We pulled together four studies and each vendor's own documentation to compare how ChatGPT, Perplexity, Claude, and Gemini cite brands. Perplexity searches on every answer. Claude and Gemini each decide per prompt whether to search, and Profound measured Claude doing that on about a third of prompts. OpenAI doesn't document when ChatGPT decides to search. That's also why your brand can appear on one of these AEO models and not another for the same question.
How Each Model Decides Whether to Search
Claude and Gemini both choose, on their own terms, whether to answer from what they already know or to search first. Perplexity always searches. That choice shapes which of your content the model draws on for that answer.
According to Perplexity's own Help Center, it "uses advanced AI to search the internet in real-time," and "every answer includes citations." There's no training-only mode to plan around.
Claude searches selectively. According to Profound's study of Claude's citation behavior, Claude triggered a web search on 36.6% of tested prompts. Queries with "best," "near me," or a current year tended to trigger a search. Basic "what is" or "how does" questions tended to stay inside the model.
In the Gemini API, the model itself decides whether to search. Google's own documentation says "the model analyzes the prompt and determines if a Google Search can improve the answer." Nobody has published a measured trigger rate for Gemini, the way Profound measured one for Claude. This documentation also covers the API rather than the consumer Gemini app.
OpenAI doesn't document how or when ChatGPT decides to search. What its bot documentation does confirm: OAI-SearchBot is the crawler that surfaces websites in ChatGPT's search results. A site that blocks it won't be shown in ChatGPT's search answers, though it can still appear as a navigational link. You can allow or block OAI-SearchBot and GPTBot, OpenAI's training crawler, independently. If you allow both, OpenAI says it "may use the results from just one crawl for both use cases" to avoid duplicate crawling.
Where Each Model Pulls Its Citations From
Once a model decides to search, it still has to pick a source, and that pick shapes what shows up in the answer.
Claude's web search runs through Brave. According to Google Cloud's own documentation of Claude on its platform, Brave Search is the third-party service Anthropic's web search feature calls. According to Profound's study, 79.2% of Claude's cited URLs also appeared in Brave's own top 10 results for the same query.
In the Gemini API, grounding runs through Google Search. OpenAI doesn't publish which search provider sits behind ChatGPT's results; what it documents is that blocking OAI-SearchBot keeps a site out of ChatGPT's search answers. Perplexity's Help Center describes real-time web search on every query, without naming a specific backend.
What Kind of Source Each Model Prefers
Even models searching the same web don't cite the same kind of source. According to Chen et al.'s unreviewed preprint from the University of Toronto (Sept 2025), earned sources make up 63.4% of Gemini's citations for well-known brands. Earned sources means independent media, review, and comparison sites. The share rises to 66.4% for niche brands. Perplexity's earned-source share runs 67.4% and 73.4%. Claude's runs 87.3% for well-known brands and drops to 86.3% for niche brands. Claude is the only one of the four where that share falls instead of rises. ChatGPT leans on earned sources most heavily: 93.5% for well-known brands, 95.1% for niche ones.
The picture shifts for local, buyer-intent questions specifically. Yext, which sells AI-visibility and listings software, analyzed 155.5 million citations from local-intent questions. For those questions, a brand's own website is the top source type overall. That's 48.8% of OpenAI's citations, 52.0% of Gemini's, 50.9% of Perplexity's, and 42.6% of Anthropic's. Sources a brand can keep current, its own site plus the directory listings it manages, cover most citations across all four. That's 85% for OpenAI, 81% for Gemini, 80% for Perplexity, 69% for Anthropic. With AI Citation Tracking, we run buying-intent prompts across each major LLM every week. We analyze which brands appear, in what context, and with what sentiment.
How Often Each Model's Citations Overlap Google's Top 10
A brand that ranks well on Google isn't guaranteed the same standing in an AI answer. We cover that mismatch in more depth in why your brand ranks on Google but not in ChatGPT or Perplexity. According to Ahrefs' analysis of AI-cited URLs (Aug 2025), Perplexity's citations overlap Google's top 10 most often among the assistants tested, at 28.6%. Gemini's overlap is 8.6%, and ChatGPT's runs 8.0% for in-text citations and 6.1% for its separate references list. Claude wasn't part of this particular study, so it has no comparable figure here.
Why Your Brand Might Appear on Perplexity but Not ChatGPT
The gap usually traces back to whether the model searches at all, and what kind of source it counts once it does. Perplexity searches on every query, so a page that's live and crawlable has a real shot at appearing. Perplexity attaches citations to every answer. ChatGPT splits its citations into an in-text list and a separate references list. Ahrefs measured each one separately: 8.0% and 6.1% overlap with Google's top 10. ChatGPT leaves out of its search answers any site that blocks OAI-SearchBot. It also leans harder on earned third-party coverage than the other three engines: 93.5% against Perplexity's 67.4%, for well-known brands.
Perplexity also cites community sources that ChatGPT leaves out entirely. According to Chen et al., 23.8% of Perplexity's citations for well-known brands came from Reddit, YouTube, and forums, against none for ChatGPT. If your brand is discussed mainly in communities rather than in the press, you're more likely to show up on Perplexity than on ChatGPT. We go deeper on tactics for each of these three engines specifically in how to get your brand cited across Claude, Gemini, and Perplexity.
Comparison at a Glance
| ChatGPT | Perplexity | Claude | Gemini | |
|---|---|---|---|---|
| When it searches | Not documented by OpenAI | Every answer | ~36.6% of prompts (Profound) | Per prompt, decided by the model (Gemini API) |
| Search backend | Not published by OpenAI | Not published by Perplexity | Brave Search | Google Search (Gemini API) |
| Earned-source share, well-known brands (Chen et al.) | 93.5% | 67.4% | 87.3% | 63.4% |
| Overlap with Google's top 10 (Ahrefs) | 8.0% (in-text) | 28.6% | Not in Ahrefs' study | 8.6% |
FAQs
Why does my brand appear on Perplexity but not ChatGPT?
Perplexity searches live on every query and cites community sources like Reddit and forums that ChatGPT doesn't cite at all for well-known brands. ChatGPT leaves out any site OAI-SearchBot is blocked from crawling, and it leans more heavily on earned third-party coverage than Perplexity does. A brand discussed mainly in communities rather than the press can clear Perplexity's bar without clearing ChatGPT's.
Does Claude search the web like the other three engines?
Less often than Perplexity, which searches on every answer. In Profound's tests, Claude searched on about 36.6% of prompts. OpenAI and Google don't publish a comparable rate for ChatGPT or Gemini. When Claude does search, it runs through Brave Search.
Which AI model relies most on a brand's own website?
For local, buyer-intent questions, Gemini and Perplexity draw the most from a brand's own site, at 52.0% and 50.9% of citations respectively, per Yext's research. For broader category questions, Chen et al.'s preprint shows all four engines leaning toward earned third-party sources instead, with ChatGPT leaning hardest.
Do OpenAI, Anthropic, and Perplexity crawl a site the same way?
Every main crawler from all three respects robots.txt, including Anthropic's ClaudeBot and Claude-SearchBot, OpenAI's OAI-SearchBot, and Perplexity's PerplexityBot. The difference shows up in the fetchers a live user session triggers. Perplexity-User "generally ignores" robots.txt, ChatGPT-User's rules "may not apply," and Anthropic's Claude-User still respects it. If you set a crawler block years ago and forgot about it, that's worth checking.
The Bottom Line
Perplexity searches on every answer, so it rewards pages it can crawl today. ChatGPT and Claude lean hardest on independent coverage, so they reward being written about elsewhere. We'll show you where your brand stands on both fronts with a free AI visibility assessment. We test your brand on ChatGPT and Gemini and audit what AI crawlers can read on your site.
