For GEO, the engines worth tracking are ChatGPT, Perplexity, Gemini, Claude and Grok, and the reason to include each one is not how popular it is. It is that each gets your content in a different way and cites sources in a different way, so the same page can be visible in one and missing in another. There is no reliable public usage share to rank them by, so this answer does not invent one; it compares what you can actually check and control.
Why "most popular" is the wrong question
A popularity ranking would tell you where people ask, not whether your pages can reach the answer. What decides that is mechanical: which crawler fetches content for the engine, whether you can allow or block it by name, and whether the engine shows its sources. Those facts are documented by the vendors themselves, and they differ enough that a single "AI search" setting does not exist. The overview on AI engines collects them per engine.
The engines, and how each one gets your content
| Engine | Named agents | What to know |
|---|---|---|
| ChatGPT (OpenAI) | GPTBot, OAI-SearchBot, ChatGPT-User | Three jobs: GPTBot gathers training content, OAI-SearchBot builds the index behind ChatGPT search, ChatGPT-User fetches a page because someone just asked. Blocking OAI-SearchBot removes you from ChatGPT search results; blocking GPTBot does not. |
| Perplexity | PerplexityBot, Perplexity-User | Indexing is separate from user-triggered fetches. Perplexity shows sources on nearly every answer, so an access problem here shows up in citation data quickly. See the Perplexity page. |
| Gemini (Google) | Google-Extended, Googlebot | Google-Extended is a usage control, not a crawler; the fetch is Googlebot. Blocking Google-Extended keeps you in Google Search but removes you from grounded Gemini answers. |
| Claude (Anthropic) | ClaudeBot, Claude-SearchBot, Claude-User | ClaudeBot collects training content, Claude-SearchBot feeds the search results Claude users see, Claude-User fetches on a user's request. Each can be allowed or blocked deliberately. |
| Grok (xAI) | None documented | xAI does not publish a crawler user agent, so you cannot allow Grok by name. What you control is whether your pages are readable to general crawling: real HTML, and no bot rule returning 403 to unknown agents. |
Two more surfaces are worth knowing about even if they are not in your first set. Apple documents Applebot for search across Spotlight, Siri and Safari, and Meta documents separate agents for Meta AI search and for user-requested fetches. Microsoft Copilot is the gap: no Copilot-specific crawler could be verified in Microsoft's public webmaster documentation, which is different from saying none exists.
How to choose which engines to measure
- Start with the engines your buyers use, which you can only learn from your own audience, not from a public ranking.
- Check access per engine before measuring anything. If OAI-SearchBot or PerplexityBot is blocked, a low score is an access problem, not a content problem. The breakdown in AI crawler user agents shows what each block costs.
- Include at least one engine that shows its sources, such as Perplexity, because citations tell you which pages the answer was built from.
- Report each engine separately. Engines name different numbers of brands per answer and cite differently, so an average across them hides where the work is.
Common mistakes
- Blocking every AI agent with one robots.txt rule to opt out of training, and removing yourself from search-style answers in the process.
- Assuming blocking Google-Extended affects Google Search ranking. Google's documentation says it does not.
- Writing a guessed Grok user agent into robots.txt and believing access is now granted.
Vendor references: OpenAI crawlers, Perplexity bots, Google common crawlers.