Do ChatGPT, Google AI Mode and Perplexity recommend the same companies?
Unity Horizon4 min read
Rarely. Asked eight provider questions three times each, they named 164 businesses, and 85% of the names came from one engine only. Checked 2 November 2026.
Rarely. We asked ChatGPT, Google AI Mode and Perplexity the same eight questions about hiring a provider, three times each, on 2 November 2026. The 72 answers named 164 different businesses. 85% of the names a question produced came from one engine only, and in 15 of the 24 question runs not a single business was named by all three.
These are our own runs, and the method is at the end. We sell AI visibility work in Greece and these are questions our own buyers ask, so weigh the numbers with that in mind.
Do the three engines name the same businesses?
Almost never the same set. For each question we took the businesses two engines named and asked what share both of them named.
| Engines compared | Named by both, of all named by either | Runs with no overlap at all |
|---|---|---|
| ChatGPT and Google AI Mode | 11.8% | 6 of 24 |
| ChatGPT and Perplexity | 11.6% | 4 of 24 |
| Google AI Mode and Perplexity | 8.4% | 12 of 24 |
Add the three engines together and each question run surfaced about 16 businesses. At least two engines agreed on 58 of the 393 names counted that way, 14.8%. Someone who asks only ChatGPT and someone who asks only Perplexity get two different shortlists for the same need, and neither of them knows it.
Does the same engine give the same answer twice?
Not within the hour. Same question, same engine, three runs between 01:30 and 02:10 UTC.
| Engine | Businesses named per answer | Named in all three runs | Named in only one run |
|---|---|---|---|
| ChatGPT | 9.5 | 29% | 45% |
| Perplexity | 5.6 | 15% | 68% |
| Google AI Mode | 4.1 | 5% | 76% |
The percentages are of the distinct businesses an engine named for a question across its three runs, summed over the eight questions: 124 for ChatGPT, 92 for Perplexity, 76 for Google AI Mode.
ChatGPT is both the steadiest and the most generous. Google AI Mode names four businesses and rarely the same four: of 76 it named, four came back in every run. A business that appears once in AI Mode has learned close to nothing about its position.
Why do they disagree?
Because they read different pages. Google AI Mode showed 3.1 source domains per answer on average, ChatGPT 12.2 and Perplexity 16.8. Across the 24 question runs the engines showed 685 domains between them, and 73 of those, 10.7%, were shown by more than one engine.
What they read differs in kind as well as count. Directory listings were 15% of the domains Perplexity showed and 5% of those ChatGPT and Google AI Mode showed. In the two questions asked in Greek, sites on .gr domains made up 48% of the sources, against 31% for the six asked in English. We cannot see inside the engines, but an answer built from three pages and one built from seventeen were always going to disagree about who is worth naming, and the language of the question changes which pages are in the pool.
Is anyone named consistently?
Yes, and the gap between first and fifth is wide. Across the four questions about GEO providers in Greece, 36 answers in all, the most-named provider appeared in 28. The fifth appeared in 8. Sixty-one businesses were named at least once.
Our reading of who this arrangement suits: providers with a page in the buyer's language that states what they do and where, and profiles on the directories the engines read. Who it costs: a provider with one English service page and nothing else. On 2 November that described us, and we were named in none of the 36.
How was this measured?
- Eight questions a buyer asks when looking for a GEO or AI provider in Greece, six in English and two in Greek. Two questions about our own name were also asked and are not counted here.
- ChatGPT and Google AI Mode through SearchApi.io, which returns what the logged-out consumer pages show, with AI Mode set to Greece. Perplexity through its API, Sonar with web search, located in Athens.
- Businesses were extracted from each answer by a language model (Gemini) at temperature 0, and merged by hand where one business appeared under two names. Sources were compared by domain, not by URL.
- One market, one day, one family of questions, 72 answers. Every share above will move with a different week or sector.
How should you measure AI visibility, then?
Count the share of answers that name you, per engine, over repeated runs of a fixed question set. A position in one answer is a sample of one from a list that changes within the hour, so ask each question at least three times. Our own tracker asks ten fixed questions of all three engines, and we rerun it rather than trust one result.
Track the sources too. If an engine shows three domains and none is yours or a directory that lists you, rewording your homepage will not put you in that answer. What makes a page quotable once it is in the pool is in What is Generative Engine Optimization (GEO) and how is it different from SEO?. The free AI visibility check asks the same three engines the questions your customers ask and counts how many answers name you.
How many of the three engines named your business the last time someone asked, and has anyone in your company asked twice?
Common questions
Do ChatGPT, Google AI Mode and Perplexity recommend the same businesses?
Rarely. In 72 answers to eight provider questions, asked three times each on 2 November 2026, 85% of the businesses a question surfaced were named by one engine only, and in 15 of 24 runs no business was named by all three.
How stable is an AI assistant's list of recommended businesses?
Not very. Asked the same question three times within 40 minutes, Google AI Mode named 76% of its businesses in only one run and kept 5% in all three. ChatGPT was the steadiest and still named 45% only once.
How many sources does each AI search engine show per answer?
In our runs ChatGPT showed 12.2 source domains per answer on average, Perplexity 16.8 and Google AI Mode 3.1. Of 685 domains shown across 24 question runs, 10.7% appeared in more than one engine.
How should a business measure its AI visibility?
As the share of answers that name it, across several engines and repeated runs of the same questions. One answer from one engine is a sample of one from a list that changes within the hour.
Related service
AI visibility
Turn AI answers into customers. From invisible to recommended.