One AI engine mentioned this brand half as often as the others

We ran 46 buyer-intent prompts against four AI engines for one brand. The blended mention rate came out to 28 percent. But that single number averaged together one engine that mentioned the brand 33 percent of the time and another that mentioned it 17 percent. If you track one AI visibility score, you cannot see which engine you are losing.

The audit

We ran a full prompt panel for a brand in a locally-competitive services category. The panel was 46 buyer-intent prompts, each asked of four AI engines that use live web search: ChatGPT (OpenAI Responses API), Claude, Gemini, and Perplexity. That is 184 responses in total.

For each response we recorded whether the brand was mentioned at all. The blended mention rate across all 184 responses was 28 percent. That is the kind of headline number most AI visibility tools report. We then split it by engine.

Mention rate by engine

Three of the four engines clustered between 30 and 33 percent. Perplexity sat at 17 percent, close to half the rate of the other three. The brand was nearly twice as likely to surface in a ChatGPT or Claude answer as in a Perplexity answer for the same set of questions.

Why the engines diverge

The four engines do not read the web the same way. Each one runs its own search step, pulls a different set of pages, and weights them differently before writing an answer. An engine that leans heavily on directory and aggregator pages will surface whichever brands those directories list. An engine that pulls more from forums, news, or the brand's own pages will produce a different cast of names.

So a brand can be well represented in the sources one engine favors and nearly absent from the sources another favors. The result is the spread you see above. It is not noise. Re-running the panel produces the same shape, because it reflects a real difference in where each engine looks.

Mentioned is not the same as recommended

Mention rate measures whether the brand showed up at all. It is a low bar. The stricter measure is whether the engine actually recommended the brand as an answer to the buyer's question, rather than listing it in passing.

For this brand, the blended recommendation rate across all 184 responses was under 1 percent. It was mentioned in 52 responses and recommended in almost none of them. A brand can clear the mention bar and still never be the answer. Both numbers matter, and both need to be read per engine, not blended.

What this means for your strategy

Your buyers are not all using the same AI. Some open ChatGPT, some use Perplexity, some get AI answers inside Google through Gemini. If your visibility is strong on one engine and weak on another, a blended score will read as "average" and tell you nothing actionable. The work is to find the engine where you are weakest and the reason for it.

That reason is usually the source mix. The engine where you score lowest is pulling from pages you are not on. Fixing your own website can lift the engines that read your site, and do almost nothing for an engine that answers from directories and forums. You have to match the fix to the engine.

This is also why a one-time check is not enough. The engines update their search behavior independently. A brand can gain ground on one and lose it on another in the same month, and a single blended number will move very little while the underlying picture changes.

How we measure it

Every CitedWell audit reports mention rate and recommendation rate for each engine separately, not just a blended figure. The same prompt panel runs against all four engines, and the per-engine columns show you exactly where the gap is. From there, the source analysis shows which pages each engine is pulling from, so the fix list is matched to the engine that needs it.