Beket AI beket.ai
Cover: Which publications does AI cite most? 10 unbranded US buying questions, 252 citations across ChatGPT, Perplexity and Google AI Overviews.

Which Publications Does AI Cite Most? A 2026 Study of ChatGPT, Perplexity and AI Overview Sources

By Beket AI team

Study date: November 15, 2026

Cover: Which publications does AI cite most? A Beket AI study of 10 unbranded US buying questions and 252 citations across ChatGPT, Perplexity and Google AI Overviews.

Ask three AI engines the same ten shopping questions and you get three different reading lists. Perplexity sends you to review publishers. ChatGPT sends you to the brands themselves almost as often as to the press. Google AI Overviews sends you to YouTube.

That’s the short version of a study we ran on November 15, 2026. We asked ChatGPT, Perplexity and Google AI Overviews ten unbranded “what’s the best…” questions as a US, English-language user, then logged every source each answer cited: 252 citation events in total. To check whether those patterns hold beyond ten questions, we also looked at the latest run of 166 audits stored in Beket AI, counting each audit once.

These are the answers the engines returned on one day. They aren’t our ranking of any publication, and AI answers change over time.

Key numbers

Study date November 15, 2026
Market United States, English
Questions 10 unbranded buying questions (full list below)
Engines ChatGPT, Perplexity, Google AI Overviews
Answers recorded 28 of 30 (AI Overviews answered 8 of 10)
Citation events 252 (Perplexity 117, ChatGPT 72, AI Overviews 63)
Distinct domains (host names) cited 124
Most cited domain rtings.com, 21 citations (8.3%)
Domains cited by all three engines 3 (rtings.com, techradar.com, wired.com)

The most cited domains

One domain stood well above the rest. RTINGS, the lab-testing review site, was cited 21 times, on all three engines and for five of the ten questions (robot vacuums, running shoes, college laptops, VPNs and mattresses). The next ten domains took between 5 and 13 citations each.

Stacked bar chart of the 11 most cited domains in a 10-question Beket AI study, by engine: rtings.com 21 (ChatGPT 4, Perplexity 16, AI Overviews 1), youtube.com 13 (all AI Overviews), pcmag.com 11, techradar.com 9, forbes.com 8, healthline.com 7, cnbc.com 6, google.com 6 (all AI Overviews), wired.com 5, nytimes.com 5 (all Wirecutter, all Perplexity), creditcards.chase.com 5.

Rank Domain Citations Share of 252 Type Engines that cited it
1 rtings.com 21 8.3% Publisher All three
2 youtube.com 13 5.2% User content AI Overviews only
3 pcmag.com 11 4.4% Publisher Perplexity, AI Overviews
4 techradar.com 9 3.6% Publisher All three
5 forbes.com 8 3.2% Publisher Perplexity, AI Overviews
6 healthline.com 7 2.8% Publisher ChatGPT, Perplexity
7 cnbc.com 6 2.4% Publisher ChatGPT, Perplexity
7 google.com 6 2.4% Search pages AI Overviews only
9 wired.com 5 2.0% Publisher All three
9 nytimes.com 5 2.0% Publisher Perplexity only
9 creditcards.chase.com 5 2.0% Brand Perplexity only

Those 11 domains account for 96 of the 252 citations (38.1%). The other 113 domains share the remaining 156, and 98 of the 124 domains were cited by only one engine.

Two details stand out. All five New York Times citations were Wirecutter review pages, so the “newspaper” in this list is really a product-testing desk. And the only brand in the top 11 is a Chase credit card site, cited five times by Perplexity for the travel credit card question. Brands still got plenty of citations overall (79 of 252), but they were spread across 57 vendor sites, most with one or two links each.

Across all 252 citations, publishers took 138 (54.8%), brand and vendor sites 79 (31.3%), user content such as YouTube and Reddit 20 (7.9%), and other pages 15 (6.0%).

Engine by engine

The overall table hides how differently the three engines behave.

100% stacked bar chart of each engine's citations by source type. ChatGPT: publishers 42%, brand and vendor sites 56%. Perplexity: publishers 79%, brand sites 21%. Google AI Overviews: publishers 25%, brand sites 24%, user content 29%, other 22%.

Engine Citations Publishers Brand sites User content Other Top domains
Perplexity 117 92 (78.6%) 24 (20.5%) 1 0 rtings.com 16, pcmag.com 9, forbes.com 6
ChatGPT 72 30 (41.7%) 40 (55.6%) 1 1 rtings.com, runningwarehouse.com and techradar.com, 4 each
Google AI Overviews 63 16 (25.4%) 15 (23.8%) 18 (28.6%) 14 (22.2%) youtube.com 13, google.com 6, reddit.com 4

Perplexity cited the most sources (about 12 per answer) and most of them were editorial. It is the engine that behaves most like a reader of review publications: RTINGS, PCMag, Forbes, CNBC, Healthline and Wirecutter carried most of its answers.

ChatGPT cited fewer sources (about 7 per answer) and was the only engine where brand sites outnumbered publishers. For the CRM and project management questions it linked straight to vendors such as HubSpot, Pipedrive, Zoho, Asana, Monday.com, Trello and Notion. No single domain got more than 4 of its 72 citations, so its sources were the most spread out.

Google AI Overviews answered 8 of the 10 questions; it returned no answer for the travel credit card and college laptop questions, so those two are missing rather than counted as zero. Where it did answer, more than a fifth of its citations went to YouTube, and it was the only engine that cited YouTube, Reddit or Google’s own search and Shopping pages in any volume. That matches public data: Searchable’s study of 3.2 million cited pages (Oct 12 to 25, 2026) found 8.6% of AI Overviews sources were posted by individual users, against about 1.1% for ChatGPT and Perplexity.

Only rtings.com, techradar.com and wired.com were cited by all three engines. Another 23 domains were cited by exactly two.

Who led each question

The winner changed with the topic. Review labs led the gadget questions, finance media led credit cards, health publishers led therapy, and brands led meal kits.

Question Most cited domains (citations) Citations, all engines
Best CRM for a small business pcmag.com 3, youtube.com 3 25
Best project management software asana.com 4, youtube.com 4 26
Best running shoes for beginners rtings.com 5, runningwarehouse.com 4, google.com 4 30
Best robot vacuum for pet hair rtings.com 10, techradar.com 3 28
Best travel credit card cnbc.com 5, creditcards.chase.com 5, nerdwallet.com 4 25
Best VPN cnet.com 3 24
Best meal kit delivery service everyplate.com, greenchef.com, healthline.com, hellofresh.com, 2 each 29
Best laptop for college students rtings.com 3 18
Best mattress for back pain ncoa.org 3 23
Best online therapy service healthline.com 5 24

The college laptop and travel credit card totals are lower partly because AI Overviews didn’t answer them.

What 166 audits add

Ten consumer questions are a narrow window, so we also looked at citation data already stored in Beket AI: audits run between June 11 and November 15, 2026 across eight engines. Raw counts would be misleading here. Twenty-nine audits re-run every day, and together they produced 67.7% of all stored citation events. So we kept only the latest run of each audit, dropped demo audits and the sites of the brands being tracked, and ranked domains by how many of the 166 audits cited them.

Bar chart of the share of 166 Beket AI audits whose latest run cited each domain: google.com 75%, reddit.com 68%, linkedin.com 65%, youtube.com 60%, facebook.com 48%, instagram.com 37%, clutch.co 28%, g2.com 26%, forbes.com 25%, yahoo.com 24%.

The picture is very different from the consumer study. Google pages were cited in 75% of audits, Reddit in 68%, LinkedIn in 65% and YouTube in 60%. Forbes was the only publisher in the top 10, at 25%. Review and directory sites such as Clutch and G2 ranked higher than any newsroom.

The reason is the mix of questions. Most audits in Beket AI are for software and professional-services businesses, where buyers ask about agencies, vendors and local providers. Those categories have few lab-style review publications, and in this data the engines leaned on platforms, directories and vendor pages instead. Publishers were only 9.9% of latest-run citation events in that data, against 54.8% in our consumer study.

Counting each audit once also changed the ranking. On raw event counts, Semrush, HubSpot, Microsoft and OpenAI were all in the top 10, because a few daily audits cited them on every run. Counted once per audit, all four fell out of it, and Facebook, Instagram, Forbes and Yahoo moved in. If you track your own citations, check whether a few repeated prompts are driving your totals in the same way.

What the public research says

Our numbers line up with larger public datasets in some ways and differ in others.

  • Muck Rack’s October 2026 “What Is AI Reading?” release reports that earned media still drives more than 84% of AI citations. It also found journalism was 22% of AI citations in September, down from 27% in May, and only 212 domains were cited by all four major AI providers.
  • The same release found ChatGPT repeated just 41% of its cited domains when the same prompt ran twice in a day, against 94% for Perplexity.
  • First Page Digital’s Singapore tracking study (164,540 answers, Oct 9 to Nov 7, 2026) found ChatGPT cited only about 50% of the previous day’s sites for the same question, Google AI Overviews about 40% and Perplexity about 88%.

That last point matters for reading any single-day study, ours included. Perplexity’s sources are fairly stable from day to day, so its preference for review publishers is probably a durable pattern. ChatGPT’s and AI Overviews’ lists shift much more, so treat their exact top domains as a snapshot.

What this means if you want to be cited

Find out which publications your category runs on. For gadgets it was RTINGS and PCMag; for credit cards, CNBC and NerdWallet; for therapy, Healthline. Run your own buyer questions and roll the cited URLs up to outlets, as in our guide on building a media list from AI citations.

Pitch differently for each engine. Perplexity favored editorial reviews. ChatGPT linked to vendor pages almost as often, so your own comparison and pricing pages matter there. Our guide on how to get cited by ChatGPT covers the on-site side.

Don’t ignore video and community content for Google. In our study, AI Overviews was the engine most likely to cite YouTube and Reddit. A useful product video can count there in a way it won’t on Perplexity.

For B2B and local services, check the platforms first. In the 166-audit data, LinkedIn, Reddit, Clutch and G2 showed up far more often than any newsroom. We wrote more on this in your AI citations are coming from somewhere you aren’t looking, and our digital PR for AI search guide covers how to earn coverage on the sources that matter.

Method and caveats

  • Audit: one Beket AI market audit, run Sunday, November 15, 2026, 4:39 to 4:48 PM Pacific Time, targeting the United States in English. Engines: ChatGPT, Perplexity and Google AI Overviews.
  • Questions: 10 unbranded buying questions across ten consumer and software categories, chosen for real buyer intent. Each was asked once per engine.
  • Coverage: 28 of 30 answers were generated. Google AI Overviews returned no answer for the travel credit card and college laptop questions; those are listed as missing, not counted as zero.
  • Counting: a citation event is one cited URL in one answer. Shares are of the 252 events. Domains are counted by host name, so creditcards.chase.com and chase.com are counted separately.
  • Source types: every cited domain was labeled by hand as publisher, brand or vendor, user content (YouTube, Reddit, forums, review platforms) or other (Google pages and small sites). Some calls are judgment calls.
  • 166-audit data: aggregate figures only, from audits stored in Beket AI between June 11 and November 15, 2026. We used the latest run of each audit, excluded demo audits and tracked brand sites, and ranked by share of audits. No customer, brand or prompt details are included. That data is mostly software and professional-services audits, so it isn’t a sample of all searches.

This is the data the engines returned in a single run, and AI answers change over time. If you’d like this study run again, or run for your own category, you can ask Beket AI for another run.

All 10 questions

  1. What’s the best CRM for a small business?
  2. What’s the best project management software?
  3. What are the best running shoes for beginners?
  4. What’s the best robot vacuum for pet hair?
  5. What’s the best travel credit card?
  6. What’s the best VPN?
  7. What’s the best meal kit delivery service?
  8. What’s the best laptop for college students?
  9. What’s the best mattress for back pain?
  10. What’s the best online therapy service?

Bottom line

There isn’t one list of “the publications AI cites.” In our ten-question study, review labs like RTINGS dominated when Perplexity answered, ChatGPT split its links between the press and the brands, and Google AI Overviews leaned on YouTube. In the wider Beket AI data, platforms like LinkedIn, Reddit and Clutch mattered more than any newsroom. The useful question is which sources each engine trusts for your category.

Beket AI’s free audit runs 25 of your buyer questions across three engines and shows which sources each one cites, so you can see your own list.

Related posts