Short answer
AI assistants pull answers from their own indexes and their own fetches: ChatGPT queries an OpenAI-built index where only 1.5% of results overlap Bing’s top list, Copilot builds on Bing, Claude on Anthropic’s own search crawler, and Perplexity on its own. Profound’s analysis of 11.84 billion citations shows 57% point at brand-owned websites.
A common assumption is that ranking in Google automatically puts you in every AI answer. In 2026 that became measurably wrong: the assistants have built their own retrieval chains, and they differ on points you can act on.
Which index does each assistant use?
ChatGPT now searches an index OpenAI operates itself. A technical analysis in Search Engine Land in August 2026 found that only 1.5% of its results appear in Bing’s top 20 for the same queries, so Bing rank no longer predicts ChatGPT visibility. The analysis also found that pages over 4 MB are rejected, that JavaScript is never executed, and that the opening lines of the content carry the most weight.
Copilot builds on Bing’s index, and in February 2026 Bing wrote generative engine optimization into its official guidelines: the NOARCHIVE tag removes a page from Copilot answers entirely, while NOCACHE limits display to title and link. Anthropic documents its own search crawler, Claude-SearchBot, and states plainly that blocking the fetch agent Claude-User can cost visibility in Claude’s answers. Perplexity runs its own index through PerplexityBot, and Meta documents a search index of its own, built by its Meta-WebIndexer bot.
Which sources actually get cited?
Profound analyzed 11.84 billion citations from April through July 2026 and found 57% point at brand-owned websites: Gemini highest at 69%, ChatGPT lowest at 47%. Your own site is the most common citation source, which makes AEO work on your own pages the base investment. On May 7, 2026, ChatGPT also began linking brands directly in its answers, and Profound’s measurement shows referral traffic roughly doubling overnight.
What can you measure for free?
Bing Webmaster Tools shipped its AI Performance report in February 2026, with citations and the actual grounding queries, and June added Citation Share, your slice of the citations per query. Together with Google’s generative AI reports, AI visibility in both major ecosystems is now trackable on real numbers.
What works, according to the research?
A systematic review on arXiv in July 2026 covering 45 GEO studies found that no reviewed technique shows a stable effect on discoverability across platforms and time. What holds are the fundamentals: being retrievable, answering early and self-contained, and being the source with its own numbers. Promises of guaranteed citations simply have no support.
Common questions
- Does ranking well in Google put my site in ChatGPT?
- There is no automatic link. ChatGPT searches OpenAI’s own index, and the Search Engine Land analysis from August 2026 showed its results rarely coincide even with Bing’s top list. Check your visibility by asking your most important customer questions directly in ChatGPT.
- How do I measure citations in Copilot?
- With AI Performance in Bing Webmaster Tools, launched in February 2026. The report shows how often your pages are cited in Copilot and Bing’s AI answers and which grounding queries retrieved the content, and since June, Citation Share shows your slice of the citations per query.
- Do all assistants cite the same kinds of sources?
- No. In Profound’s analysis of 11.84 billion citations, 69% of Gemini’s citations went to brand-owned sites against 47% for ChatGPT, which lifts third-party sources more often. Your own pages and presence on credible third-party surfaces therefore complement each other.
- What does the NOARCHIVE tag do to my visibility in Copilot?
- It removes the page from Copilot answers entirely, per Bing’s February 2026 guidelines. NOCACHE is the middle setting: Copilot may then only show title, link and a short snippet. Treat the tags as deliberate controls, because tags inherited by accident cost visibility silently.
- Do the first lines of a page matter extra for ChatGPT?
- Yes. The analysis of OpenAI’s retrieval chain shows the system weighs the content right after the main heading most heavily, and the recommendation is that the first two hundred characters carry the page’s core message. That coincides with the rule of opening with a self-contained direct answer.
- Do I get paid when AI cites my content?
- As a rule, no. Perplexity shares revenue with enrolled publishers through Comet Plus, and large media houses sign licensing deals, but for an ordinary company site the compensation is the visibility itself. You collect the value by making the answers lead on to pages that convert.
Sources
- Inside ChatGPT’s retrieval stack — Search Engine Land
The August 2026 analysis: own index, 1.5% Bing overlap, the 4 MB limit, and the weight of the opening lines.
- Bing Adds GEO To Official Guidelines — Search Engine Journal
Bing’s February 2026 guidelines, with the NOARCHIVE and NOCACHE effects in Copilot.
- Introducing AI Performance in Bing Webmaster Tools — Bing Webmaster Blog
The February 2026 launch of citation reporting and grounding queries.
- New AI Visibility Insights in Bing Webmaster Tools — Bing Blog
Citation Share and the expanded AI reports from June 2026.
- Does Anthropic crawl data from the web? — Anthropic Help Center
The documentation of Claude-SearchBot and the statement that blocking Claude-User may reduce visibility, updated April 2026.
- Meta web crawlers — Meta for Developers
The documentation of Meta-WebIndexer, the bot building Meta AI’s search index.
- Where do AI citations come from? — Profound
The analysis of 11.84 billion citations: 57% to brand-owned sites, 69% for Gemini and 47% for ChatGPT.
- Optimizing Visibility in Generative Engines: A Critical Survey — arXiv
The review of 45 GEO studies: no technique shows a stable, cross-platform effect on discoverability.
Tell us what you want to build
Thirty minutes, free of charge, and a straight answer on whether we are the right studio for it.