Sources checked 8 November 2026.
On 8 November 2026, only one of the five companies behind the main AI answer engines had published what makes a page more likely to be chosen as a source. Microsoft has. Google says its AI features run on normal Google Search and need nothing extra. OpenAI, Anthropic and Perplexity document how to keep your site out, and say nothing about how they pick what goes in.
That’s my reading of their documents on one date (original analysis, not an industry finding). Everything below is what each company says, linked to the page I read, with a label for how solid it is.
Why this question matters
Most advice on “getting cited by ChatGPT” is written as if the engines have published a rulebook. They haven’t. If you’re deciding where to spend time, the useful split is between what a company has committed to in writing and what someone has inferred from watching its outputs. The first changes when the company changes its page. The second can be wrong on day one.
The five engines, side by side
| Engine | Where it gets results | What blocking controls | What it says about choosing sources | What it reports to site owners |
|---|---|---|---|---|
| ChatGPT search | Third-party search providers plus content from partners (OpenAI, Oct 2024) | OAI-SearchBot for search, GPTBot for training, ChatGPT-User for user requests (OpenAI bots) | Any public site can appear. Nothing on ranking (OpenAI FAQ) | utm_source=chatgpt.com on referral links |
| Claude | A search tool that grounds answers in the live web. Brave Search is reported, not stated in Anthropic’s help pages (Claude help; Simon Willison) | ClaudeBot for training, Claude-SearchBot for search, Claude-User for user requests (Anthropic) | Nothing in the pages I read | Nothing I found |
| Google AI Overviews and AI Mode | Google’s own index. Pages must be indexed and eligible for a snippet (Google) | Googlebot for Search; snippet and noindex controls; Google-Extended for some other Google AI systems | No extra requirements, no special schema, no AI text files (Google guide) | Counted in Search Console’s Performance report |
| Perplexity | Searches the web in real time (Perplexity help) | PerplexityBot for search, not training. Perplexity-User generally ignores robots.txt (Perplexity) | Calls its sources “authoritative”. No criteria | Nothing I found |
| Microsoft Copilot | Bing’s index (Bing guidelines) | NOINDEX, NOARCHIVE and NOCACHE each limit Copilot use differently | Single-topic pages, facts stated outright, key information near the top, consistent names for people and organisations | AI Performance report: citations, cited pages, sample grounding queries (Bing, Feb 2026) |
What each company commits to
OpenAI says ChatGPT search “leverages third-party search providers, as well as content provided directly by our partners”. It doesn’t name the providers on the pages I read. Its publisher FAQ says “Any public website can appear in ChatGPT search” and doesn’t say how results are ranked. One detail matters for anyone who blocks: if you disallow OAI-SearchBot, ChatGPT can still show your page as a bare link and title when it gets the URL from a third-party provider. To stop that you need a noindex tag, which the crawler has to be allowed to read.
Anthropic runs three crawlers with separate jobs. Blocking ClaudeBot signals that future content should be left out of training. Blocking Claude-SearchBot stops indexing for search and “may reduce your site’s visibility”. Blocking Claude-User stops Claude fetching your page when a user asks. Anthropic also warns that blocking its IP addresses may not work as an opt-out. On how Claude picks between sources it has found, the help pages say Claude “processes multiple sources” and that every web search answer has citations. That’s all.
Google is the most specific about eligibility and the least specific about anything AI-only. A page has to be indexed and allowed a snippet. Beyond that, Google’s guide says you don’t need new machine-readable files, markup or chunked content, and that Search doesn’t use AI text files such as llms.txt. Google also says AI Mode and AI Overviews may use different models, so the links they show will vary. Read plainly: Google’s position is that AI search is Google Search, and its ranking systems decide.
Perplexity‘s crawler page is about access: one bot builds the index, the other fetches pages for users and generally ignores robots.txt because a person asked. Neither is used to train foundation models. Its help centre describes sources as authoritative and top-tier without saying what earns that.
Microsoft is the exception. Bing’s webmaster guidelines now treat grounding and citations in Copilot as outcomes of Bing indexing, and say which pages are “more likely to be selected”: pages on a single topic, pages whose facts and definitions are explicit, pages with essential information near the top, and pages that name entities clearly and consistently. They also say structured data “does not guarantee visibility”. Microsoft is also the only one of the five that reports citation counts and the queries behind them to site owners, through the AI Performance report in Bing Webmaster Tools. Google reports how often links to your pages were shown in its AI features, without queries or clicks.
My reading
Four of the five companies have documented the exit and not the entrance. That isn’t necessarily secrecy. Ranking systems change constantly, and a published list becomes a list people game. But it means two things for anyone advising a business.
First, any claim that “ChatGPT prefers X” or “Perplexity rewards Y” is an inference from outputs, not a rule from the company. Ask how it was measured. The site’s review of 16 studies on backlinks and AI citations found none was a controlled test.
Second, the controls are asymmetric. You can reliably ask most engines to stop training on your pages. You can’t reliably stay out of their answers. ChatGPT and Claude both draw on other companies’ indexes, so blocking an engine’s own crawler doesn’t remove your pages from every index it searches. See what blocking actually means and why robots.txt is a request.
What I’m not claiming
I’m not claiming the four are hiding a rulebook, or that Bing’s list explains Copilot’s choices in full. Bing describes what makes selection “more likely”, not a formula. I’m also not claiming these pages are complete: companies publish in many places, and a page I didn’t find may say more. Google’s pages cover AI Overviews and AI Mode in Search; I didn’t find an equivalent page for the Gemini app, so the Gemini app isn’t covered here. If you find a page I missed, tell me and I’ll add it with the date.
The obvious counterargument is that Google’s position (“it’s just SEO”) is itself selection guidance, since Google has published a great deal about how Search ranks. That’s fair for AI Overviews and AI Mode. It doesn’t tell you anything about the other four.
What to do with this
- Check that each engine’s search crawler can reach you: OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot and Bingbot. Check your CDN’s bot settings as well as robots.txt. Bing warns that content it can’t reliably render may not be selected for grounding.
- If you want out of training but in answers, block the training tokens (GPTBot, ClaudeBot, Google-Extended) and leave the search tokens open. Read training bot, search bot, fetcher first.
- Use the two reports that exist: Search Console (what it can and can’t tell you) and Bing’s AI Performance report. Filter analytics on
utm_source=chatgpt.com. - Treat Bing’s list as the only first-party description of selection. It’s also ordinary good practice, so following it costs little.
Conclusion
The engines have told you how to leave and, Microsoft aside, almost nothing about how they choose. Anyone selling certainty about AI citations is selling their own inference. I’ll re-read these five sets of documents every quarter and log what changes on this page.
Frequently asked questions
Does ChatGPT use Bing?
OpenAI says ChatGPT search uses “third-party search providers” and partner content. The pages I read on 8 November 2026 don’t name them, so I won’t either.
Does Claude use Google or Brave?
Anthropic’s help page names Bing only for image search. Brave Search was added to Anthropic’s subprocessor list in March 2025, and an independent test matched Claude’s citations to Brave results. That’s reported, not stated by Anthropic.
If I block GPTBot, will ChatGPT stop citing me?
No. GPTBot is OpenAI’s training crawler. Citations in ChatGPT search depend on OAI-SearchBot and on third-party providers.
Do I need schema or llms.txt to appear in Google’s AI features?
Google says no special schema and no AI text files are needed, and that Search doesn’t use llms.txt.
Which AI engine tells you when it cites you?
Microsoft reports citation counts and sample grounding queries in Bing Webmaster Tools. Google reports how often links to your pages were shown in its AI features, without queries or clicks. OpenAI tags clicks with a UTM parameter. Anthropic and Perplexity report nothing I could find.
How often is this page checked?
Every quarter. Each change goes in the log below with the date.
Change log
- 8 November 2026: first published. Sources checked 8 November 2026.
Sources
All opened 8 November 2026.
- OpenAI, Introducing ChatGPT search, 31 Oct 2024
- OpenAI, Overview of OpenAI crawlers
- OpenAI, Publishers and developers FAQ
- Anthropic, Does Anthropic crawl data from the web?
- Anthropic, Enabling and using web search
- Simon Willison, Anthropic use Brave, 21 Mar 2025
- Google, AI features and your website, last updated 10 Dec 2025
- Google, Guide to optimizing for generative AI features, last updated 10 Jul 2026
- Perplexity, Perplexity crawlers
- Perplexity, How does Perplexity work?
- Microsoft, Bing Webmaster Guidelines (undated page)
- Microsoft, Introducing AI Performance in Bing Webmaster Tools, 10 Feb 2026