Five engines, one question, different answers: how much do AI engines agree?

Ask ChatGPT, Gemini, Claude, Perplexity and Google's AI Overviews the same Turkish question and the five engines usually name different brands: in the questions we measured, 70.7% of the brands they named were named by only one engine. GeoMono is a platform that measures how visible brands are in AI answers. This post uses the questions of 15 brands to show how much the engines agree, what the same engine says to the same question a second time, and what that means for a brand, with the figures.

The short answer: the engines agree little, and the same engine does not fully agree with itself either. That is not a flaw of measurement but the nature of what is measured. A brand's place in AI is told not by a single screenshot but by a measurement repeated in every engine.

What did we measure?

Between 16 August and 31 October 2026, for 15 brands, we asked five engines 3,438 Turkish questions written the way a buyer would ask, without the brand's name. A classifier extracted the brands and companies named in each answer; different spellings of the same brand counted as one brand, and the engines themselves (ChatGPT, Google, Perplexity and the like) did not count as brands. If the measured brand itself was mentioned, it entered the list too.

Of 17,190 planned answers, 501 were not counted: 317 calls returned an error, 24 never ran, and the brand list of 30 answers could not be extracted. For 130 questions Google showed no AI Overview. None of these was counted as "named no brand"; that engine had no readable answer for that question. Of the 3,238 searches we could measure, Google showed an AI Overview for 3,108 (96.0%): for buyer questions of this kind, an AI Overview was nearly always on the page.

How often do the engines name a brand at all?

The first difference is how willing each engine is to name brands. Some of our questions were informational ("what should I look out for"), so in every engine a large share of answers named no brand at all:

ChatGPT was the most cautious engine: more than half of its answers named no brand. AI Overviews had the fewest answers naming no brand. That alone says something: the visibility rates measured for the same brand in ChatGPT and in AI Overviews cannot be compared directly, because the engines' habits of naming brands differ.

Do the five engines name the same brands?

To measure agreement fairly, we looked only at questions where all five engines were read and all five named at least one brand. 906 of 2,960 questions met that condition; they were mostly "which one should I buy" or "which firm should I work with" questions.

So for a typical buyer question the five engines name twenty-two different brands between them, and most of those appear in only one engine. In two questions out of three, there is not a single brand all five engines agree on.

Leaving AI Overviews out and looking at the four chat engines does not change the picture: in the 1,010 questions where all four named a brand, 74.2% of brand mentions were in only one engine, and 374 of the questions (37.0%) had a brand all four named.

An example

We picked a "which would you recommend" question in a consumer product category; the number of different brands the five engines named for it equalled the median of the 906 questions. We do not say which question or which brands; only the counts:

For this question the five engines named 22 different brands in total. Of these, 16 were named by only one engine, 3 by two, 2 by three and 1 by four. No brand appeared in all five. For this question, "which brand does AI recommend?" has no single answer; it has five, one per engine.

Which engines are closest to each other?

To compare two engines we measured this: of all the brands two engines named for the same question, the share both named. We averaged it over the questions where both engines named at least one brand:

The closest pair is Gemini and AI Overviews, both of which rest on Google's search infrastructure. Google's AI features documentation says its AI answers rest on pages in Google Search's index; two surfaces fed from a common source share more brands. Even so, in the closest pair three quarters of the brands the two named were not shared. ChatGPT overlapped least with every other engine.

Next step

Read how we measure

The method page explains which answers are counted, what happens when an engine does not answer, and when we do not publish a figure.

Read the method

What does the same engine say to the same question a second time?

Part of the difference between engines comes from variation within each engine. To see it, we looked at the 5 brands whose questions were measured more than once with the same question text, and compared each engine's answers in the first and second measurement. There were 1,380 pairs in which both answers named at least one brand.

Only 63 of the 1,380 pairs (4.6%) gave an identical brand list. Perplexity was the most consistent with itself, ChatGPT the most variable. This fits other research: in SparkToro's study published in January 2026, the same brand list rarely came back for the same question, and the April 2026 paper "Don't Measure Once" gives the warning in its title. Our data shows it holds for Turkish questions as well.

By sector: consumer and business questions

For questions where all four chat engines named a brand, we split two groups:

In business questions the engines agree even less. Nearly half of consumer questions had a brand all four engines shared; for business questions the figure is just over one in five. For a business brand the meaning is clear: being visible in one engine does not mean being visible in the others.

What should a brand take from these figures?

1. Measure each engine separately. Most brands the engines name appear in only one engine. A measurement in one engine says very little about the others.
2. Repeat the questions. When the same engine is asked the same question a second time, the two answers' brand lists overlap by only about a third on average. One answer is a moment, not a trend.
3. Read frequency, not rank. "We are third in ChatGPT" may not hold for the next answer. A pair such as "mentioned in 12 of 40 questions that do not name us" is sturdier (example numbers).
4. Do not merge the engines into one number. The average of engines whose brand-naming habits differ this much describes none of them.

How we apply these rules in our own measurement is on our method page. How the engines look at the web is in why the same Turkish question brings different brands, and why measurement tools give different figures is in why AI visibility tools give different results.

Frequently asked questions

Do AI engines recommend the same brands for the same question?

Mostly not. In the 906 questions where all five engines named a brand, 70.7% of the brands named were named by only one engine, and only 32.3% of the questions had at least one brand all five named together. For a typical buyer question, the five engines name twenty-two different brands between them, so saying "AI recommends us" after looking at one engine is misleading.

Which two engines are most alike?

Gemini and Google AI Overviews. In the 1,427 questions where both named a brand, 25.4% of the brands they named were shared on average. ChatGPT overlapped least with every engine (between 14.8% and 15.8%). Both rest on Google's search infrastructure, so they draw on shared sources. Even in this closest pair, three quarters of the brands they named were not shared.

If I ask the same engine the same question again, will I get the same answer?

Most likely not. When the same question for the same brand was asked again in the same engine in a later measurement, the two answers' brand lists overlapped by 36.1% on average; only 4.6% of 1,380 pairs gave an identical list. Perplexity was the most consistent with itself (50.4%), ChatGPT the most variable (22.3%). So visibility should be read from repeated measurement, not from a single answer.

Do these figures hold for my sector too?

These figures come from the questions of 15 brands and are not a market sample. To see how much the engines agree in your sector, you need a measurement with your own questions, separate for each engine and repeated. Splitting consumer and business brands showed a clear gap: in business questions, the share of questions where all four chat engines named a common brand (22.6%) was half that of consumer questions (45.4%).

Why might I look good in one engine and be absent in another?

The engines look at different sources and have different habits of naming brands: ChatGPT named no brand in 59.1% of its answers, AI Overviews in 41.7%. Being on the pages one engine reads does not put you on the pages another reads. So the sources each engine reads have to be measured separately, and content work planned engine by engine.