ChatGPT vs Perplexity vs Google AI Mode: How Each Picks Sources

At a glance: seven dimensions

Each cell is drawn from the vendor documentation linked in the sections below. Where a vendor says nothing, the cell says so.

Sources: OpenAI, Overview of OpenAI crawlers; Perplexity Help Center, How does Perplexity work?; Perplexity, PerplexityBot and Perplexity-User documentation; Google, AI Mode in Search announcement (May 20, 2025); Google Search Central, AI features and your website. Accessed October 2026.

ChatGPT search: one documented gate, no documented ranking

OpenAI’s crawler documentation is narrow and useful. It names the agent that controls whether a site can appear in ChatGPT search, and it states the consequence of blocking it. OAI-SearchBot “is used to surface websites in search results in ChatGPT’s search features,” and “sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers,” though they can still appear “as navigational links” (OpenAI, Overview of OpenAI crawlers). A robots.txt change takes “~24 hours” to propagate on OpenAI’s side.

Two more agents matter. “When users ask ChatGPT or a CustomGPT a question, it may visit a web page with a ChatGPT-User agent,” and “because these actions are initiated by a user, robots.txt rules may not apply.” That agent “is not used to determine whether content may appear in Search.” GPTBot is the training crawler, and OpenAI describes each setting as “independent of the others,” so a GPTBot block neither removes you from search nor protects you from being omitted if OAI-SearchBot is blocked. IP ranges are published for all three.

What OpenAI does not document is how ChatGPT ranks the pages it can reach, how fresh its index is, or what the index is built from. Any claim about ChatGPT preferring a word count, a schema type or a domain age is inference. The documented work is access: allow OAI-SearchBot, return a plain 200 to ChatGPT-User, and be indexable. The full reading is in how ChatGPT search picks sources.

Perplexity: live retrieval, visible citations, two agents

Perplexity describes its answering process in plain terms: it “uses advanced AI to search the internet in real-time, gathering insights from top-tier sources,” and “each answer includes numbered citations linking to the original sources” (Perplexity Help Center, How does Perplexity work?). That visible citation list is the reason Perplexity is the easiest engine to audit: ask a question, read the numbers, record the domains.

Its crawler documentation names two agents. PerplexityBot is “designed to surface and link websites in search results on Perplexity” and “is not used to crawl content for AI foundation models.” Perplexity-User “might visit a web page to help provide an accurate answer” and may “include a link to the page in its response”; it “generally ignores robots.txt rules” because a person requested the fetch (Perplexity, PerplexityBot and Perplexity-User documentation). For firewall rules Perplexity recommends you “combine both User-Agent string matching and IP address verification.”

Perplexity does not publish how it ranks or selects among sources beyond calling them top-tier and authoritative. The levers that follow from the documentation are being indexable by PerplexityBot, being fetchable by Perplexity-User without a challenge page, and having a direct answer near the top of the page. See how Perplexity chooses the sources it cites.

Google AI Mode: query fan-out on top of ordinary indexing

Google is the most explicit about mechanics. AI Mode “uses our query fan-out technique, breaking down your question into subtopics” and issuing “a multitude of queries simultaneously on your behalf.” Deep Search “uses the same query fan-out technique but taken to the next level” and “can issue hundreds of searches” (Google, AI Mode in Search announcement (May 20, 2025)). A page is cited when it is the best answer to one of those sub-queries, not only to the head question.

Eligibility is ordinary search eligibility. Google’s documentation says a page must be indexed and eligible to show with a snippet, that “there are no additional technical requirements,” and that “there’s also no special schema.org structured data that you need to add” (Google Search Central, AI features and your website). Googlebot’s robots rules govern AI features; the controls that limit display are nosnippet, data-nosnippet, max-snippet and noindex. Google-Extended is not the control. Traffic from AI Overviews and AI Mode is reported in the Performance report under the Web search type, and recrawl “can take anywhere from several days to several months.”

Google’s own advice for these surfaces is to “focus on making unique, non-commodity content,” and it notes that AI Overviews “show a wider range of sources on the results page” and that clicks from them are “higher quality, where users are more likely to spend more time on the site” (Google Search Central blog, Top ways to ensure your content performs well in Google’s AI experiences (May 21, 2025)). What Google does not document is how sub-queries are generated or which page wins each one. The practical response is one clear sub-answer per section, which we walk through in how Google AI Mode picks sources.

Claude, briefly

Claude is a fourth engine worth a paragraph. Anthropic documents that “Claude determines when to search based on the prompt,” that searches can “repeat multiple times throughout a single request,” that citations are “always enabled,” and that each cited passage is “up to 150 characters of the cited content” (Anthropic, Web search tool documentation). Three agents are named: Claude-SearchBot indexes for search, Claude-User fetches for users, ClaudeBot trains, and all honor robots.txt (Anthropic, Does Anthropic crawl data from the web?). The 150-character window is the most concrete writing instruction any vendor has published. More in how Claude finds and cites sources.

What all three need from your website

Strip away the differences in retrieval and the documentation converges on four conditions. They are the spine of the AI Search Playbook, and every engine above fails a site on at least one of them before any ranking question arises.

1. Reachable

   OAI-SearchBot, PerplexityBot, Claude-SearchBot and Googlebot must be allowed, and the user-initiated agents must get a normal page rather than a challenge. The content must exist in the HTML as delivered; none of the non-Google vendors documents JavaScript rendering. See which AI crawlers visit your site.

2. Unambiguous

   The page states who you are, what you offer, where and for whom, and the same facts appear in structured data and on your profiles. An engine that cannot resolve the entity cannot attribute the answer.

3. Quotable

   A sentence that answers the question, short enough to lift. Anthropic’s 150-character citation window and Perplexity’s numbered citations both reward a direct answer near the top of the page over a paragraph that implies one.

4. Corroborated

   Profiles, directories, reviews and press that repeat your facts. Google’s advice to produce unique, non-commodity content and Perplexity’s preference for sources it calls authoritative both point at the same thing: being confirmed by someone other than yourself.

Where to focus first, by business type

This is reasoning from the documentation above, not measured share. Your own prompt panel should overrule it within two months.

Whatever the starting engine, the sequence is the same: fix access, state identity once, give each key page a quotable answer, earn corroboration, then measure monthly. How that differs from classic SEO is covered in SEO vs GEO and GEO vs AEO.

Keep going