We replaced estimated AI visibility scores with measured data from Microsoft Copilot, Google AI Overviews and ChatGPT. Here is what we built, what it showed us, and how AI transformation delivered it.

From Simulated Scores to Measured Citations: How We Rebuilt AI Visibility Reporting

Every business now wants to know the same thing: when a customer asks ChatGPT, Copilot or Google for a recommendation, do we come up?

For most of the last two years, the honest answer from the AI visibility industry has been "probably". Tools, including our own, estimated visibility by running prompts through AI models and scoring the answers. The scores looked precise. They were not measurements.

This is the story of how we replaced estimated AI visibility with measured AI visibility in Thursday Intelligence, our client reporting platform, and how we used AI transformation to build it.

The problem with prompt tracking

Thursday Intelligence originally included a Prompt Tracker. It asked AI models a set of questions a client's customers might ask, recorded whether the client appeared, and turned the results into an AI Visibility Score out of ten.

The trouble was what sat underneath. The answers were modelled, not observed. Two runs of the same prompt could produce different results. The score could not tell a client how many real people had seen their brand in an AI answer, which pages the AI had used, or what it had searched for. Clients were being shown a precise-looking number built on an estimate.

We decided a modelled score had no place in a client report. So we retired it.

What the AI platforms actually publish

The next question was what could be measured instead. The answer is more than most people realise, though it is spread across three different places:

Microsoft Copilot. Bing Webmaster Tools now includes an AI Performance report. It shows how often Copilot cited a site's pages as a source, which pages were cited, and the grounding queries: the searches Copilot ran behind the scenes before it answered. It also shows the intent and topic of those searches and the site's share of citations. Microsoft samples this data, and there is no API for it, only a manual export.

Google AI Overviews and AI Mode. Google Search Console reports how often a site's pages appeared in Google's generative AI features, broken down by page, country and device. It reports impressions only, with no clicks or positions, and again the report has to be exported by hand.

ChatGPT. OpenAI does not publish citation data for websites. What can be measured is the result: visits arriving from ChatGPT, which Google Analytics 4 records as referral traffic from chatgpt.com.

None of these is a rank. There is no such thing as a published position in ChatGPT or Copilot. But each is a real, first-party measurement from the platform itself, and together they answer the question clients actually care about.

What we built

We rebuilt the AI section of Thursday Intelligence around those three sources.

One import, one click. Each month, the Bing and Search Console exports are dropped into a single import screen in our CRM. It recognises each file from its column headers, fills in the reporting period where the file records it, and shows a summary before anything is saved. One click then saves the data, pulls the latest ChatGPT visits directly from the client's GA4 property, and publishes the report to the client's portal.

A tab for each engine. Clients now see separate Copilot, Google AI Overviews and ChatGPT tabs, each with a daily trend, the pages that engine uses most and, for Copilot, the grounding queries behind the citations. Each tab only appears once there is real data behind it.

A summary that compares engines. The AI Visibility tab brings the three together: monthly trend charts for each engine, headline figures, a combined view of the pages each engine relies on, and the top searches Copilot ran before citing the client.

Honest numbers. Every figure is labelled with its source. Citations, impressions and visits are never added together, because they measure different things. Where data is missing, the portal says "Not available" rather than showing a zero. Where a platform samples its data, the portal says so.

What it showed us straight away

The first client loaded into the new reporting was a travel and experiences business. Within minutes the data showed something a single visibility score never could: different AI engines rely on different content.

Microsoft Copilot leaned most heavily on the client's practical guides, such as entry requirements and activity guides. Google's AI Overviews leaned on a different set entirely: the client's seasonal weather guides. And the grounding queries showed, in the customers' own words, exactly which questions were leading Copilot to the site.

That changes the advice we give. Instead of "improve your AI visibility score", the recommendation becomes specific: which pages to strengthen for which engine, and which real questions to answer next.

How AI transformation built it

We practise what we recommend. This project was delivered as an AI transformation of our own reporting operation, with an AI engineer working inside our CRM and client portal under clear rules:

• Changes that clients would see were built and checked in preview, then approved by a person before going live.
• Every change was verified against the live database and the live portal, not taken on trust from the build tools.
• Nothing was published with invented figures, and anything that could not be measured was labelled as such.

The result is a monthly reporting routine that takes minutes per client, and a report that answers the questions clients ask with evidence rather than estimates.

What this means for our clients

If you work with us on generative engine optimisation, your reporting now tells you:

• whether Copilot, Google's AI features and ChatGPT are using your website, measured by the platforms themselves
• which of your pages each engine relies on
• what AI is searching for when it finds you
• how all of that is changing month by month

And our recommendations start from that evidence.

See where you stand

Want to know how AI engines see your business today? Request a free GEO audit, or read more about our generative engine optimisation service and AI transformation.