Data Infrastructure

Marketing tracking audit checklist: GA4, ads and CRM

A tracking audit reviews a live setup step by step: tags, consent, GA4, campaign tags, ad platform conversions, the CRM handoff, and one period of reconciled numbers.

Founder of MetricMaven. Previously managed Google Tag Manager and Google Analytics across 800+ sites at HCA Healthcare, and built attribution and eCommerce reporting at RYZE Superfoods and Mars.

A marketing tracking audit checks whether a running setup measures what the business needs, and whether the numbers agree across the tools that report them. Work through it in order: tags and consent, conversions and deduplication, campaign tags and click IDs, the CRM handoff, what the ad platforms learn from, and one closed period of reconciled counts. Record each finding with an owner and a fix.

Run this audit in Claude or ChatGPT

With Maven, your AI assistant can run the data checks in this list against your own numbers. Maven is the connector: through the MCP, it puts each client’s connected marketing data into Claude or ChatGPT.

A chat window with message bubbles and a magnifying glass, linked by a cord to a checklist card with four check marks and two flags.
The data checks run in your assistant and come back as a short list.

It works in three steps:

  1. Connect your sources. Connect GA4, Google Ads and the client’s other ad, analytics and CRM sources to their Maven workspace.
  2. Add the MCP to your assistant. It works in Claude, ChatGPT, Claude Code or Cursor.
  3. Ask the checks. Use the prompts in this post against the client’s own numbers.

Two things to know first:

  • Tag checks run in Maven. Connect the client’s Google Tag Manager, and Maven’s tracking audit finds broken, degraded and stalled pixels and events, tags with no trigger, and GA4 measurement-ID mismatches. Any container change is staged with the exact diff and QA evidence, and goes live only after you approve it.
  • The MCP checks data, not your website. It reads what your platforms report. It can’t see which tags fire before consent or whether a hidden form field is filled. For those, use the checklist.

Each checklist group below ends with a Check it with Maven prompt and the Maven operation that answers it.

What is a marketing tracking audit?

An audit looks at the setup as it is today. Tags added years ago, settings changed by a previous agency, and conversions nobody reviewed all show up here.

It is different from the tracking QA checklist before launch, which tests one release before it goes live. Use QA for each release. Use an audit to find the problems that built up between releases.

The marketing tracking audit checklist

Seven icon tiles strung on a cord: code with a shield, a target with a flag, a link with a tag, a contact card, a megaphone, a bar chart and a calendar.
The seven checklist groups, in the order to work through them.

Work through the groups in order. Each one depends on the one before it: there’s no point reconciling counts if one event reaches the same platform twice.

1. Tags and consent

Start with a complete list of where data goes. Most problems in a running setup come from tags added outside the tag manager, or from one event that reaches the same platform twice.

An auditor traces each website tag to its destination and flags duplicates.
  • Export every container in the tag manager. List each tag, trigger and variable with its status.
  • Search page templates and the website builder for hardcoded analytics or ad tags that bypass the container.
  • Write down every destination: GA4 properties, each ad account, Meta, the CRM, and any other endpoint.
  • Mark any tag with no current owner or purpose for removal after the audit, not before.
  • Test the consent banner in both directions. Load the site in a clean browser session and record which tags fire before any choice. Then accept, then reject, and compare each time. The default state and the reject path matter as much as the accept path.
  • Check that the choice is applied on every page, and that a visitor who rejects doesn’t later trigger a tag that ignores it.
  • Write down what each choice enables, so the next release can be checked against the same list.

For how Google’s consent signals should work with your tags, read the Consent Mode v2 explainer.

2. Conversions and deduplication

Check what counts as a conversion, and that each one is counted once.

  • Look for duplicate GA4 configuration tags and duplicate ad conversion tags. They double count when they fire together.
  • Confirm each GA4 key event is a real business outcome, such as a booked consultation or a purchase. A common finding is page_view or scroll marked as a key event, so every visit looks like a conversion.
  • If a Meta Pixel and the Conversions API both send the same event, check that they share the event name and event ID. Meta counts them once only when both match, and Meta’s deduplication guide says events are deduplicated only when received within 48 hours. See the Meta deduplication post.
  • Agree what counts as a conversion before you fix anything. The fixes depend on it.

GA4 also has a handful of settings that quietly change reported numbers. Check each one in the property’s admin area and keep a dated record of what was set.

What to checkHow to checkCommon finding
Internal trafficDefine the internal traffic rule, then check the data filter is Active, not only Testing.Filter left in Testing for months, so office and staff visits still count. Filters don’t change historical data.
Data retentionGA4 data retention settings. Standard properties can set event and user data retention to 2 or 14 months.Retention at the shortest setting, so explorations older than two months return no data.
PII in URLs and titlesCrawl key pages and read the page paths, titles and query strings sent to GA4.Email addresses or form values in query strings, which Google’s policy does not allow in GA4.

Verify each setting against the current Google help pages, since menu labels change. Google’s key events page explains how key events are set and the limit on how many a property can have.

3. UTMs, click IDs and channel grouping

Click one ad from each platform and follow it to the landing page, then to the form. Check that the click ID and campaign values survive the trip.

  • Confirm auto-tagging is on in Google Ads. It’s the setting that appends the Google Ads click ID to landing page URLs. The auto-tagging help page shows where to check.
  • Look for redirects that strip query parameters before the landing page loads.
  • Check UTM values for differences in case or spelling across ad sets, and for platforms that mix manual UTMs with auto-tagging. For a naming scheme that stays consistent, see dynamic UTM parameters.
  • Check cross-domain measurement. In the data stream’s tag settings, confirm every checkout or booking domain is listed, then follow a link between domains and look for the _gl parameter. If a domain is missing, sessions restart and paid traffic is credited to direct.
  • Check the unwanted referrals list. If a payment provider shows up as a referral, it takes credit for sales your campaigns drove. Exclusions don’t reach back in time. See Google’s unwanted referrals page.
  • Baseline your unknown or unattributed traffic share and your UTM coverage, so you can show the fixes worked.

4. CRM and offline conversions

Submit a test lead through each form, then find that lead in the CRM.

  • Check that campaign values and click IDs are stored on the CRM record, not only on the page URL. Hidden form fields are the usual place they’re kept.
  • If offline conversions are imported, confirm each imported stage matches a CRM record, and that the click ID is the one captured at submit.
  • Name a system of record for each metric, such as revenue, orders and channel spend.

The offline conversions guide covers the import side. The GoHighLevel offline conversions guide shows the same handoff in one CRM.

5. What the ad platforms learn from

Each ad account has its own list of conversion actions, and the settings decide what bidding learns from.

  • In Google Ads, a primary action in a standard goal is used for bidding when the campaign bids on that goal. A secondary action in a standard goal is reported but not used for bidding. Custom goals are the exception: their actions are used for bidding whether primary or secondary. See primary and secondary conversion actions.
  • Check each campaign’s selected standard and custom goals, as well as each action’s status.
  • Make sure duplicate actions for the same event can’t both feed bidding.
  • Check that real outcomes, such as qualified leads or closed deals, reach the platforms, not only form fills. For the Google side, see the Enhanced Conversions guide.

6. Reporting reconciliation

Pick a closed period, such as the last full calendar month. Pull the counts for the same conversion from each source: ad platform, GA4 and the CRM (or Shopify, for a store).

  • Decide before you look what gap you’ll accept, then investigate every difference above it.
  • Expect the numbers to differ. A gap with a known cause, such as attribution windows or consent, is different from one nobody can explain.
  • Record your server-side setup alongside the rest, since server-side tagging changes which fields reach each platform. The server-side tagging guide covers the approach and its common mistakes.

The guide to why Google Ads and GA4 conversions differ covers the most common causes. For stores, start with why Shopify and GA4 revenue doesn’t match.

7. Document it and repeat it

An audit is only useful if the next person can see what was checked and what changed. Keep one log with the date, the check, the expected result, the actual result, the owner and the fix. Repeat the same reconciliation each month, so drift shows up as a trend rather than a surprise.

Before you call the audit done, check you have:

  1. An inventory of tags, destinations and owners, dated.
  2. Consent behaviour recorded for accept, reject and no choice.
  3. GA4 key events, domains, filters, retention and PII findings logged.
  4. Click IDs and campaign values confirmed on the form and the CRM record.
  5. Conversion actions and each campaign’s selected goals confirmed in every ad account.
  6. One closed period reconciled, with each gap explained or assigned.
  7. A next review date, and a list of changes that trigger a re-run.

Keep it checked: schedule the daily brief

An audit is a snapshot. Tracking can break again the next time someone edits a template, tag or form. Maven’s daily brief catches that between reviews.

A checklist card in an open envelope with check marks, one flag and one paused item, beside a morning sun and a coffee cup.
The daily brief lists tracking issues first, then budget pacing.

Tracking health comes first: each issue labelled broken or stalled, with an owner, what changed and what to check. Then monthly budget pacing: spend to date, pace, projected month end, and over or under flags. A clean day says so. You can schedule it as a daily Claude task.

Your assistant makes two calls. First it checks each platform’s last successful sync, so it doesn’t brief on half-loaded data:

workspace tool: check freshness before reading the brief
{
  "target": {"workspace_id": "<workspace_id from workspace list>"},
  "operation": "freshness",
  "options": {},
  "context": {
    "user_request": "Is my tracking healthy today?",
    "agent_goal": "Check each platform's last successful sync before reading the brief.",
    "turn_id": "turn_audit_0001",
    "fidelity": "exact"
  }
}

That goes to the workspace tool, which needs a workspace_id when your connection can see more than one workspace. Then it calls notification with operation: "brief" and a new agent_goal. The brief can take an empty target, which covers every workspace your connection can see. Add options.date (YYYY-MM-DD) for a past day.

Maven also runs tracking health checks daily, with alerts when a pixel, event or tag stops working. See Maven’s tracking audit.

What a full tracking audit includes

A full audit needs a boundary: which decisions it supports, which systems it covers, and what evidence counts as a finished fix. It ends with decisions, not just findings.

Scope

An audit covers the systems that feed the decisions you care about. Typically that means:

  • The tag manager (GTM), pixels and conversion events.
  • GA4 admin settings.
  • Each ad account’s conversion and linking settings.
  • Server-side tracking, if you have it.
  • The website’s forms and templates.
  • The CRM, store or other backend, with an export of leads and outcomes for one period.

Deliverables

A useful audit produces:

  • Findings ranked by severity and impact, so you fix the gaps that skew decisions first.
  • A prioritized fix list, with a named owner for each item.
  • Baseline metrics, such as unknown traffic share, UTM coverage and the platform-to-backend reconciliation gap, so you can show the fixes worked.
  • A UTM standard and naming taxonomy, with a QA checklist the team can run before each launch.
  • Quick-win fixes where they’re safe to make during the audit.
  • A walkthrough of the findings.

Example scope: a 30/60-day attribution audit

Here’s how one larger engagement was scoped, for a direct-to-consumer brand selling on Shopify and Amazon. Most audits are smaller, but the order of work is the same. The brand started with last-touch attribution only, a high share of unknown Shopify traffic, and no link between Amazon sales and marketing.

An audit moves through inspection, repair, and handover.

Four workstreams

WorkstreamWhat it covers
Foundation auditUTM governance, root causes of unknown Shopify traffic, GA4 configuration, a source of truth per metric, and what Amazon data is available
Channel auditPaid account structure and conversion setup, affiliate tracking and payouts, email attribution, social tagging
Multi-touch evaluationWhat GA4 can deliver, plus Northbeam and Triple Whale compared on cost, effort, accuracy and Amazon coverage
Insight and operating rhythmDecisions on spend and fixes, workflows and owners, and a reporting cadence built around questions

The first two fix what is broken. The last two decide what to do with it, and who owns the decisions. Without owners and a cadence, findings fade.

Timeline

  • Days 1–30, audit and fix: access, the UTM audit, unknown traffic diagnosis, GA4 review and channel audits. The month ends with a prioritized issues list and quick fixes live in production.
  • Days 31–60, build and hand off: the multi-touch decision, one source of truth with reconciliation running, and a readout with budget recommendations. By day 60 the team runs UTM governance, monthly QA and the reporting cadence without the consultant.

Success was measured against the client’s own baseline, with agreed thresholds: less unknown Shopify traffic, higher UTM compliance, a smaller GA4-to-Shopify revenue gap, and a named owner on every recurring process.

Why the tool decision came last

The brand was evaluating Northbeam and Triple Whale, and the scope treated that as a real option, after the audit. Buying a platform on top of unknown traffic and broken UTMs imports the same gaps into a more expensive system. Amazon coverage was likely to decide it, since a platform that can’t see Amazon leaves the marketplace out. For the options, read the multi-touch attribution guide.

Should you audit it yourself or hire someone?

You can run most of this checklist yourself, and the data checks in your AI assistant through Maven’s MCP. The logic of the example above still applies:

  1. Define the business questions first, before choosing any tool or report.
  2. Baseline unknown traffic share and the reconciliation gap against orders or leads.
  3. Name a system of record for each metric.
  4. Decide on any attribution tool after the audit, not before.
  5. Put a named owner on each recurring process, including UTM governance and monthly QA.

Bring in help when:

  • A gap persists after the checklist, and nobody can explain it.
  • The audit needs systems your team doesn’t control, such as the CRM, offline imports or several ad accounts.
  • You run server-side tracking, several domains, or a storefront and a marketplace together.
  • You’re about to buy an attribution platform and want clean inputs first.

Want it done for you? MetricMaven runs full tracking audits and quotes the fixes from the findings.

Related reading: the tracking QA checklist before launch, why Google Ads and GA4 conversions don’t match, and the measurement guide.

Frequently asked questions

Is a GA4 audit the same as a tracking audit?

A GA4 audit is one part of it. GA4 settings such as key events, cross-domain measurement, unwanted referrals, internal traffic filters and data retention all change reported numbers. A tracking audit also covers the tag manager, consent, the ad platforms and the CRM.

How is a tracking audit different from tracking QA before a launch?

QA tests one change before it goes live. An audit reviews the whole running setup, including tags nobody remembers adding.

Can Maven help with a tracking audit?

Yes, for the tag manager. Connect each client’s Google Tag Manager and Maven’s tracking audit finds broken, degraded and stalled pixels and events, tags with no trigger, and GA4 measurement-ID mismatches. Container changes are staged with the exact diff and QA evidence and go live only after you approve them. A full audit also covers the CRM, offline conversion imports, ad account settings and reconciliation across systems.

Can I run a tracking audit in Claude or ChatGPT?

You can run the data checks. Connect GA4, Google Ads and the client’s other sources to their Maven workspace, then ask through the MCP in Claude, ChatGPT, Claude Code or Cursor. It checks your data, not your website, so consent still needs a browser test.

How long does a tracking audit take?

It depends on how many channels, domains and data sources are in play. In the attribution engagement above, the audit and quick fixes took the first 30 days.

How much does a tracking audit cost?

It depends on how many domains, ad accounts and data sources are in scope.

How often should you audit tracking?

After any change to the website, tag manager, consent banner or conversion settings, and on a fixed schedule the team will keep. No ad platform or analytics vendor sets a required interval.

Sources and references

Jamie Isabel

Founder of MetricMaven. Previously managed Google Tag Manager and Google Analytics across 800+ sites at HCA Healthcare, and built attribution and eCommerce reporting at RYZE Superfoods and Mars.

Updated October 23, 2026

Start your 14-day trial

Set up your first client, build their report by asking Claude or ChatGPT, and publish it to their portal.

  • Free for 14 days, no credit card
  • Your first client workspace and portal
  • Every report template, ready to clone
Start a 14-day trial

Nothing is charged unless you subscribe.

Prefer a walkthrough? Book a demo on one of your own clients.