Which Publications AI Search Engines Cite Most in 2026, Led by Forbes in 3.8 Percent of Answers

The publications AI search engines cite most are Forbes, TechRadar, Yahoo News, Healthline and NerdWallet. In the Machine Relations Index release of October 20, 2026, Forbes is cited in 3.82% of 28,201 monitored answer runs across six engines, first of 1,039 editorial publications, while Reddit, YouTube and LinkedIn are each cited more often than any publication.

Which publication an engine cites depends on the question. Forbes leads five of the six buyer question shapes and reaches 8.13% of runs on best-tools questions; TechRadar leads problem-first research; TechCrunch is cited almost only on news questions; and in 18 of the 20 measured categories the most-cited publication is not Forbes but a specialist such as NerdWallet, Healthline, TechTarget or MarTech.

The October 20, 2026 Machine Relations Index release, observation window May 10 to October 20, 2026 (157 days), 1,227 monitored prompts, six engines: ChatGPT, Claude, Gemini, Google AI Mode, Google AI Overviews and Perplexity. Rebuilt October 20, 2026 on the full release. Live standings: the Machine Relations Index.

The 15 publications AI search engines cite most, October 20, 2026 #

Rank is by cited runs among the domains the release classes as editorial publications. A run is one answer from one engine to one monitored prompt, and the rate is cited runs divided by all 28,201 runs. The engine column counts the engines observed citing the publication at least once in the window.

Seven publications clear 1% of runs. Five carry the Index's top confidence grade: Forbes, TechRadar, Yahoo News, NerdWallet and TechTarget. Another 25 carry grade B and 73 grade C. The other 936 of the 1,039 are still collecting evidence, cited too rarely or on too few days to grade, so most of any list of publications AI engines cite is a list of thin observations.

Platforms are cited more often than every publication #

Across all 32,242 cited domains, Forbes is the fourth most-cited source. Reddit (8.81% of runs), YouTube (6.89%) and LinkedIn (5.37%) are cited more often, and Medium (3.49%), which the release classes as a community platform, sits between Forbes and TechRadar. arXiv (2.67%), G2 (2.41%) and NIH (2.29%) are each cited more often than every publication except Forbes and TechRadar, and Gartner (1.91%) more often than every publication but the top three.

By source class, editorial publications account for 16,386 of the 181,052 times a domain was cited in a run, 9.1%. That share is a floor. In this release 27,839 of the 32,242 cited domains (86%) carry no identity evidence yet and publish as Other observed source, holding 65.4% of those citations, and reading a domain's site can only move it out of that bucket into a named class. The Index logs each reclassification on its correction record. What the release does settle is the order inside the confirmed editorial class, which is the ranking above.

What kind of publication gets cited depends on the question shape #

The Index files every monitored prompt under a question shape. Shortlist questions concentrate citations on a few general publications; research and news questions spread them across specialists.

• Shortlist questions favor general business and technology publications. On best-tools questions Forbes is cited in 8.13% of runs and TechRadar in 5.62%; on top lists Forbes reaches 6.25%. Forbes draws 507 of its 1,076 cited runs from those two shapes.
• Evaluation questions add trade and reference titles. On is-it-worth-it questions TechTarget (2.27%) sits beside Forbes and TechRadar. The New York Times is cited almost entirely on buyer questions: 292 of its 294 cited runs.
• Problem-first research pulls in product-review and health reference publications. TechRadar leads at 2.15%, then Tom's Guide (1.53%) and Healthline (1.50%), while Forbes falls to 1.06%.
• News questions reward news desks. TechCrunch is cited on 160 of its 172 runs on news and current-events questions, 2.87% of 5,581 news runs and second only to Forbes at 3.01%. It is cited on 12 buyer-question runs and on no best-tools, top-list or comparison question in the window.
• A platform is the most-cited source on every shape: Reddit on five of the six buyer shapes, YouTube on comparisons (13.00%) and LinkedIn on news (9.46%).

Each category has its own lead publication #

Forbes, TechRadar and Healthline reach the top of the overall list in different ways. Forbes is cited in 22 of the Index's 25 categories and TechRadar in 20, while Healthline's 491 runs come from four. Inside a category the leader is usually a specialist, and Forbes leads only 2 of the 20 categories with published segments.

A publication's category rate is its cited runs in that category's published segments divided by the runs observed in them. Professional Services, Sales and GTM, VC and Private Equity and Physical Consumer have no published segment yet, and the legacy news-topic bucket is left out of this table. Each category page on the Index carries the full leaderboard by question shape.

B2B and consumer questions draw different publications #

On B2B questions, 14,146 runs in eleven business categories, TechRadar leads at 2.49%, then Forbes (2.07%) and TechTarget (1.76%). On consumer questions, 8,093 runs in seven categories, Forbes leads at 7.77%, then Healthline (5.93%) and NerdWallet (4.63%). Publications built for one buyer barely register with the other: TechTarget is cited in 0.04% of consumer runs, and Healthline in none of the B2B runs.

Reddit (7.55%), LinkedIn (7.25%), YouTube (5.80%), arXiv (5.10%) and Medium (4.72%) are each cited on more B2B runs than any publication. The B2B cut covers AI infrastructure, AI security and privacy, AI visibility and GEO, cybersecurity, deep tech and hardware, enterprise software, fintech, HR and talent, industrial, logistics and freight, and martech and advertising. The consumer cut covers consumer finance, consumer health, consumer products, family software, emergent prosumer, iGaming and betting, and local services. Healthcare services, education and training and the legacy news bucket sit in neither cut.

Which engines cite which publications #

Engines do not share one publication list. Of the 1,039 publications in the release, 49 are cited by all six engines and 536 by only one. Among the 25 most cited, 11 are cited by all six, and each of the rest is missing from at least one engine's answers.

• The New York Times (294 runs) is observed only in Google AI Mode, Google AI Overviews and Perplexity answers, not in ChatGPT, Claude or Gemini answers.
• Tom's Guide (237 runs) is observed only in ChatGPT, Claude and Perplexity answers.
• PCMag and CNET are cited by Gemini, both Google surfaces and Perplexity, and not by ChatGPT or Claude.
• Yahoo News, NerdWallet and TechCrunch are cited by five engines, every one but Gemini.
• The wire services barely register: Reuters is cited in 23 runs by three engines and AP News in 12 runs by ChatGPT alone, both still collecting evidence.

Two limits govern this section. The release records which engines cite each domain, not a citation rate per engine, so it can show that an engine was not observed citing a publication but not how often each engine cites it. And engine coverage is uneven: Gemini produced a citation on 2,153 of 6,384 returned runs, and Google AI Mode and Google AI Overviews answered on 115 and 98 of the 157 days, so a publication missing one of those engines may be missing a quiet engine rather than a rejection. The release manifest carries each engine's day and run counts.

Why rankings of AI-cited sources disagree #

Lists that put Wikipedia, Reuters or The New York Times first are usually measuring a different population, not doing different arithmetic.

• Every domain, or publications only. A ranking of every cited domain is led by Reddit, YouTube and LinkedIn in this release too. Wikipedia is cited in 1.00% of runs (282) and is filed under academic and government sources, not publications.
• News questions, or buyer questions. Studies of news citations rank wire services and national papers highly because they sample news questions. On the six buyer shapes here Reuters and AP News are barely cited.
• One engine, or six. A list built on one engine inherits that engine's gaps, such as The New York Times's absence from ChatGPT and Claude answers in this window.
• A synthesis, or one denominator. A ranking assembled from studies with different prompts, windows and units cannot state a rate. This one gives every publication a rate on the same 28,201 runs, recomputable from the public file.

Why engines select the sources they do, from retrieval to the cited passage, is covered in how answer engines select sources. Week-over-week movement is in the weekly Index report.

Methodology: how the Index measures it, and what it does not #

• Corpus. 28,201 answer runs from 1,227 monitored prompts across six engines, May 10 to October 20, 2026, with 205,362 citation events and 32,242 cited domains. Google AI Mode and Google AI Overviews count as two engines because Google ships them as separate products.
• Rate. Cited runs divided by observed runs. Shape, category, B2B and consumer rates sum runs_cited and runs_observed over published segments. A segment publishes once it clears the evidence floor of 10 observed runs on 7 distinct dates; 123 of 151 collectable segments are published and 28 are still collecting.
• Editorial class. The release's source_role field, value editorial_media: 1,039 domains. Medium, Reddit and LinkedIn are filed as community platforms, and PR Newswire and Business Wire as wire distribution.
• Confidence. Each domain is graded A, B, C or collecting by how much evidence stands behind it and how consistently it is cited across the window.
• Scope. The prompts are a fixed basket of buying and research questions, not a sample of all user traffic. The class share is a floor while domain identity coverage is incomplete. The release carries no per-engine rate per domain. A citation records that an engine selected a source in an answer; it is not an endorsement, and it says nothing about what caused the selection.

Method note. Every figure is read from the public release at machinerelations.ai/data/machine-relations-index.json, the October 20, 2026 Machine Relations Index release: domains[].methodology v2.overall for the overall ranking and engines, and domains[].methodology v2.strata[] summed by question_shape or category over published segments for the cuts. A zero on this page, such as TechCrunch on best-tools, top-list and comparison questions or Healthline on B2B questions, means no observed run in those segments cited the domain in this window, not that it never can be.

FAQ #

Which publication do AI search engines cite most? #

Forbes. In the Machine Relations Index release of October 20, 2026, Forbes is cited in 1,076 of 28,201 monitored answer runs (3.82%) by all six engines, first of 1,039 editorial publications, with the top confidence grade. TechRadar (2.81%) and Yahoo News (2.00%) follow.

Which publications get cited by AI search engines? #

Seven publications are cited in 1% or more of monitored runs: Forbes (3.82%), TechRadar (2.81%), Yahoo News (2.00%), Healthline (1.74%), NerdWallet (1.51%), TechTarget (1.30%) and The New York Times (1.04%), in the October 20, 2026 release. Reddit (8.81%), YouTube (6.89%) and LinkedIn (5.37%) are each cited more often than any of them. 936 of the 1,039 publications in the release are cited too rarely to grade.

Do ChatGPT, Perplexity and Gemini cite the same publications? #

Partly. Forbes, TechRadar, Healthline, TechTarget, CNBC, People Managing People and WebMD are cited by all six engines. The New York Times is observed only in Google AI Mode, Google AI Overviews and Perplexity answers, and Tom's Guide only in ChatGPT, Claude and Perplexity answers. Of 1,039 publications, 536 are cited by one engine only.

Is TechCrunch cited by AI search engines? #

On news questions, yes: 160 of TechCrunch's 172 cited runs are news and current-events questions, where it is cited in 2.87% of runs, second to Forbes. On buyer questions it is cited in 12 runs and on no best-tools, top-list or comparison question in the October 20, 2026 release.

Which publications do AI engines cite on B2B questions? #

TechRadar (2.49% of 14,146 B2B runs), Forbes (2.07%) and TechTarget (1.76%), then People Managing People (1.33%). Reddit, LinkedIn, YouTube, arXiv and Medium are each cited on more B2B runs than any publication.

How often does this ranking change? #

The Index publishes a release daily, and every publication's profile and every category page update with it. This page reads the October 20, 2026 release. Ranks at the top move slowly; category leaders and thin publications move more often.

First published March 30, 2026 as a ranking of the publications AI engines cite; rebuilt October 13, 2026 on a B2B cut and October 20, 2026 on the full release, with the B2B cut kept as a section. Method: the Machine Relations Index.