves on Perplexity and AI Mode before ChatGPT
Mixed, the common B2B case
Movement on retrieval surfaces only
What success looks like instead: →Retrieval slots gained, memory slots unchanged
=Day 90 target = day 10 share of consideration set + movement your route test permits
Setting that expectation on day one is what keeps the program funded through quarter 2. A parametric category that produces no headline movement in 90 days has not failed, and a team that promised movement has.
1
Run the route test. 5 prompts, clean sessions, search off then on
Move on whenYou know which decision-relevant facts are invisible to a non-rendering system
Steps 1 through 5 take under a day of working time between them and they determine everything that follows. Do not skip to tactics before they are complete, because the most common expensive mistake in this discipline is funding Authority for a brand whose site returns 403s.
6
Fix crawler access. Allow search and user-triggered bots at the firewall, make the training bot decision deliberately, document both
Move on whenThe same facts appear consistently everywhere a system will check
Steps 6 and 7 are engineering work with a marketing consequence, and the sequence stalls here more than anywhere else. Raise them as revenue tickets rather than SEO requests.
Step 9 is the highest return per hour in the whole plan. It costs a week and it fixes a class of problem that no amount of content or outreach can reach.
11
Log every cited source across your priority prompts. Calculate source concentration and authority gap
Move on whenThe program survives the person who started it
Step 19 gets skipped more than any other. Without an identical re-run under the same conditions you have no evidence the program worked and no basis to ask for the next quarter's budget.
What 90 days can and cannot deliver
✓Realistic in 90 days✕Not realistic in 90 daysWhat the strategy should be
✓Realistic in 90 days: Fixing every discoverability failure
✕Not realistic in 90 days: Changing what a trained model already believes about your category
What the strategy should be: Close the gate fast and stop. Discoverability does not compound, so anything past a clean audit is spend with no return
✓Realistic in 90 days: Correcting how AI describes you, where retrieval dominates
✕Not realistic in 90 days: Displacing an incumbent from parametric memory
What the strategy should be: Compete for the retrieval slots rather than the memory slots. Fix the Truth Sheet, then get it onto the sources your prompts already retrieve
✓Realistic in 90 days: Appearing in sources you already had a right to
✕Not realistic in 90 days: Building the kind of earned coverage that compounds over years
What the strategy should be: Take everything free first. Directories, partner listings, marketplaces, and review platforms need permission from nobody
✓Realistic in 90 days: Measurable movement in recommendation rate
✕Not realistic in 90 days: Measurable movement in a parametric category
What the strategy should be: Report recommendation rate rather than share of consideration set in quarter one, because it moves on evidence you can produce yourself
✓Realistic in 90 days: A target list, an owner, and a cadence
✕Not realistic in 90 days: A finished authority position
What the strategy should be: Treat quarter one as building the machine and quarter 2 onward as running it. Budget accordingly
If your route test returns a parametric category, the first 90 days produce a working system and very little movement in the headline number. That is the correct outcome for that starting position, and the third column is what you report against instead.
10
2 min
Chapter 10: The final readiness checklist
This is one instrument covering all 5 layers. Every line is a yes or no with a name next to it. Anything you cannot answer yes to is either work in progress or work nobody owns.
All 10 priority URLs return 200 to Googlebot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot, and Bingbot, tested from outside your network
Content and decision-relevant facts survive with JavaScript disabled
Crawler policy separates training bots from search bots, deliberately and in writing
Priority pages are indexed
Organization schema exists with sameAs links to every external profile
No markup contradicts the visible page
Server logs show AI crawlers arriving
No decision-relevant fact sits behind a form
Someone owns re-running the curl test on every release
AI describes your category, buyer, and fit criteria correctly in a clean session with memory off
The Brand Truth Sheet is complete, including who you are not for
The Truth Sheet names when a competitor is the better choice
The same facts appear across your site, schema, review listings, professional profiles, and partner listings
Every claim in the Truth Sheet names where its public evidence lives
Someone owns updating it when the product or positioning moves
All 6 prompt families have at least one retrievable asset
A worked fan-out on your top 3 prompts shows no unanswered sub-question
Pricing is readable without JavaScript
Integrations are readable without JavaScript
Priority pages answer the question in the first third
Priority pages carry sourced data rather than adjectives
Category content is structured as evaluative comparison rather than as an essay
Every source cited across your priority prompts is logged
Source concentration and authority gap are calculated
Every profile you are entitled to is claimed and current
A review generation program is running on the platforms your prompts cite
You appear in the majority of your top 5 retrieved sources
Authority gap is falling quarter on quarter
New third-party mentions appear without you commissioning each one
Every material commercial claim has public independent evidence
Corroboration rate is measured and rising
Each priority claim has public evidence less than 12 months old
Your limitations and fit criteria are public
Proof has moved from logos and testimonials to named situations with figures
Monthly prompt tracking runs at 5 runs per prompt across at least 2 platforms
Conditions are identical every time, with memory off and logged out
Claude is in the prompt set, not only ChatGPT and Perplexity
The Retrieval Map, authority gap, and corroboration rate are rebuilt quarterly
The 10-URL audit re-runs on every site release
Program survival rate is 1.0, meaning every recurring item has a named owner
Security or infrastructure knows that bot management is a revenue decision
What this comes down to
You cannot control what an AI system says about you. You can control how much accurate, well-evidenced, publicly available information exists for it to work from.
Every layer here does the same thing in a different place. Make the facts findable, unambiguous and relevant to the question being asked. Get them corroborated by people who are not you, and make the limitations as public as the strengths.
Brands that do that get recommended. Brands that publish more and hope do not.
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.