AI Search Engines: How 7 of Them Find Sources

AI search engines are tools that answer a question in prose, built from web pages they retrieve while you wait, and cite the pages they used. The seven that matter most for a business in the US right now are ChatGPT search, Google AI Overviews and AI Mode, the Gemini app, Microsoft Copilot, Perplexity, Claude with web search, and Brave. They look alike on the surface. Underneath, they pull from different indexes, send different crawlers, and cite in different ways, and that is what decides whether your site shows up.

This page is for owners and marketers who want to know where each engine gets its sources, checked against each vendor’s own documentation on September 22, 2026. It is not a “best tool” ranking. If you want tactics for a single engine, we link the deep dives as we go.

On this page

• What makes something an AI search engine?
• Where each AI search engine gets its sources
• The engines one by one
• Which crawlers feed AI search engines?
• How each engine cites
• What this means for your website
• Getting found across AI search engines

What makes something an AI search engine?

An AI search engine retrieves pages at the moment you ask, has a language model read them, and writes an answer that links back to its sources. That last part is the line between a search engine and a chatbot answering from memory. A model answering only from training data cannot cite a page it never fetched, and it cannot know what changed last week.

Most of these products now search on their own when a question needs current information. OpenAI’s help center says ChatGPT may search the web automatically when a question would benefit from current information, and Anthropic’s help center says that in the newer Claude experience there is no web search toggle and Claude searches when it helps. So the question for a website owner is no longer “does this engine search?” It is “which index does it search, and am I in it?”

The mechanics of how an engine narrows a big pool of retrieved pages down to the few it quotes are covered in retrieval vs citation. This page is about the step before that: where the pool comes from.

Where each AI search engine gets its sources

Each engine draws on one of three kinds of source: Google’s index, Bing’s index, or an index built by its own crawler. Some combine two. The diagram below maps what each vendor documents.

Two things stand out. First, Bing matters far more than its classic market share suggests, because Copilot runs on it and OpenAI names Microsoft among ChatGPT’s search partners. Second, three engines send their own crawlers, which means your robots.txt and your CDN rules decide whether you are even a candidate.

The engines one by one

Each engine below is described only from its own vendor’s pages. Features change often, so treat anything with a date as a snapshot.

ChatGPT search. It has the widest reach of the chat-first engines: OpenAI says its ad-supported free tier helps keep ChatGPT available to more than 1 billion weekly active users. OpenAI’s help page lists web search on Free, Go, Plus, Pro, Business, Enterprise and Edu, and for people who are not signed in. When it uses partner providers, ChatGPT “typically rewrites your query into one or more targeted queries” and may send follow-up queries. The providers whose privacy policies OpenAI links are Microsoft and Shopify. To be eligible, OpenAI says to allow OAI-SearchBot and make sure your host or CDN allows its published IP addresses. Tactics are in how to rank in ChatGPT search.

Google AI Overviews and AI Mode. These reach the most people: at I/O in May 2026, Google said AI Overviews had over 2.5 billion monthly active users and AI Mode had passed 1 billion. Google’s AI features guide for site owners says both “may use a ‘query fan-out’ technique”, issuing related searches across subtopics. To be a supporting link, a page must be indexed and eligible to show with a snippet, and “there are no additional technical requirements.” AI Overviews only appear when Google judges them additive, “and as such, often don’t trigger.” The deeper mechanics are in Google AI Mode explained.

Gemini app. Google reported 950 million monthly active users for the Gemini app in July 2026. Google’s common crawlers documentation describes grounding in Gemini Apps as “providing content from the Google Search index to the model at prompt time.” The Google-Extended robots.txt token controls whether your content is used for that grounding and for Gemini training. It does not affect Google Search inclusion or ranking.

Microsoft Copilot. Bing’s webmaster guidelines state that Bing and Copilot “rely on the same core crawling, indexing, and ranking foundation as traditional search,” and recommend IndexNow so Copilot does not cite outdated URLs. More on that in our Bing SEO guide.

Perplexity. Perplexity’s help center says it searches the internet in real time and that each answer includes numbered citations. Its crawler documentation says PerplexityBot surfaces and links sites in results and “is not used to crawl content for AI foundation models.” How it picks among candidates is covered in how Perplexity chooses sources.

Claude with web search. Anthropic’s help center says every web search response includes citations and that image results are powered by Bing. The article does not name the provider behind text results, so neither do we. Anthropic runs its own search crawler, Claude-SearchBot. Details are in Claude web search.

Brave. Brave says it delivers results from its own, built-from-scratch index, with AI answers through Ask Brave on the same results page. Of the engines here, it is the one that states most plainly that it depends on neither Google nor Bing.

Which crawlers feed AI search engines?

Three vendors split their bots into a search crawler, a user-triggered fetcher and a training crawler, and blocking one does not block the others. That split is the most useful thing a site owner can learn about AI search engines, because it lets you opt out of training without disappearing from answers.

The specifics, from each vendor:

• OpenAI. Per its bots page, sites that opt out of OAI-SearchBot “will not be shown in ChatGPT search answers,” and changes take “~24 hours” to apply. Disallowing GPTBot signals that content should not be used for training. ChatGPT-User handles user actions, and “robots.txt rules may not apply.”
• Anthropic. Its crawler help article says blocking Claude-SearchBot “may reduce your site’s visibility and accuracy in user search results,” while blocking ClaudeBot excludes future content from training.
• Perplexity. Perplexity-User “generally ignores robots.txt rules” because a person requested the fetch.

The full robots.txt syntax, with a sample file, is in robots.txt for AI crawlers. Whether you should block training at all is a separate business call.

How each engine cites

Every engine here links its sources, but the format changes how much traffic a citation sends. The table sums up what each vendor documents.

The measurement column is where things have changed most, and it matters because a citation is not a click. A Pew Research Center study found Google users clicked a traditional result on 8% of visits with an AI summary, against 15% without one. Google says clicks from AI Overviews and AI Mode are counted in Search Console’s Performance report under the Web search type, not split out. Microsoft went further: in February 2026 it launched AI Performance in Bing Webmaster Tools, a public preview that shows total citations, cited pages and the “grounding queries” behind them across Copilot and Bing’s AI summaries. For the other engines you are left with referral data and manual prompt checks, which we set up in tracking AI traffic in GA4 and Search Console.

What this means for your website

The practical takeaway is that three indexes cover almost every AI search engine: Google’s, Bing’s, and the engines’ own crawlers. Get into all three and you are a candidate everywhere. Miss one and a whole group of engines can’t cite you, however good the page is.

A few honest caveats:

• Being indexed is necessary, not sufficient. Eligibility gets you into the candidate pool. Being quoted depends on whether a passage answers the sub-question cleanly.
• Nobody publishes ranking weights. Google says there are no extra technical requirements for AI features. OpenAI says ChatGPT “ranks search results using multiple factors” and placement “is not guaranteed.” Anyone promising a guaranteed spot is guessing.
• Some people want none of this. A real share of searchers now look for search engines without AI, and classic blue-link rankings still matter to them.

Getting found across AI search engines

Getting found across AI search engines comes down to four steps in order: be in Google’s index, be in Bing’s index, let the engines’ own search crawlers in, then measure what gets cited. The diagram shows the sequence and what each step covers.

Steps one and two are ordinary SEO hygiene, and most sites have gaps in the second. Step three is where accidental blocks tend to hide, often in a CDN bot rule rather than in robots.txt, which is why Google lists CDN access in its own AI features guidance. Step four keeps you honest, because what an engine cites changes month to month.

If you want the whole picture across engines rather than one at a time, our GEO service starts with exactly this audit: index coverage in Google and Bing, crawler access at the CDN, and a prompt set that shows where you are cited and where a competitor is.