ChatGPT, Perplexity, Gemini, AI Overviews compared for businesses
Crawlers to allow, links to sources, trackable clicks: what each assistant officially documents, and how to measure each one separately.
By Sébastien Monnier · · 6 min read · Lire en français
ChatGPT, Perplexity, Gemini and Google AI Overviews all answer your customers’ questions, but not in the same way. This comparison sticks to what their publishers officially document, and to what can be measured. Where there is no public documentation, we say so.
The comparison
| Criterion | ChatGPT (search) | Perplexity | Google AI Overviews | Gemini (app) |
|---|---|---|---|---|
| Crawler to allow in order to be cited | OAI-SearchBot | PerplexityBot | Googlebot (normal indexing) | No separate rule published |
| What the publisher says | A site that blocks OAI-SearchBot is not shown in search answers | PerplexityBot is used to surface and link websites in results | The page must be indexed and eligible to show with a snippet; no special markup | — |
| Links to sources | Yes, in answers that use search | Yes, numbered sources | Yes, links in the overview | Varies with the answer |
| Clicks identifiable on your site | Often utm_source=chatgpt.com | Referrer perplexity.ai | Counted with Google Search (Search Console, “Web” type) | Referrer gemini.google.com when it is passed on |
| How Recoia measures it | Real answers via DataForSEO | Perplexity’s official API | Google results page via DataForSEO | Real answers via DataForSEO |
Table compiled on 26 September 2026 from the documentation of OpenAI, Perplexity and Google. Behaviour changes over time: check the date.
What they all have in common
- They need to read your site: if the relevant crawler is blocked in your
robots.txt, nothing can be cited. - Answer precisely: one page per service, with the facts at the top (prices, lead times, area).
- Be present elsewhere: assistants cross-check your site with directories, reviews and media.
Why results differ from one assistant to another
Each assistant has its own sources, its own search engine and its own way of writing. A business can be recommended by Perplexity and missing from ChatGPT for the same question. That is why a serious measurement follows these practices:
| Good practice | Why |
|---|---|
| Measure each assistant separately | Gaps between assistants show where to act first |
| Ask each question several times | A single answer is a lottery draw, not a measurement |
| Keep the raw answers | To check what was said, word for word |
| Date every measurement | Assistants change: you compare weeks with each other |
Where to start
- Check your
robots.txt(the guide). - Measure your key questions on all four assistants.
- Work first on the assistant where the gap with your competitors is widest.
Updated 26 September 2026.