The AI visibility report lands and it says one thing: cited in 5,700 answers. It is a true number. I have one just like it for a publication I run, and on its own it would tell me nothing about where to spend the next month. Split by engine, the same site is a leader on one, a footnote on another and unstable on a third. The split is the report. The total is the cover page.
The numbers
A luxury fashion publication I built and run: 120 pages, live since February 2026. Measured on 28 September, 90 of those pages had been cited inside AI answers, 57 of them in Google’s AI Overviews, across 5,700 answers in total. By engine:
| Engine | Answers citing the site |
|---|---|
| Perplexity | about 2,000 |
| Gemini | 907 |
| Google AI Mode | 870 |
| Copilot | 493 |
| ChatGPT | 268 |
Same pages, same entity, same schema, same everything. Perplexity cites the site seven times more often than ChatGPT. And ChatGPT is the one that moved: down 705 over the period, with no change on the site that explains it. The full case has the rest.
Why the engines differ
Each engine decides on its own what to fetch and what to quote, and the companies say so in their own documentation. Google’s AI Overviews and AI Mode run on the Google index, and Google’s documentation on AI features says there are no additional technical requirements beyond being indexed. Perplexity is built around live search and shows sources on every answer. ChatGPT shows sources only when it decides to search, and its search agent is a separate crawler from its training one, which is why which AI crawlers to allow matters there more than anywhere else.
So a page that is well indexed and well structured gets picked up by Google’s AI features almost by default, gets cited on Perplexity if it answers cleanly, and has to earn its place on ChatGPT. Three different mechanisms, one line in a report.
What a blended number hides
Three things, in the order they cost you money.
Where your buyers are. A company selling to enterprise procurement teams that live in Microsoft 365 should care about the Copilot count and almost nothing else. A consumer brand should care about ChatGPT and Google. The publication above is strongest on Perplexity, which is fine for a publication and would be the wrong engine for a B2B vendor. A total of 5,700 does not say which of these you are.
Where you moved. A drop of 705 on one engine, inside a total that went up, is invisible in the total. It was the most important thing in the data that month, and I still do not know the cause. A blended score would have reported growth.
Where the work goes. The fix for a weak Google AI Overviews count is indexing and page structure. The fix for a weak ChatGPT count is usually robots.txt, entity consistency and third-party mentions. The fix for a weak Perplexity count is the first sentence of each page. Different problems, different hours. A single number gives you no way to choose.
What to do with this
- Pick the two engines your buyers use and treat the rest as context. Ask your last ten customers which assistant they use at work. That is your list.
- Baseline per engine, not in total. The method is in how to measure AI search visibility: a fixed set of buyer questions, asked monthly, results logged per engine. For Google itself there is now a first-party number, covered in Google’s AI performance report.
- Read the drops before the growth. A total that rises while one engine falls is the case where you are about to lose the engine that matters.
- Match the fix to the engine. Do not rewrite fifty pages because ChatGPT went quiet, if the pages are being cited everywhere else. Check the crawler access and the entity first.
Where this sits
Per-engine measurement is the first report I build in AI search optimization work, before any page is touched, because it decides which pages get touched. If your report has one number on it, tell me which engine your buyers use and I will tell you what that number is hiding.