Guide · AI search

Do a few websites win most AI Mode and Perplexity citations?

Yes: in every study we reviewed, a small group of websites collects a large share of AI search citations, while thousands of others are cited once or twice. How unequal it is depends on the engine and the industry. In one Swiss study, Google AI Mode was the most top-heavy engine and Perplexity the least, so targets should be set engine by engine.

The short version

  1. In a 2026 Swiss study of four engines over 45 to 46 days, AI Mode scored 0.782 on a 0-to-1 inequality scale and Perplexity 0.671, the lowest of the four.
  2. In a 2026 audit of four engines, the 25 most cited websites took 23.8% of all citations, while 59.1% of cited websites appeared only once.
  3. In ChatGPT’s answers to 100 health questions, 10 organizations took 52.8% of the 615 citations, led by Wikipedia, Mayo Clinic and Cleveland Clinic.
  4. For news, the top 20 outlets took 67.3% of OpenAI models’ news citations in 2025, against 31.9% for Google’s and 28.5% for Perplexity’s.
  5. In our own US data, AI Mode spread its citations across more websites than AI Overviews did, so the engine ranking is not settled.

How concentrated are AI search citations?

Very: in most studies a handful of websites takes a large share, followed by a long tail. The exact degree depends on what is measured.

The clearest test comes from Schulte and colleagues (opens in a new tab) at the University of St. Gallen; one author is also affiliated with a company that supplied the data. They asked ChatGPT, Gemini, Google AI Mode and Perplexity eight Swiss-German questions in each of four industries, daily from 24 January to 20 March 2026. They scored inequality on a scale where 0 means every website gets an equal share and 1 means one website gets them all.

The average across all engines and industries was 0.715. The authors describe this as “a highly unequal citation landscape in which a handful of domains captures most of the AI-generated visibility.” Reference sites such as Wikipedia are often in that handful; see Wikipedia’s outsized role in AI search.

Which engine concentrates citations the most?

In the Swiss study, Google AI Mode was the most concentrated and Perplexity the least. Other studies rank engines differently.

EngineInequality score (0 = equal, 1 = one site gets all)
Google AI Mode0.782
Gemini0.723
ChatGPT0.684
Perplexity0.671

Source: Schulte and colleagues, Swiss-German questions, January to March 2026.

Other measurements disagree on the order. Allaham and Diakopoulos (opens in a new tab) audited the web versions of four engines on politics, health and the environment. On the same scale, ChatGPT was the most concentrated at 0.648 and Copilot the least at 0.492. Our own comparison of AI Mode and AI Overviews, on 400 US searches in September 2026, found AI Mode spreading its citations more widely than AI Overviews: the equivalent of 93.5 equally cited websites against 58.8. Market, language, topic and method all change the answer.

Which websites tend to win?

The winners are usually large reference, institutional or platform sites, and they change by topic. There is no single list for every industry.

In health, Jacques and colleagues (opens in a new tab) put 100 consumer health questions to ChatGPT 5.2 Pro. Ten organizations took 52.8% of all citations: Wikipedia had 10.7%, Mayo Clinic 9.9% and Cleveland Clinic 9.8%. The same study asked whether cited pages need author bylines. In news, Yang (opens in a new tab) analyzed more than 366,000 citations from a public AI search comparison site in 2025. OpenAI models cited reuters.com in 22.8% of their news citations and apnews.com in 12.2%.

In commercial searches, video can lead. In our AI Overview study of 481 US AI Overviews, YouTube was cited in 64.2% of them and accounted for 15.7% of all 4,051 citations. Industry also matters within one market: in the Swiss study, telecommunications scored 0.750 and sporting goods 0.680.

Is there still room for smaller websites?

Yes: the long tail is large, even though the top is crowded. Most cited websites appear only rarely, but they do appear.

In the Allaham and Diakopoulos audit, the 25 most cited websites took 23.8% of citations and the other 76.2% were spread across a long tail. Within that tail, 59.1% of websites were cited only once. A 2026 study across twelve European languages (opens in a new tab) found a similar shape: 80% of citations came from 3,878 of 21,077 websites, or 18.4%.

Google’s AI Overviews show the same shape. Xu and colleagues (opens in a new tab) tracked 55,393 trending US searches over 40 days in 2026. The 10 most cited sites took 29.7% of AI Overview citations, while 56.2% of the sites cited appeared exactly once. The authors conclude that breadth, not concentration, is the main shape of AI Overview citations, and that the regular first page of results is more concentrated.

Breadth also differs by engine. In Yang’s news data, Perplexity models cited 1,430 different news sources, against 881 for Google’s and 707 for OpenAI’s. A long tail means a smaller site can be cited, but rarely for any single question.

Do the same few websites win every day?

No: even the leading sources move from run to run. Concentration describes the pattern over many answers, not any single answer.

In repeated runs of the same questions within 24 hours, the Swiss study found that only 32 to 43% of cited sources overlapped between runs, on average. Sielinski (opens in a new tab), who works for an AI visibility company, warns that a single run gives a misleadingly precise picture of how websites rank. In practice, a website’s share of citations is an average that needs many runs to estimate.

What should you do about it?

Find out which websites dominate your category on each engine, then earn a place on them. Concretely:

  1. List the most cited websites for your buyers’ questions on each engine you care about, from repeated runs, not one.
  2. Set separate targets per engine; a share that is good on Perplexity may be weak on AI Mode.
  3. Prioritize coverage on the sites that already dominate your category, such as major reference, review or institutional sites.
  4. Keep publishing your own useful pages; the long tail is real, but each page wins only occasionally.
  5. Recheck the list regularly, because the leaders move.

If you want help mapping the dominant sources in your category, see our generative engine optimization service.

What does the research not tell us yet?

The research shows that concentration is high, but not how stable or universal the ranking of engines is. The gaps:

  • The engine ranking conflicts between studies, and no study compares the same questions across several markets.
  • The Swiss study covers four industries in one market and language; the US picture may differ.
  • Health and news studies cover single topics and dates.
  • No study shows whether winning a top source makes an engine more likely to recommend your brand.
  • Several datasets come from monitoring companies, and two key papers have authors linked to them.

Frequently asked questions

Which AI search engine relies on the fewest websites?

In one Swiss study, Google AI Mode did: it scored 0.782 on a 0-to-1 inequality scale. Our US data found the opposite for AI Mode against AI Overviews, so check your own market.

Is Perplexity less concentrated than other AI engines?

In the Swiss study it was the least concentrated of four engines, at 0.671. In Yang’s 2025 news data it also cited the most different news sources, 1,430.

What share of AI citations go to the top websites?

It varies by topic. In one health study, 10 organizations took 52.8% of ChatGPT’s citations; in a broader audit, the top 25 websites took 23.8%.

Can a small website still get cited by AI search?

Yes, but rarely for any one question. In one audit, 59.1% of cited websites appeared only once.

Sources

Free strategy call

Some questions are easier to answer about your own business.

Bring the one that matters most. On a free 30-minute call we’ll take a first look at it and send you a short written read afterward.