Skip to content

Checks

UTM check

Checks your UTMs against GA4's channel rules, finds paid traffic reported as organic and everything in Unassigned, then writes one naming scheme with an old to new map.

Other tools

Add this skill to Claude

Claude takes skills as a ZIP file. Your download has started; if it didn't, use the button below.

  1. 1In Claude, open Customize, then Skills.
  2. 2Choose +, then Upload a skill, and pick utm-check.zip.
  3. 3Ask for it by name, or let Claude use it when a task matches.

Skills need code execution turned on in Claude's settings. On a Team or Enterprise plan, an admin can share it with everyone.

Download the folder

Try it now, without installing

See the skill work on a made-up input before you add it. Nothing is installed, and you can swap in your own input.

Open in the Claude app

Opens a new chat in the Claude desktop app with the prompt filled in. Nothing is sent until you press enter.

Includes the whole SKILL.md and the sample input. Paste it into Claude, ChatGPT or any chat.

Influenced by
Default channel group
Needs
Your UTMs or a GA4 traffic acquisition export
Works with
Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md
Version
1.0.0 · Last checked · License MIT
Maintained by
Sabih AhmedMarkOps
SkillMarkOps

UTM check

Example output

Where your links land
Source / mediumLands inCall
linkedin / paid-socialPaid SocialKeep
linkedin / social_paidOrganic SocialFix
newsletter / newsletterUnassignedUse email

Checks your UTMs against GA4's channel rules, finds paid traffic reported as organic and everything in Unassigned, then writes one naming scheme with an old to new map.

The skill

The whole skill, in plain text.

Nothing is hidden. Read it, change it to fit your team, or take the whole folder.

Scripts
Optional Python scripts for the counting. It works without them and says which way it ran.
Network
Opens web pages or your tools if your agent can. Works from pasted data if not.

3 test cases

utm-check/SKILL.md

In the folder

  • utm-check/SKILL.md
  • utm-check/references/ga4-channel-rules.md
  • utm-check/scripts/channel_check.py
---name: utm-checkdescription: >-  Checks a B2B SaaS company's UTMs, or a GA4 traffic acquisition export, against GA4's default channel rules in the  order GA4 applies them, finds paid traffic that lands in organic channels and everything that falls into  Unassigned, flags naming that splits one channel into several rows, and writes one naming scheme with an old to new  map. Returns PASS or FAIL with severities. Use it before a quarter's campaigns launch, when a channel report looks  wrong, or when Unassigned grows. Not for conversion tracking or offline imports (use tracking-audit) or for  pipeline by channel (use pipeline-report).license: MITcompatibility: >-  Works in Claude, ChatGPT, Codex, Cursor and any agent that reads SKILL.md. Reads a connected GA4 property if the  agent has one, otherwise an export or pasted rows. Python 3 is optional, for the channel check.metadata:  author: SHAKEUP  version: "1.0.0"  stage: check  plugin: markops  homepage: https://www.shakeuphq.com/skills/utm-check--- # UTM check Shows where each tagged link really lands in GA4, and gives the team one naming scheme that lands every channel where it belongs. ## Purpose - One job: check UTM values against GA4's default channel group and write one scheme that fixes what fails.- Stage: check. It ends in a verdict with severities, plus the scheme as its main output.- It never edits an ad, a link, a GA4 setting or a CRM field. It writes the map; people change the values.- It never guesses traffic. Sessions come from the reader's export; a link list without sessions counts each link once and says so. ## Use when / not | Situation | Use || --- | --- || Before a quarter's campaigns launch | this skill, on the planned links || A channel report looks wrong, or Unassigned grows | this skill, on the GA4 export || A new tool, agency or team starts tagging links | this skill, then share the scheme || Conversions, enhanced conversions, consent or imports | `tracking-audit` || Pipeline per channel from the CRM | `pipeline-report`, which reads the scheme this skill writes | ## Inputs | Input | Required | Ways to provide, in order | If missing || --- | --- | --- | --- || UTM rows: source, medium, campaign | yes | a connected GA4 property (Traffic acquisition: session source / medium, session campaign, sessions, session default channel group, last 90 days); a CSV export; pasted rows; the planned links or each platform's tracking template | ask once || Sessions per row | no | the same GA4 export | count each row once; Unassigned share by rows, labelled || GA4's reported channel per row | no | the same export | predicted channels only; mismatches not checked || The channels the reader reports on | no | `## Channels` in the profile; the reader | Unassigned compared with every channel instead || An existing naming rule | no | an accepted `utm-check` learning; the reader's document | start from the scheme in `references/ga4-channel-rules.md` || CRM source and medium picklists | no | CRM settings; pasted values | the map covers GA4 only; open item | Use 90 days of GA4 data by default: one quarter of campaigns, long enough for a monthly newsletter to appear 3 times. ## Profile read From `company-profile.md`: `## Channels` (active and testing channels, to map onto GA4 channels), `## Accounts and data` (analytics tool, CRM, whether GA4 is connected), `## Company` (Brand terms, so the brand's own domain isn't read as a referral partner), and every `## Learnings` entry for `utm-check` with status `noted`, `fact` or `accepted` (an accepted scheme is the scheme this run checks against). No profile: ask for the channels the team runs and the analytics tool, in one message. Say that `company-profile` saves this next time. ## Steps 1. **Apply learnings.** An accepted scheme or an accepted exception ("we keep `newsletter` as the medium for our own list") overrides the defaults below. Name each one that fires. 2. **Map the reader's channels to GA4 channels.** From `## Channels`: Google Ads and Microsoft Ads to Paid Search; LinkedIn, Meta, Reddit and X ads to Paid Social; YouTube ads to Paid Video; SEO to Organic Search; founder and company posts to Organic Social; newsletters to Email; partners and review sites to Referral; podcasts to Audio. These are the channels the reader reports on. 3. **Classify every row** with the script:    ```   python3 scripts/channel_check.py --csv ga4_export.csv --report-channels "Paid Search,Paid Social,Organic Search,Email" \       --campaign-pattern "^\d{4}-q[1-4]_[a-z0-9-]+_[a-z0-9-]+$"   ```    Add a `paid` column (yes or no) when the reader says which links are ads: a medium like `video` doesn't show it.    Without Python, use the ordered table in `references/ga4-channel-rules.md`: lowercase the source, medium and campaign, test each rule from the top, and stop at the first match. Use the site lists there; a source not on them is `unknown_source`. Say which way you ran it. 4. **Record findings per row.**    | Finding | When | Severity | Why |   | --- | --- | --- | --- |   | `paid_outside_paid` | The medium or source says paid (paid, cpc, ppc, cpm, ads, sponsored, boosted, retargeting, promoted), or the reader says the link is for an ad, but the row lands outside the paid channels | critical | Paid spend shows as free traffic: organic looks better and paid looks worse |   | `unassigned` | The row lands in Unassigned | warning | Nobody reports on it |   | `paid_other_known` | Paid traffic from a known platform lands in Paid Other because the source value isn't on Google's list (`meta`, `google.com`, `x.com`) | warning | The platform's channel undercounts |   | `shop_trap` | The campaign name sends the row to a Shopping channel | warning | B2B traffic reported as shopping |   | `case` | Capitals in source, medium or campaign | warning | 2 rows for one value in every report |   | `spaces` | Spaces, "+" or "%20" in a value | warning | Values break in links and split rows |   | `no_campaign` | A paid row with no campaign name | warning | Spend can't be tied to a campaign |   | `campaign_pattern` | The campaign doesn't follow the scheme's pattern | warning | Reports can't filter by quarter, theme or audience |   | `split_source` | One platform under several source values | warning | One channel spread over several rows |   | `unknown_source` | The source isn't on the script's subset of Google's lists | note | Check it against Google's full list before trusting the channel |   | mismatch | GA4's reported channel differs from the predicted one | note | GA4 used a rule the values can't show (auto-tagging, its referrer lists) or the subset is missing the source | 5. **Test Unassigned against the channels that matter.**    | Condition | Severity | Why |   | --- | --- | --- |   | Unassigned sessions are at least the sessions of a channel the reader reports on | critical | Unassigned traffic could change which channel looks bigger |   | Unassigned above 0, smaller than every reported channel | warning | Small now, and it grows with each new tool |   | No Unassigned rows | pass | |    Rows the reader has accepted as known Unassigned (events, print) still count here, labelled. 6. **Write the scheme.** Start from the scheme in `references/ga4-channel-rules.md`. Keep every existing value that already lands in the right channel. Change only values with a critical or warning finding. Use the reader's campaign pattern if 80% or more of campaign rows already follow one pattern (most of the team already uses it); else use the scheme's `<year>-q<quarter>_<theme>_<audience>`. 7. **Write the map.** One row per value that changes: old source and medium, new source and medium, the channel it lands in now and after, the sessions affected, and where to change it (the ad platform's tracking template or URL options, the link builder, the email tool, the CRM picklist and hidden fields). 8. **Name what the scheme can't fix.** Traffic types with no default channel (events, print, QR codes), and channels the reader wants reported apart (ChatGPT Ads land in Paid Other): the default group can't be edited, so the fix is a rule in the property's primary channel group. Recommend it; don't make it. ## Check and verdict | Verdict | Condition (first match wins) || --- | --- || `BLOCKED` | No UTM rows after asking || `FAIL` | Any critical finding: paid traffic outside the paid channels, or Unassigned at least the size of a reported channel || `PASS WITH NOTES` | Warnings or notes only || `PASS` | No findings | Named stop: `BLOCKED` asks for the rows. No other stops: a FAIL still delivers the scheme and the map. ## Learn - The reader accepts the scheme: write a `fact` learning with the scheme's sources, mediums and campaign pattern in one line, so later runs check against it.- The reader keeps a value the check flagged ("our own newsletter stays `newsletter`"): write a `proposal` to accept it as known Unassigned.- A source GA4 classified differently from the script: write a `fact` naming the source and GA4's channel. Example: `- 2026-10-24 | utm-check | fact | Scheme accepted: linkedin/paid-social, google/cpc, newsletter/email, campaign <year>-q<n>_<theme>_<audience> | Why: reader confirmed in this run | Status: noted` ## Output ```Verdict: <PASS | PASS WITH NOTES | FAIL | BLOCKED> · Coverage: <complete | partial (no sessions, no reported channel)> · Check: <script | by hand>Rows: <n> · Sessions: <n or "not available"> · Unassigned: <n sessions, n% (or rows)>; reported channels not larger: <list or none>Counts: <n> critical, <n> warnings, <n> notes | Source / medium | Campaign | Sessions | Lands in | Finding | Severity | Fix |(critical first, then by sessions) Split sources: <platform: values>The scheme: <table: use, utm_source, utm_medium, GA4 channel; campaign pattern; content and term rules>The map: <table: old, new, channel now, channel after, sessions, where to change it>Can't be fixed by naming: <list, with the primary channel group rule to add>Learnings applied: <list, or none>``` Then the handoff block. ## Hand off ```handoffskill: utm-checkversion: 1.0.0verdict: FAILcoverage: completefor: <property or link list>, <period>outputs:  unassigned_share: 0.07  paid_outside_paid_sessions: 1240  scheme:    linkedin_ads: linkedin / paid-social    google_ads: google / cpc    newsletter: newsletter / email    campaign: <year>-q<quarter>_<theme>_<audience>  values_to_change: 6next: [pipeline-report, crm-feedback-loop]open: [CRM picklists not checked]``` `pipeline-report` reads `scheme` to group CRM sources the same way GA4 does. `self-reported-attribution` reads it to name the software channel beside each answer. ## Failure modes | Problem | Do this || --- | --- || The export has "Session source / medium" in one column | Split on " / "; the script does this || GA4's reported channel and the prediction disagree on auto-tagged Google Ads rows | Trust GA4: those rows use platform rules || The reader asks what share of Unassigned is normal | There's no such figure in this skill; compare with their own last 90 days and with their smallest reported channel || A source looks like a search or social site but isn't on the subset | Mark `unknown_source`; ask the reader to check Google's downloadable list, or pass it with `--sources` || The brand's own domain appears as a referral | Note it as self-referral (a cross-domain or payment step); it's a tracking setup issue for `tracking-audit` || Planned links only, no traffic yet | Check every link the same way; the Unassigned test uses rows, labelled | ## Influenced by - Google's documentation of the GA4 default channel group, the ordered rules and source lists this check applies: https://support.google.com/analytics/answer/9756891?hl=en- SHAKEUP's UTM builder naming rule: source and medium from fixed lists, campaign as quarter, theme and audience, and the same values in the CRM.

How it thinks

Inputs in, a verdict out, and a stop only where you decide.

What we built

We check UTMs against the rules GA4 actually applies, in its order, instead of a style guide. The scheme keeps what already works, because renaming working values breaks year-on-year reports.

It reads your company profile if you have one, or asks for only what it needs. Then it works through its steps, checks its own work and ends in a verdict.

The steps

  1. 01

    Read your learnings first

    A scheme your team accepted before becomes the rule this run checks against.

  2. 02

    Classify every row the way GA4 does

    Its script applies GA4's default channel rules in order, first match wins, so you see where each link really lands.

  3. 03

    Find what reports wrong

    Paid ads landing in organic channels, sources GA4 doesn't know, campaign names that trip the shopping rule, capitals that split one channel into 2 rows.

  4. 04

    Test Unassigned against your channels

    If Unassigned holds as many sessions as a channel you report on, it could change which channel looks bigger, and the check fails.

  5. 05

    Write one scheme and the map

    It keeps every value that already lands right, changes only what fails, and lists where each change happens, down to CRM picklists.

What you need

  • Required

    Your UTMs or a GA4 traffic acquisition export

    A connected GA4 property, last 90 days → A CSV export → Rows or planned links you paste

  • Optional

    Sessions and GA4's reported channel per row

    The same GA4 export

  • Optional

    The channels you report on

    The Channels section of your company profile → Your request

  • Optional

    Your existing naming rule

    A scheme your team accepted in an earlier run → Your document

What it can return

  • PASS

    Every row lands in the channel it should, with clean naming.

  • PASS WITH NOTES

    Naming issues or a small Unassigned share, but no paid traffic outside the paid channels.

  • FAIL

    Paid traffic lands in an organic channel or Unassigned, or Unassigned is as large as a channel you report on.

  • BLOCKED

    No UTMs or export after one question.

Where it stops for you

  • When there are no UTMs or export to check

Example output

What you get back.

A run for a made-up company. Your output depends on your inputs.

Here's a run for a made-up B2B SaaS company that runs LinkedIn Ads and Google Ads. Its channel report showed organic social growing while paid social stayed flat.

Verdict

FAIL

Coverage: complete

Check: script

Source / mediumSessionsLands inFindingFix
LinkedIn / social_paid610Organic SocialPaid ads outside the paid channelslinkedin / paid-social
meta / paid-social240Paid OtherGA4 doesn't know "meta"facebook / paid-social
newsletter / newsletter310UnassignedGA4 doesn't know this mediumnewsletter / email
partner / partner95UnassignedGA4 doesn't know this mediumpartner / referral
Unassigned

405 sessions, smaller than every channel the team reports on, so a warning, not a fail.

The scheme

linkedin / paid-social for LinkedIn Ads, google / cpc for Google Ads, newsletter / email for the newsletter. Campaigns read year, quarter, theme, then audience: 2026-q4_problem-ad_revops.

The handoff block the next skill reads
skill: utm-check
version: 1.0.0
verdict: FAIL
coverage: complete
for: GA4 property, July to September 2026
outputs:
  unassigned_share: 0.06
  paid_outside_paid_sessions: 610
  values_to_change: 4
next: [pipeline-report, crm-feedback-loop]
open: [CRM picklists not checked]

What good looks like

  • Every paid link lands in a paid channel
  • Unassigned compared with the channels you report on, not an average
  • One source value per platform, all lowercase
  • A scheme that changes only what fails
  • A map that says where each change is made

Install

Use it in the tool you already have.

Pick your tool. Each tab has the shortest path we know of.

  1. 01
    Download utm-check.zip.
  2. 02
    In Claude, open Customize, then Skills.
  3. 03
    Choose +, then Upload a skill, and pick the ZIP. Code execution must be on in your settings.
  4. 04
    Ask for it by name, or let Claude use it when a task matches.

Questions

Before you add it.

Why does paid traffic show up as organic?

GA4 reads a medium as paid only when it contains "cp" or starts with "paid". A medium like social_paid or sponsored doesn't, so a LinkedIn ad lands in Organic Social.

Why does "meta" land in Paid Other?

GA4 matches the source against Google's list of social sites, which holds facebook and instagram but not meta. Use facebook or instagram as the source.

Can it change GA4's channel rules?

The default channel group can't be edited. Where a channel needs its own row, such as ChatGPT Ads, it recommends a rule in your property's primary channel group.

Does it work on links we haven't launched?

Yes. It checks planned links the same way and counts each link once, since there are no sessions yet.