News, business, technology and lifestyle
Grab The Data

How AI Search Engines Choose Which Websites to Cite

Google AI Overviews, ChatGPT and Perplexity each pick their sources differently. Here is what the evidence in 2026 actually shows, and what it means for anyone who publishes on the web.

A growing share of searches now end with an AI-written answer rather than a list of ten blue links. Google shows AI Overviews above its results, ChatGPT answers with live web search, and Perplexity was built around cited answers from the start. Each of these answers carries a handful of links, and for publishers those links are quickly becoming the new front page of search.

So how do these systems decide which websites to cite? The honest answer is that none of the companies publishes its ranking formula, and most of the statistics in circulation come from SEO software vendors measuring the systems from the outside. But enough independent measurement now exists to draw a reasonably clear picture.

Google AI Overviews: still built on Google’s index

Google is the only one of the three that has documented the basics. In its guidance for site owners, Google says a page needs to be “indexed and eligible to be shown in Google Search with a snippet” to appear as a supporting link in AI Overviews or AI Mode, and that “there are no additional technical requirements.” It also says there is no special structured data or AI-specific markup to add.

What is less obvious is that AI Overviews do not simply quote the top results. Google describes a technique it calls query fan-out: the system runs several related searches behind the scenes and gathers supporting pages from all of them. That is why a page can be cited for a question it never ranked for.

Outside measurement backs this up and suggests the effect is growing. In July 2025, Ahrefs analysed 1.9 million citations from one million AI Overviews and found that 76% of cited pages ranked in Google’s top 10 for the same query. A repeat analysis reported in 2026, covering 863,000 keywords, put that share at roughly 38%, with the rest split almost evenly between pages ranking 11 to 100 and pages outside the top 100. Ahrefs notes that part of the change comes from improved parsing, so the two numbers are not perfectly comparable, but the direction is clear: ranking well still helps, and it no longer decides the outcome on its own.

ChatGPT: a different index, different winners

ChatGPT’s search feature retrieves pages through its own search pipeline, which draws heavily on Microsoft Bing’s index, and then selects which pages to quote. Two consequences follow.

First, ranking on Google matters less here than on Google’s own product. Several vendor studies report that a large share of ChatGPT’s citations come from pages that are not in Google’s top results; one widely quoted analysis puts it near 90%, while another finds most ChatGPT citations line up with Bing’s top ten. The studies use different methods and samples, which is why they disagree, but both point away from Google rankings as the deciding factor.

Second, ChatGPT leans heavily on encyclopedic and reference sources. Wikipedia appears near the top of nearly every published list of its most-cited domains.

Perplexity: freshness and community discussion

Perplexity runs its own index and visibly favours recent material and community discussion. Vendor measurements consistently report Reddit among its most-cited sources and a strong preference for content published in the last few weeks. Treat the precise percentages with caution, since they come from single companies selling optimisation services, but the pattern is consistent across them.

The engines mostly disagree with each other

Perhaps the most useful finding is how little overlap there is between the three. Estimates published this year suggest only a small minority of domains are cited by both ChatGPT and Perplexity for the same topics. A page that dominates one engine can be invisible in another, which means there is no single “AI SEO” checklist that wins everywhere.

There is one common thread. A Muck Rack study found that the large majority of links cited by AI systems point to earned media: independent coverage that brands neither own nor pay for. Across all three systems, being the kind of source other people reference appears to matter more than any formatting trick.

What this means for publishers

The evidence supports a few practical conclusions, and rules out a few popular myths:

  • Get the basics right first. For Google, a page that is not indexed or cannot show a snippet cannot be cited at all. Blocking crawlers, or adding nosnippet, removes you from the running.
  • Answer the question directly. Because AI Overviews gather pages from related searches, pages that answer a specific question clearly, near the top, have more chances to be picked up than pages that bury the answer.
  • Keep facts current and dated. Freshness clearly matters to Perplexity and helps everywhere. Visible publication and update dates make that easy to judge.
  • Be worth citing. Original data, first-hand reporting and clear explanations attract links from other sites, and earned coverage is what AI engines lean on most.
  • Ignore magic formulas. Google states plainly that no special schema or AI-only file is required. Claims that one tag or file guarantees AI citations are not supported by the evidence.

AI search is changing quickly, and the numbers above will move again. But the underlying logic is stable: these systems are trying to find the clearest, most trustworthy page for a question, and the publishers who keep being cited will be the ones who keep being worth citing.

Sources