ChatGPT vs Perplexity vs Gemini: how does each platform choose its sources?

ChatGPT, Perplexity and Gemini choose their sources in the same way, but from a different index. All three turn your question into search queries, fetch candidate pages, select the most useful passages from them and cite those. The difference is where they search: Gemini uses Google’s index, Perplexity its own index of more than 200 billion URLs, and ChatGPT in October 2026 combines its own OpenAI index with Google results and search partners such as Bing.

Below: how that selection process works, which index and crawler each platform uses, how little their sources overlap and what you need to do to be mentioned in all three.

How does an AI assistant choose its sources?

An AI assistant chooses sources in five steps, and only in the last step is it decided who gets cited.

1. Search or not. The model decides whether the question needs current information. OpenAI writes that ChatGPT searches the web automatically when the question benefits from it. Without a search, the model answers from its training data, usually without sources.
2. Create search queries. The question is translated into one or more search queries. For AI Overviews and AI Mode, Google calls this query fan-out: several related searches on subtopics at once.
3. Fetch candidates. A search index returns pages that match those queries. This is where the platforms differ most.
4. Select passages. Not whole pages, but pieces of text are judged and re-ranked. Perplexity describes that it splits documents into standalone pieces and ranks them in several rounds, first on words and meaning, then with a heavier model.
5. Answer and sources. The model writes an answer from the selected passages and attaches sources.

Of all the pages fetched in step 3, only a small share makes it to step 5. Which signals decide that is further down.

Which search index does each platform use in October 2026?

	ChatGPT	Perplexity	Gemini
Search index	Mix of its own OpenAI index, Google results through third parties and search partners such as Bing	Its own index, more than 200 billion URLs	Google’s search index
Crawler for search	OAI-SearchBot	PerplexityBot	Googlebot
Runs JavaScript	No	No	Yes, through Googlebot
Opt out of training	Block GPTBot	PerplexityBot is not used for training	Block Google-Extended
Sources in the answer	In 96% of answers, 5 on average	Numbered sources with every answer	In 82% of answers, 8 on average

The details and sources per platform are below; the numbers on sources in the answer come from the Muck Rack measurement from May 2026. Note: none of the three publishes exactly how selection works. What is here is based on the platforms’ own documentation and independent measurements.

ChatGPT: its own index, Google results and Bing

ChatGPT does not use one fixed search engine, but a mix whose split differs per mode. The old advice “ChatGPT works with Bing” is outdated.

• What OpenAI says. The help page on ChatGPT search names “third-party search providers such as Bing”. OpenAI gives no further details.
• What measurements show. The French agency Resoneo analyzed 1,249 ChatGPT answers from July 2026, Search Engine Journal reported. For paid accounts in thinking mode, about 75% of search results came from Google results fetched through third parties and about 24% from OpenAI’s own index. For free accounts, most came from that own index, including small sites without a content deal with OpenAI.
• Caveat. Resoneo reconstructed this from network traffic. Around July 21, 2026 OpenAI removed the label that showed the source of a result, so the ratios can no longer be checked.

Two practical lessons from the same measurement: the text ChatGPT stores from a page usually starts at the page content, not at the meta description, and for 83.6% of cited pages with an H1, that H1 was in the fragment. So your H1 and first paragraph do the work.

Crawler and robots.txt. OAI-SearchBot powers search. OpenAI writes in its crawler documentation that sites blocking OAI-SearchBot do not appear in ChatGPT’s search answers. GPTBot is the crawler for training; you can block it without losing visibility in ChatGPT search. ChatGPT-User fetches pages when a user asks and is not a crawler.

More on ChatGPT in getting found in ChatGPT.

Perplexity: its own index with numbered sources

Perplexity searches its own index and shows numbered sources with every answer, so you see fastest whether you are in it. Perplexity wrote in September 2025 that its index tracks more than 200 billion unique URLs, with capacity for hundreds of billions more.

• Standalone pieces. Perplexity splits pages into pieces that are retrieved and ranked separately. A section that makes sense without the rest of the page has its own chance.
• Freshness. A model decides which pages are indexed and how often they are revisited, based on importance and how often they change. Outdated content is filtered out early in ranking.
• Crawlers. According to Perplexity’s documentation, PerplexityBot fills the search results, respects robots.txt and is not used to train AI models. Perplexity-User fetches pages when a user asks a question and generally ignores robots.txt, because it is a user request.

More on Perplexity in getting found in Perplexity.

Gemini: Google’s index

Gemini gets current information from Google’s search index. Google calls that grounding: for a question, content from the Google index is passed to the model to make the answer more factual and relevant (Google’s documentation on crawlers). There is no separate Gemini crawler; Google uses its existing crawlers.

It follows that:

• Good SEO is the foundation. If you are indexed and score well on the quality signals Google uses, you have the best chance in Gemini. Google writes that its AI features need no extra requirements or special markup.
• JavaScript is less of a problem than with ChatGPT and Perplexity, because Googlebot renders pages.
• Google-Extended is a choice with consequences. If you block Google-Extended in robots.txt, Google may not use your content for Gemini training, nor for grounding in the Gemini apps. Your place in Google Search does not change.

Google AI Overviews and AI Mode are part of Google Search and use the same index. More in getting found in Gemini and getting found in Google AI Overviews.

How much do the platforms’ sources overlap?

The platforms cite largely different pages. In a Rankability analysis of eleven AI platforms across 31 topics (August 2026), no pair of platforms shared more than 24.1% of cited pages. Some other results from that analysis:

• Pages in positions 1 to 3 in Google were cited by at least one AI platform 98.9% of the time.
• 44.8% of the Google top 10 were also among the first ten AI citations.
• 55.2% of the first ten AI citations had no top-10 position in Google.

Citation behavior differs too. Muck Rack analyzed more than 25 million links in May 2026: ChatGPT gave sources in 96% of answers, five on average, and Wikipedia was the most cited domain. Gemini gave sources in 82% of answers, eight on average, with Reddit as the most cited domain.

And the preference for new content differs. In an Ahrefs analysis of almost 17 million citations (July 2025), ChatGPT cited the newest content on average, more than a year newer than regular Google results. Gemini and Perplexity sat in between. Google AI Overviews cited slightly older pages on average than the regular results.

The lesson: a good position in Google helps everywhere, but guarantees nothing. Measure per platform where you stand.

Which signals count on all three platforms?

Despite the different indexes, all three reward the same kind of content. Five signals come back in every study.

1. A direct answer. Pages that introduce the topic in the first 100 words were among the first ten AI citations 22.1% of the time in the Rankability data, against 8.7% without.
2. Verifiable facts. In the GEO paper (Aggarwal et al., KDD 2024), statistics, quotations and cited sources gave the biggest gains in visibility, up to about 40%.
3. Authority through others. 84% of citations in the Muck Rack measurement came from earned media, such as journalism and research, not from the sites of the businesses being mentioned.
4. Freshness. AI assistants cite newer content on average than Google, ChatGPT most strongly.
5. Something that exists nowhere else. Google asks in Creating helpful content whether you offer original information, research or analysis. Ten copies of the same advice give a model no reason to choose you.

How to work these signals into your writing is in content that AI cites. How to build authority outside your own site is in E-E-A-T for AI.

Why are you mentioned in one platform and not in another?

Usually it is the technical setup or the index, not your text. Go through these points before you start rewriting content.

1. Robots.txt. Do you allow OAI-SearchBot and PerplexityBot? If you block them, you do not appear in ChatGPT’s and Perplexity’s search answers. See how AI crawlers read your site.
2. JavaScript. The crawlers of OpenAI, Anthropic and Perplexity do not run JavaScript; Googlebot does (Vercel). If your text only appears after scripts load, ChatGPT and Perplexity see an empty page while Gemini reads it. The technical scan shows what a crawler sees without JavaScript.
3. Google index. If the page is not in Google, you miss Gemini, AI Overviews and the part of ChatGPT that leans on Google results. Check indexing in Search Console.
4. Bing. Bing Webmaster Tools takes little time and covers Bing and Microsoft Copilot.
5. Freshness. If your page has not been updated in years while competitors published recently, you have less of a chance, especially in ChatGPT and Perplexity.

What to do per platform

Everywhere

• Important content as plain text in the HTML.
• The answer in the first 100 words, H2s as questions, sections that stand alone.
• Numbers with source and date.
• Mentions on strong external sites.

ChatGPT

• Allow OAI-SearchBot. You may block GPTBot if you do not want to contribute to training.
• Make sure your H1 and first paragraph together form a standalone answer.
• Good Google SEO, because part of the results come from Google.

Perplexity

• Allow PerplexityBot.
• Keep pages current; outdated content drops out early.
• Write each section so it can be cited on its own.

Gemini

• Regular Google SEO: indexing, quality, speed.
• A deliberate choice about Google-Extended.
• Author and business information Google can link.

Which platform do you start with?

Start with what works on all three, then measure per platform. A good position in Google counts twice: Gemini and AI Overviews draw directly from the Google index, and ChatGPT got a large share of its results from Google too in the Resoneo measurement. Perplexity and ChatGPT additionally need access for their crawlers and text that is readable without JavaScript.

Because the sources overlap so little, you cannot extrapolate from one platform to another. Test the same questions in all three. GreatCompany’s prompt tracker measures every day whether ChatGPT and other AI models mention your business.

Read also

• Content that AI cites: how to write for ChatGPT, Perplexity and Google AI
• E-E-A-T for AI: build authority search engines and AI recognize
• Getting found in AI search engines
• GEO glossary