Engine by engine

ChatGPT vs Perplexity vs Gemini: who cites your brand?

They don't answer alike, and they don't cite alike. Some search the web and name their sources; some answer from memory and cite nothing. The practical consequence is blunt: there is no single "AI optimization" — the brand that an engine names depends on how that engine builds its answer. Here is the head-to-head, measured.

The short version: two camps, not one "AI"

Across our measured answers, engines fall into two behaviours, and the behaviour decides everything about how you get named:

The comparison, measured

How often each engine's answers rested on a web source (a directory or page) in one measured niche, and how often it named a company at all. The first column is the tell: a high number means a web-grounded engine you can influence through sources; a low one means an engine answering from memory.

EngineCampLeaned on a web sourceNamed a company
PerplexityRetrieval53.1%high
Yandex (search answer)Retrieval86.5%91.0%
Gemini / AI OverviewsRetrieval*web-heavy (not measured)not measured
ChatGPT (from memory)Memory~0%varies
GigaChatMemory4.6%72.4%
DeepSeekMemory-leaninglow74.6%

* Gemini / AI Overviews: we do not measure it directly and publish no numbers for it — it has no measurement API we use. It is placed by documented behaviour (a retrieval system that leans on web sources, Reddit and forums prominent among them), not by invented figures. The measured shares are from buyer questions in the Russian-speaking market; the mechanism — retrieval engines cite the web, memory engines cite little — holds across markets. Open data by niche: /barometr.json.

The part that surprises people: the sources barely overlap

Because the engines build answers so differently, a source that feeds one rarely feeds another. Perplexity may cite a review aggregator; Yandex a map and a city directory; ChatGPT nothing at all but whatever it absorbed in training. So a brand can be named reliably by one engine and missing from the next — not because it's weaker, but because the engines read different things. Chasing a single "get cited by AI" checklist optimizes for an average that no real engine uses.

So what do you actually do — per engine

The honest through-line: we don't sell a position on any of them, and no one can promise one. We measure which engines name you, which name a competitor instead, and what each built its answer from — so the work goes where your buyers actually ask.

See the split on your own brand

Free, no sign-up: your buyers' questions across ChatGPT, Perplexity, DeepSeek, GigaChat and Yandex, with each engine's answer, sources, and whether you were named.

Run a free check

Common questions

Do ChatGPT, Perplexity and Gemini cite the same brands?
No — often not even the same kind of source. Retrieval engines (Perplexity, Yandex, AI Overviews) search and cite listings; memory engines (ChatGPT from memory, GigaChat) answer from training and cite almost nothing. A source that feeds one reaches few of the others, so the named brand changes by engine.
Which AI engine should I optimize for?
The one your buyers use — the work differs. For retrieval engines, presence in the sources they search. For memory engines, being written about widely and long enough to be in training. No single "AI optimization" covers both.
Do you measure Gemini and Google AI Overviews?
Not directly — no measurement API we use, so we publish no numbers for it and won't pretend to. We place it by documented behaviour (retrieval, web-heavy, Reddit-prominent) and measure ChatGPT, Perplexity, DeepSeek, GigaChat and Yandex ourselves.