A GA4 audit for paid media checks whether the property can be trusted to judge campaigns: one clean stream per site, a short list of real key events, unique purchases, no test or internal traffic, UTMs that land in the right channels, a working Google Ads link and 14-month retention. Run it before your first report on an inherited property, and tell the client what is wrong with past data before they find it.
- Audit in the order that breaks paid reporting: purchases, key events, channels, then settings.
- Compare GA4 transactions with the backend for two weeks before you quote any ROAS.
- Most fixes only apply going forward; say so in the audit, not after.
- Test filters in Testing state first. An active data filter permanently drops data.
Say you’re handed a client’s GA4 property on Thursday and the first performance report is due Monday. You don’t have time for a 200-point audit. You need to know which numbers you can put in front of the client and which come with a warning. This is the pass we run, in the order that matters for paid media.
1. Property basics
- Time zone and currency should match the ad accounts you report beside. A time zone change only affects new data and can leave a flat spot or spike on the day; a currency change also restates history.
- Data retention at 14 months (the standard-property maximum). It only affects explorations and funnel reports, but a two-month setting quietly breaks year-on-year analysis.
- Reporting identity and Google signals. Note what’s set; it changes user counts and thresholds, not what gets collected.
- Staging properties kept separate and named as such, so nobody reports from the wrong one.
2. Streams and tags
- One web stream per site, and the same measurement ID on every template, including checkout and the thank-you page.
- If the journey crosses domains (a booking engine, a payment page), cross-domain measurement is set and the payment provider is in "List unwanted referrals". Otherwise the sale is credited to the payment site.
- Enhanced measurement reviewed: form interactions can create noisy form_submit events on sites with many forms.
- The event list has no junk names, such as page titles firing as event names, or three versions of the same login event.
3. Key events and purchases
- Fewer than the 30-event limit, and only real outcomes. Legacy key events such as page_view or click inflate every report.
- With Ads personalization on and a Google Ads link, GA4 also marks add_to_cart, add_to_wishlist, begin_checkout, session_start, view_item, view_item_list and view_search_results as key events. Know which ones are there before anyone reads the "Key events" total.
- Counting method: "Once per event" is recommended. Migrated properties often still use "Once per session (Legacy)".
- Only one purchase-type event is a key event. Two events that fire on the same confirmation page double every conversion.
- Every purchase carries a unique, non-empty transaction_id. Check for repeats (duplicate purchases).
4. Test, internal and bot traffic
Run a free-form exploration with Hostname, Event name and Event count for the last 90 days. Staging, QA and preview hostnames sending purchases into production are more common than you’d think, and those sessions usually land in Direct or Unassigned, which inflates "organic" sales. While you’re there, look for a browser or country with an absurd events-per-user ratio, which usually means bots or a QA tool. The fixes are in filtering test and internal traffic.
5. UTMs and channels
- Share of sessions in Unassigned and Paid Other, by week. A step change usually marks the day someone changed a URL template.
- Source fragmentation: tiktok and tiktok.com, or facebook and Facebook, as separate rows.
- Mediums that are platform names (tiktok, snapchat, meta), which push paid traffic into Organic Social or Paid Other (the UTM fix).
- One utm_campaign reused across several platforms, which makes campaign rows meaningless.
- Whether a custom channel group already exists, and who built it.
6. Links, consent and geography
- Google Ads link present, auto-tagging on in Google Ads, and you know which GA4 key events are imported and whether they are primary next to the Google Ads tag.
- BigQuery link if the client wants raw data. Standard properties cap the daily export at 1 million events; streaming has no cap.
- Consent mode status, checked in the tag setup rather than assumed.
- If the client asks for sales by city, warn them that GA4’s city is IP-based and mobile carriers can route traffic through hub cities.
GA4 audit checklist for paid media
- Time zone and currency match the ad accounts
- Data retention set to 14 months
- Staging properties separated and labelled
- One web stream per site; same ID on checkout and thank-you pages
- Cross-domain set; payment providers in unwanted referrals
- No junk or duplicate event names
- Key events: real outcomes only, under 30
- Auto-marked Ads key events known
- Counting method: once per event
- One purchase key event, unique transaction_id
- GA4 transactions vs backend orders, 14 days
- Hostname check: no test or preview hosts
- Internal traffic rules and data filters reviewed
- No bot-like browser or country spikes
- Unassigned + Paid Other share trended weekly
- No platform names used as utm_medium
- Google Ads link, auto-tagging and imports checked
- Consent mode verified in the tag setup
Score it: what to tell the client
Severity of common findings
| Severity | What to say | |
|---|---|---|
| Duplicate or test purchases | Blocks ROAS | Revenue in GA4 is overstated for these dates; we report backend revenue until it’s fixed. |
| Paid social in Unassigned or Paid Other | Blocks channel reporting | Paid social is under-counted; we’ve fixed the tags and re-labelled history with a custom channel group. |
| Junk or auto-marked key events | Fix this week | We report one named key event, not the total. |
| Missing cross-domain or referral exclusions | Fix this week | Some sales are credited to the payment site; fixed from [date]. |
| Two-month retention | Fix today | Explorations older than two months are gone; retention now 14 months. |
| City-level reporting | Caveat | Use the platforms’ location reports for geo decisions. |
Where this goes wrong
- Fixing before you take a baseline. Export the key numbers first, or you can’t show the client what changed.
- Promising to clean history. Data filters aren’t retroactive, and data deletion requests only blank text parameters; the events still count.
- Activating filters untested. An active data filter permanently drops data. Leave it in Testing for a few days.
- Auditing GA4 alone. Half of these problems live in the tag container or the ad URLs. Run the GTM audit alongside.
Do it faster with Whathead
Most of this audit is reading, which is where an assistant saves the most time. With the Whathead GA4 MCP, Claude or ChatGPT can list key events and their counting method, data streams, custom dimensions and Google Ads links, and run the hostname, channel and transaction checks as reports. Settings such as data retention and data filters aren’t covered, so check those in the GA4 interface.
- “For GA4 property [name], list key events with their counting method, the data streams, and the linked Google Ads accounts.”
- “Show purchases and revenue by hostname for the last 90 days, then list transaction IDs that appear more than once.”
Nothing changes unless you ask for a change, and Admin changes are shown as a preview first. The same connection covers the ad side; see the Meta Ads audit with Claude. Whathead is our product.
GA4 audit: common questions
What should a GA4 audit include?
Property settings, data streams, key events, purchase data, internal and test traffic, UTMs and channel groups, product links such as Google Ads and BigQuery, consent and data retention.
How long does a GA4 audit take?
A focused paid-media pass like this fits in an afternoon. Fixing what it finds takes longer, because tag changes need testing and some fixes wait on developers.
How do I check if GA4 is tracking purchases correctly?
Compare daily GA4 transactions and revenue with backend orders for at least two weeks, then look for repeated transaction IDs and purchases from non-production hostnames.
Can a GA4 audit fix historical data?
Mostly no. Custom channel groups re-label past channels, but filters and tag fixes only apply going forward. Document the affected dates.
How often should I audit GA4?
Fully when you inherit a property or before a big season, and a quick check monthly or after any site release.
Sources
- GA4 Help (Checked 19 Nov 2026): data retention, time zone, currency, reporting identity, enhanced measurement, cross-domain, unwanted referrals, property limits, key events, counting method, transaction IDs, data filters, data deletion requests, Google Ads links
Written by Ahmed Shahin, founder of Whathead. Last reviewed 19 November 2026.
Run the read-only GA4 audit from Claude
Key events, streams, Ads links, hostnames and duplicate transactions, checked in one conversation.



