Which websites ChatGPT, Google AI Overviews and Perplexity cite most

The websites ChatGPT, Google AI Overviews and Perplexity cite most change from one engine to the next, and the three lists have few names in common even for the same questions. ChatGPT, Google AI Overviews and Perplexity share only 2 of their top 10 cited sources when answering the same questions about the same brand, according to Agenzy’s 2026 source study (n=109 business projects).

The study counted 2,283,923 citations behind 394,180 AI answers about 130 businesses, tracked as 138 market projects in Peec AI, an AI search tracking tool, from 1 January 2025 to 7 September 2026. 104 of the 138 projects ran from Lithuania, so the websites named below describe that market first.

How much AI engines’ sources overlap

AI engines’ cited sources overlap by a handful of websites for any one brand. In the 2026 source study, a median of 4 websites appeared in both ChatGPT’s and Google AI Overviews’ top 10 (n=119 projects), 3 in both ChatGPT’s and Perplexity’s (n=128), and 4 in both Google AI Overviews’ and Perplexity’s (n=109).

The median number of websites each pair of engines shared in its top 10, against the number all three shared:

Any two of the engines share 3 or 4 of their 10 most cited websites, and all three together share 2.

For some brands the engines agree on nothing at the top: in 14.7% of projects, 16 of 109, no website appeared in all three top-10 lists. Extending each engine’s list to its 25 most cited websites lifts the three-engine median to 6 of 25 (n=103), so the gap holds further down the lists as well.

The comparison counts only projects where each of the three engines cited at least 10 websites, which is why its sample is 109 of the study’s 138 projects. For a brand, a citation on one engine says little about its standing on the other two.

The website each engine cites most

ChatGPT, Google AI Overviews and Perplexity each have a different most cited website in the 2026 source study: Reddit on ChatGPT, YouTube on Google AI Overviews and vz.lt, a Lithuanian business daily, on Perplexity. Reddit’s 3.0% of ChatGPT citations is the largest share any website gets on that engine, and YouTube takes 4.7% of Google AI Overviews citations.

Reddit and YouTube each lead on one engine only, and YouTube gets 0.03% of ChatGPT’s citations. Perplexity’s leader comes from the sample: vz.lt is among the ten most cited websites in the whole study, beside the Lithuanian government portal lrv.lt and the directory paslaugos.lt, and 104 of the 138 projects ran from Lithuania.

A US dataset puts different leaders on two of the three engines. Tomek Rudzki, GEO Expert at Peec AI, analyzed 30 million sources from US answers on ChatGPT, Google AI Mode, Gemini, Perplexity and Google AI Overviews in a study published on 31 March 2026. YouTube came first on Google AI Overviews there as well, while Wikipedia led ChatGPT and Reddit led Perplexity.

Rudzki wrote that “Reddit and YouTube appear across all five platforms”, and below those two his lists split, with Wikipedia strong on ChatGPT and Perplexity. In both datasets the leader depends on the engine, and the websites under it change with the market the questions come from.

How concentrated AI citations are

AI citations spread across a large number of websites overall and bunch up inside each category. The 10 most cited websites hold 8.7% of all 2,283,923 citations in the 2026 source study, pooled across 138 projects, while inside the median project the top 10 hold 35.3% of that project’s citations.

The pooled shares against the share inside a single project:

Pooled, even the 100 most cited websites reach only 21.9% of citations, while the top 10 inside one project hold more than a third.

The pooled list runs to at least 100,400 websites, so a global ranking of cited websites says little about any single category. A category’s own list is short enough to work through, and where a brand’s own site sits on it is the subject of how often LLM citations include a brand’s own website.

Kinds of sites each engine prefers

Each AI engine favors a different kind of site. In the 2026 source study, ChatGPT took 11.9% of its citations from official sites of government, health and EU bodies, Google AI Overviews took 10.8% from user-generated content, and Perplexity took 13.6% from news and media (n=138, 137 and 133 projects), by Peec AI’s classification of each cited website.

The three source types where the engines differ most, as a share of each engine’s citations:

Each engine leads on one type, and the other two engines sit well below it on that type:

• Official sites take 11.9% of ChatGPT’s citations, 8.5% of Perplexity’s and 5.5% of Google AI Overviews’.
• User-generated content takes 10.8% of Google AI Overviews’ citations, 4.8% of ChatGPT’s and 3.8% of Perplexity’s.
• News and media take 13.6% of Perplexity’s citations, 7.7% of ChatGPT’s and 5.2% of Google AI Overviews’.

User-generated content means forum threads, reviews, videos and posts that users write themselves. The order holds when each project is cut to its 50 most cited websites, with 14.6% of Google AI Overviews citations going to user-generated content (n=124 projects) against 4.2% of Perplexity’s (n=118).

The official sites in the data come from the study’s main market and the EU. Four of the ten most cited websites in the whole study are public bodies: the Lithuanian government portal lrv.lt, the central bank lb.lt, the tax authority vmi.lt and the European Union’s europa.eu.

Why engines choose different websites

Each engine builds its own pool of candidate pages before it writes an answer, so each one cites different websites. ChatGPT runs its own retrieval index and also pulls results from at least eight other providers, Tomek Rudzki of Peec AI reported on 4 September 2026 from ChatGPT’s server-side events. The mechanics are set out in where ChatGPT gets its information when it searches the web.

Google AI Overviews draws on Google Search, and Google’s guide to AI features says a page must be indexed to appear there as a supporting link. AI Overviews and AI Mode may also use “query fan-out”, several related searches on subtopics and other data sources for one response, a process explained in how Google AI Overviews choose the pages they show.

Perplexity crawls the web with its own bot. Perplexity’s crawler documentation says PerplexityBot finds and links websites for Perplexity’s search results and is not used to collect content for AI foundation models. Three pools of pages pass through three selection steps, and the lists split there.

Within one engine, what the search step returns counts for more than how a page is built. In a paper published on 10 October 2026, University of Toronto researchers Mahe Chen and Nick Koudas found that every page ChatGPT cited was one its own search had returned. Retrieval explained 32% of the variation between pages of the same website, against 2.8% for page properties, as covered in what drives ChatGPT’s citations.

What it means

Marketing leads should plan each AI engine as a separate channel with its own list of cited websites. That list comes from a fixed set of buyer questions run on each engine, the same method brands use to measure how often AI answers mention them. The steps, in order:

1. Put your buyers’ questions to ChatGPT, Google AI Overviews and Perplexity without your brand name, and record every website each engine cites, one list per engine.
2. Mark the websites that appear on two or three lists, since one mention there reaches more than one engine.
3. For Google AI Overviews, publish short YouTube videos that each answer one buyer question, and keep your profiles on the forums and review sites it cites up to date.
4. For ChatGPT, answer the Reddit threads on its list under your brand’s own name, and get listed in the directories and official registers it cites for your category.
5. For Perplexity, pitch the news and media sites on its list with something they can report.
6. Run the same questions every month, compare each engine’s list with the last one, and count how many websites on each list now name your brand.

FAQ

How often do the cited websites change?

Often, from one day to the next. Ahrefs followed about 37,500 citations across five AI tools for 20 days, in a study published on 29 October 2026. Between 50 and 83% were gone the next day, and about 75% of those came back within the 20 days. The findings on how stable AI citations are show why a single check gives a partial list.

Is Reddit as large in US ChatGPT answers?

Reddit’s share is larger in US data. Ahrefs’ September 2026 list of the websites ChatGPT cites most gives reddit.com 16.8% of citations among its top sources, on US questions across all topics. In the 2026 source study, run mostly from Lithuania on buyer questions, Reddit took 3.0% of all ChatGPT citations. Ahrefs counts shares of its top sources only, so the two figures sit on different bases.

Does Gemini cite the same websites?

The 2026 source study does not answer that, because Gemini ran in only 16 of its 138 projects and stayed out of the engine comparisons. Peec AI’s US analysis of 30 million sources listed Reddit, YouTube, Wikipedia, Medium and PCMag or Forbes as Gemini’s five most cited. A brand that wants Gemini’s answers needs a separate list for that engine, built the same way.