Across 2,126 AI answers about AEO and AI search, collected in November 2026, the engines rarely agreed on sources. Google AI Overviews cited YouTube in 66% of answers. ChatGPT leaned on official documentation. Claude, Gemini and Perplexity mostly cited vendor and agency blogs. Search Engine Land was the only domain every engine cited.aeo.soThe direct answer, written the way answer engines quote.
AI answers analyzed
2,126
Citations counted
4,356
Different domains cited
217
AI engines and databases
6
What are the key findings?
AI engines rarely agree on sources. Measured as the overlap between all the domains each one cited, any two engines shared between 3% and 17% of their sources. Ranking in one engine says little about the others.
Google AI Overviews favors video and community sites. YouTube appeared in 66% of 1,910 tracked answers across 10 AEO and AI search topics, and Reddit in 25%. Nearly half of Google's citations went to video and community platforms.
ChatGPT trusts official documentation. Official sources such as Google Search Central and Google's help center made up 59% of ChatGPT's citations in the citation database and 40% in our panel.
Claude, Gemini and Perplexity cite vendor blogs. Brands, tool makers and agencies made up 92% of Claude's citations and 76% of both Perplexity's and Gemini's.
One publisher reached every engine. Search Engine Land was the only domain cited by all four engines in our panel and by both citation databases. Perplexity cited it in 7 of 10 answers.
Engines often answer from memory. ChatGPT and Claude both answered "AEO vs SEO" and "What is GEO?" without searching the web, even with search turned on. For definitions, what a model already knows about a topic beats this week's blog post.
Google AI Overviews: share of 1,910 answers citing each domain (DataForSEO's tracked answers).
Brands, vendors & agencies
Trade press & news
Video & community
Official docs & reference
Courses & education
youtube.com66% · 1,252
reddit.com25% · 481
support.google.com16% · 309
linkedin.com10% · 182
semrush.com9% · 181
zapier.com8% · 152
developers.google.com7% · 141
blog.google6% · 120
pcmag.com6% · 108
search.google5% · 96
google.com
ahrefs.com
technologyadvice.com
perplexity.ai
arxiv.org
bing.com
openai.com
rankite.com
citecycle.com
measure.co
andersoncollaborative.com
dageno.ai
nytech.in
searchenginejournal.com
llmrefs.com
frase.io
hubspot.com
authoritytech.io
searchengineland.com
rosssimmonds.com
neilpatel.com
cubitrek.com
typeface.ai
surfer.ai
cxl.com
medium.com
theseoagent.ai
mediasince.com
nikiclarkmarketing.com
conductor.com
salesforce.com
coursera.org
optimizely.com
seo.com
wix.com
Percent of answers citing the domain at least once · number of answers.
Switch engines to watch the sources re-rank. Database answers: DataForSEO LLM Mentions, US, English. Panel answers: 10 AEO questions asked to each engine with web search on, November 16, 2026.
Do AI engines agree on which sources to cite?
No. The heatmap below shows the 15 domains cited most widely, and how often each engine cited them. Read across a row: almost every domain is strong in one or two engines and absent from the rest.
YouTube is the clearest case. It appeared in 66% of Google AI Overview answers and 20% of Gemini's, yet ChatGPT, Claude and Perplexity never cited it in our panel. Search Engine Land shows the opposite pattern: modest in Google, cited by every engine. Semrush and Google's developer documentation each reached five of the six sources.
Share of each engine's answers citing the domain at least once. Darker means more often; a dash means never cited.
What kinds of sites does each engine cite?
We sorted every cited domain into five types: brands and vendors (including agencies and tool makers), trade press, video and community platforms, official documentation and reference sources, and courses. The split explains most of the disagreement above.
Google AI Overviews spreads its trust: video and community sites took 49% of its citations, vendors 23% and official documentation 20%.
ChatGPT is the most conservative. Official documentation took 59% of its database citations and 40% in our panel, with Google Search Central and Google's help center on top.
Claude drew 92% of its citations from vendor and agency blogs, spread across 47 different domains in eight answers. It cited the widest range of small sites.
Perplexity returned ten sources for every answer and leaned on large marketing brands: HubSpot appeared in 6 of 10 answers and Semrush in 4.
Share of each engine’s domain citations, by type of site. Hover a type to compare it across engines.
Which pages win the "answer engine optimization" question?
In Google AI Overviews, one page dominates. HubSpot's AEO guide was cited in 18 of the 35 tracked answers to questions about answer engine optimization. Marcel Digital's AEO service page followed with 8, then Profound's definition article and G2's AEO software category with 7 each.
The topic matters as much as the engine. Here is the most cited domain for each topic in Google AI Overviews:
Topic
Tracked answers
Most cited domain
Answers citing it
AI search
1,208
youtube.com
780
AI SEO
326
youtube.com
222
Schema markup
130
youtube.com
81
AI visibility
88
youtube.com
57
Generative engine optimization
54
youtube.com
43
Answer engine optimization
35
youtube.com
22
ChatGPT SEO
27
youtube.com
22
AI Overviews
23
youtube.com
15
LLM SEO
19
llmrefs.com
11
On schema markup, Schema App (67 answers) and Google's developer documentation (65) sat close behind YouTube. On LLM SEO, a niche tool maker, LLMrefs, led every larger site.
How do the engines answer the same question?
Pick a question below to see which domains each engine cited for it. Domains cited by more than one engine are highlighted; hover one to find it in every engine's list.
Ask the engines: one question, four sets of sources
Blue chips were cited by more than one engine for the same question (×n = how many). Hover one to find it everywhere.
Live answers from ChatGPT (gpt-5.5), Claude (claude-sonnet-5-5), Gemini (gemini-3.5-flash), Perplexity (sonar-pro), November 16, 2026.
Two patterns stand out. First, ChatGPT cited the fewest sources: 28 links across 10 answers, against 100 for Perplexity. Second, three of the four engines skipped the web for some questions. ChatGPT and Claude answered the two definition questions from memory, and Gemini's first answers to "How do I get my website cited by ChatGPT?" and the writing best-practices question cited nothing.
What should you do with this data?
Pick the engine that matters most to you, then work on its favorite sources. For Google AI Overviews, that means video and community: a clear YouTube answer and real Reddit participation. See entity SEO.
For ChatGPT, align with official documentation. Cite Google Search Central and platform help pages, and make your own pages the primary source on your product. See how ChatGPT chooses sources.
For Perplexity and Claude, earn coverage on established marketing sites. Guest research, data and quotes on Search Engine Land, HubSpot or Semrush travel further than another post on your own blog. See Perplexity SEO.
Get into what models already know. Definitions often come from a model's training, not a live search. Consistent, widely repeated facts about your brand and topic are what a model learns. See AEO examples.
Measure each engine separately. One engine's results do not predict another's. See how to measure AEO.
What changed in AI search in 2026, heading into 2027?
When
What changed
2026-02
Bing Webmaster Tools launched its AI Performance report in public preview, showing citations and grounding queries
2026-05-15
Google published its guide to optimizing for generative AI features, saying llms.txt, special schema and content chunking are not needed
2026-06
Google announced a Search Console control to exclude a site from AI Overviews and AI Mode, plus AI performance reports
2026-06
Bing added intent, topic and citation share to its AI Performance report
2026-08-31
Google completed the worldwide rollout of the Search Console AI control and reports
The direction is clear: AI search is becoming measurable and controllable for site owners, and the major engines say the same basics apply. Crawlable pages, clear answers and trusted sources win.
How was the data collected?
Citation databases (2,086 answers). DataForSEO's LLM Mentions database records AI answers and the sources they cite. We pulled Google AI Overviews and ChatGPT answers in the United States, in English, to tracked questions containing any of 10 topic phrases: answer engine optimization, generative engine optimization, AI SEO, LLM SEO, AI Overviews, AI search, llms.txt, schema markup, ChatGPT SEO and AI visibility. For each topic we recorded the total answers and the 10 most cited domains, counted once per answer. Google returned 1,910 answers and ChatGPT 176.
Question panel (40 answers). We asked the same 10 AEO questions to ChatGPT (gpt-5.5), Claude (claude-sonnet-5-5), Gemini (gemini-3.5-flash) and Perplexity (sonar-pro) through DataForSEO's AI Optimization API, with web search turned on, and recorded every link each answer cited. The first answer to each question counts for every engine. The questions are listed in the explorer above.
Source types. Each domain was assigned one of five types by hand. Unlisted brand, tool and agency sites default to "brands, vendors and agencies".
Date. Pulled 16 November 2026.
Limits:
Topic counts can overlap: an answer whose question contains two topic phrases counts in both.
Database figures cover each topic's 10 most cited domains only, which under-counts smaller sources.
ChatGPT's database sample is small (176 answers), and four topics returned no ChatGPT answers at all. Google returned none for llms.txt.
Perplexity lists every page it retrieved as a source and always shows ten. Compare engines by domains, not by link counts.
Gemini's grounding links name only the site; its citations are counted by domain.
Two other Gemini models we tried (gemini-4-argon and gemini-3.1-pro-preview) did not search at all for the first question.
AI answers change often. Treat this as a snapshot, not a ranking.
Cite or reuse this data
Free to use under CC BY 4.0. Link to this page as the source and include the date the data was pulled.
Paul Andre de Vera, "State of AEO 2027: Which Sources AI Engines Cite," aeo.so, data pulled November 16, 2026. https://aeo.so/research/state-of-aeo
Which sources do AI engines cite most for AEO questions?
Each engine has its own favorites. In aeo.so's November 2026 data, Google AI Overviews cited YouTube in 66% of 1,910 answers and Reddit in 25%. ChatGPT cited Google's own documentation most. Perplexity cited Search Engine Land in 7 of 10 answers, and Gemini cited Semrush and Reddit in 4 of 10 each.
Do ChatGPT, Claude, Gemini and Perplexity cite the same sources?
Rarely. Measured as the overlap of all domains each engine cited, any two engines shared between 3% and 17% of their sources. Search Engine Land was the only domain cited by all four engines in the question panel and by both citation databases.
How was this data collected?
Two ways. DataForSEO's LLM Mentions database supplied 2,086 tracked Google AI Overview and ChatGPT answers for 10 topic phrases. aeo.so also asked 10 AEO questions to ChatGPT, Claude, Gemini and Perplexity with web search on, through DataForSEO's API, on 16 November 2026, and recorded every cited link.
Can I use this data?
Yes. The data is free under CC BY 4.0. Quote the figures with a link to this page and the date the data was pulled. The full dataset is available as CSV and JSON. AI answers change often; aeo.so plans to refresh this report and will note earlier versions on the page.