How does GEO differ by AI engine? Four engines, four rulebooks.
A page can be quotable in Perplexity and missing from ChatGPT search, because each engine reads the web with its own crawler, draws on its own index and honors its own opt-outs. Get one robots.txt line or meta tag wrong and you drop out of that one engine while the others keep citing you.
This page takes four of them in turn: ChatGPT search, Perplexity, Google's AI Overviews and AI Mode, and Microsoft Copilot. Every claim about an engine links to the vendor's own documentation. The checklist that applies to all of them is in what technical GEO covers.
- OpenAI and Google keep search and AI training on separate controls, so opting out of training does not take a page out of their answers.
- Google's AI Overviews and AI Mode draw on the regular Search index. Google-Extended doesn't decide whether you appear there; nosnippet, max-snippet and noindex do.
- Copilot runs on Bing's crawl and index, and Bing's NOARCHIVE and NOCACHE tags limit Copilot without removing the page from Bing search.
- OpenAI and Perplexity both say their user-triggered fetchers may not follow robots.txt, so robots.txt alone won't stop a live fetch.
Does GEO work differently on each AI engine?
Yes, at the access and control layer, and hardly at all at the content layer. Each vendor names a different crawler and a different set of directives, so the rule that opts a page out of one engine does nothing to the next. What makes a passage worth quoting is the same everywhere, and that part comes last on this page.
Which crawler and controls does each engine use?
The short version first. Each row is sourced below.
| Engine | Agent that reads for answers | Opt out of answers | Limit what is quoted |
|---|---|---|---|
| ChatGPT search | OAI-SearchBot | Disallow OAI-SearchBot | No directive documented |
| Perplexity | PerplexityBot | Disallow PerplexityBot | No directive documented |
| Google AI Overviews, AI Mode | Googlebot | noindex or nosnippet | max-snippet, data-nosnippet |
| Microsoft Copilot | Bingbot | NOARCHIVE or NOINDEX | NOCACHE |
How does ChatGPT search decide whether to show a page?
Through OAI-SearchBot. OpenAI's bots reference says that agent is for search, and that sites opted out of it will not be shown in ChatGPT search answers, though they can still appear as navigational links. GPTBot is a separate agent that crawls content that may be used to train OpenAI's models. You can disallow GPTBot and still appear in search.
ChatGPT-User is the third agent. It fetches pages for actions a user starts in ChatGPT, OpenAI says robots.txt rules may not apply to it, and it isn't used to decide what appears in search. A robots.txt change takes about 24 hours to reach OpenAI's search systems, by the same page.
That page documents robots.txt as the control and names no snippet-level directive, so we treat ChatGPT search as all or nothing per agent. Our guide to AI training controls in robots.txt shows how we wrote those groups on this site.
How does Perplexity find and cite a page?
It searches the web in real time and puts numbered citations in each answer, each one linking to the original source, according to Perplexity's help center. PerplexityBot is the crawler that surfaces and links sites in those results. Perplexity's crawler guide says it is not used to crawl content for AI foundation models, and recommends allowing it.
Perplexity-User visits a page when someone asks Perplexity a question that needs it. Because a person asked for the fetch, the guide says it generally ignores robots.txt. Perplexity publishes IP ranges for both agents, says changes can take up to 24 hours, and tells firewall admins to use its current ranges. Match on those ranges, not the user-agent string, which anyone can copy.
What decides whether Google AI Overviews and AI Mode cite a page?
The same things that decide classic Search. Google's AI features guide says a page must be indexed and eligible to show with a snippet, and that there are no additional technical requirements. Both features may use a technique Google calls query fan-out, which it says can show a wider and more diverse set of links than a classic results page.
The controls are the snippet directives you already have. Google's robots meta tag reference says nosnippet also stops content from being used as a direct input for AI Overviews and AI Mode, and max-snippet limits how much of it can be. The AI features guide adds data-nosnippet and noindex to the list.
Google-Extended is the wrong lever here. Google's crawler reference says the token governs Gemini training and grounding in Gemini Apps and Vertex AI, has no user agent string of its own, and does not affect inclusion or ranking in Google Search. The AI features guide points to it only for "Google's other systems". If you block it, AI Overviews can still quote you.
Where does Microsoft Copilot get its sources?
From Bing. Bing's Webmaster Guidelines say Bing and Copilot search share the same crawling, indexing and ranking foundation, and list blocking Bingbot in robots.txt as something to avoid. They also recommend IndexNow to tell Bing when a URL changes, which they say reduces outdated URL references in Copilot responses.
Bing has two tags Google ignores. NOARCHIVE keeps content out of Copilot responses and grounding results. NOCACHE limits Copilot to the URL, title and snippet, which the guidelines say reduces citation depth. Bing's announcement of the two tags says pages carrying either one still appear in Bing search, and that a page with both is treated as NOCACHE. Google's robots meta reference says Google Search uses neither. The guidelines add that nosnippet and data-nosnippet may limit Copilot citation quality.
What stays the same on every engine?
The page. We assume none of these fetchers runs JavaScript, and Bing's guidelines warn against hiding critical content behind client-side rendering, so the text ships in the server's HTML. An engine that quotes has to lift a passage, so each section gets a question heading and a direct answer in its first sentence. Bing's guidelines also say structured data must reflect the visible content. Our GEO essay covers that passage work in depth.
The engine-specific part is small by comparison: a robots.txt group per search agent, a decision on each snippet directive, and Bing's two extra tags. In our twenty-week benchmark, crawler access and server-rendered text passed on all eight sites, while 41.0 percent of the implemented work landed in content and internal linking.
Sources, and reusing this page
Claims about an engine link to the vendor's page where they are made: OpenAI's bots reference, Perplexity's crawler guide and help center, Google's AI features guide, robots meta tag reference and common crawlers reference, and Bing's Webmaster Guidelines and its 2023 announcement of the NOCACHE and NOARCHIVE controls. We read all of them on November 11, 2026. Vendors revise these pages without notice, so check the live version before you change a production robots.txt.
You may republish or adapt this comparison with attribution and a link to https://concept211.com/articles/geo-by-ai-engine/. Every heading carries a stable id, so you can link to one engine instead of the whole page.
Want each engine checked on your site?
We fetch your pages as every search agent, set the robots.txt groups and snippet directives to match what you want quoted, and confirm the result with the same fetch. That's our technical SEO and GEO service.
Photo by Rickie-Tom Schünemann on Pexels
All articles