The assistant cannot see what never arrived.
On one national retailer's site, the cookie notice has to be closed before the site can be used, and closing it leaves every consent category on. The consent field in the history is therefore a constant. The only control that turns a category off sits three pages away, below a newsletter form. On the same site, 279 of 280 tag rules carry no consent gate.
An AI reading that history answers with confidence, because from inside the data everything looks complete. It has no way to know. That is what a web analytics audit is for: finding the gap between what happened on the site and what the data says happened, before the AI turns the gap into an answer.
What a digital analytics audit finds before the AI does
Three kinds of fault reach every AI that reads analytics data. Each one is measurable on the site.
- Collection faults. A tag fires twice, or never, or inside a frame the console cannot see. On the retailer's checkout, the whole Adobe stack ran inside a child frame; a reading taken in the top window saw no Adobe at all. A checkout progress event fired six times on one pass and begin checkout fired twice.
- Taxonomy faults. The same thing carries different names or different identifiers on different pages. The AI treats them as different things, or merges things that are not the same.
- Consent faults. The consent field says one thing and the network says another. With Targeting refused on the retailer's site, the gated marketing tags stayed silent, and 30 calls to a retail-media network and a Google cookie sync went out anyway. Analytics requests marked as denied still reached Google as cookieless pings.
A general data audit looks at tables. An analytics audit looks at the site, then at the tables, then at the distance between them.
Measured on the site. Graded by analysts. Verified by code.
- Measure reality.Real sessions on real devices, per consent state. The tag container parsed rule by rule. Every beacon decoded. This is what actually reached your data.
- Ask what your analysts ask.Forty to one hundred business questions, each with the correct answer or the honest "not answerable from this data", and the evidence behind it.
- Grade and trace.Six failure classes, from "correct and supported" to "answered when it should have refused". Every wrong answer traced to the tracking, taxonomy or consent cause, with the fix.
Stone, the guideline. Wood, the graded row. Glass, the evidence. Bronze, the delivered set.
Checked, not claimed.
Five live QA runs on one retailer's consent stack, on five pages, in one evening. One published finding corrected on the record the same night, when live evidence contradicted it. Two of our own console readings retracted the same evening once the frame scan ran; recorded so the method is not repeated.
We read the vendor's own material before we grade the vendor's AI: Amplitude's documentation and its partner training, transcribed; the Adobe partner hub and its recordings, transcribed; the Shopify Help Center and developer documentation.
When a consent panel turned out to exist three pages deep on a client's site, we corrected our own finding the same night and measured what the refusal actually stopped: the gated marketing tags obeyed; the ad-delivery stack and the analytics beacons did not. That is the standard your answer audit is graded to.
Four audits, one method
Website tracking audit
What fires, per device and per consent state, against the tracking plan.
The tracking auditData quality audit
The collected table checked against the site: gaps, duplicates, mislabels.
The data quality auditIf you need an audit of one platform by name, those live on our platform sites: Google Analytics audit at ga.webclat.com, Google Tag Manager audit at gtm.webclat.com, Adobe Analytics audit at adb.webclat.com. The analytics implementation itself, the fix after the audit, is engineering work we quote per scope.
Start with forty questions.
One AI. One data source. Forty business questions, graded, with root causes and fixes. Three weeks. Fixed price, quoted per scope. You keep the set.
- Answer audit, one-off.
- Re-audit, after each container release or AI update.
- Real-session collection, what actually fires on your site, per device and consent state.