What ChatGPT cites when people ask about AEO and GEO (November 2026)
In 844 ChatGPT answers about AEO and GEO, we counted 5,613 citations across 810 domains. Homepages account for 30% of citations, and only 34% of domains in reported search results are cited in the same answer. This snapshot covers our own market, six countries and three languages. GeckoBoost was named in none.
David Dao, Founder
Leads the product and the client work at Altis Koten SRL, the Belgian company that publishes this site.
Published Updated
Key findings
- ChatGPT cites a mean of 6.65 sources per answer (median 6). Only 20 answers, or 2.4%, cite nothing.
- The 10 most cited domains account for 27% of all citations: 1,521 of 5,613.
- Homepages receive 1,711 citations, or 30% of the total. Their share is 36% in English, 25% in French and 23% in Dutch.
- Among citations from the 150 domains we classified, agencies and consultants account for 33.6%, software vendors for 32.1% and search engines and AI platforms for 20.8%.
- Only 34% of domains in an answer's reported search results are cited. Conversely, 31% of cited domains are absent from that answer's reported search results.
- Of 51 tracked brands, at least one appears in 542 answers (64%). GeckoBoost is named in 0 of 844 answers.
What questions and markets does this corpus cover?
These are our own six tracking projects about the AEO and GEO market: Belgium, France, Luxembourg, the Netherlands, the United Kingdom and the United States. They contain public answers about public sites, with no client data. We collected 844 answers to 50 distinct buyer questions, grouped into 95 question-country pairs. Some questions were asked in several countries.
- By language, with question-country pairs in parentheses: English 348 answers (39), French 316 (36), Dutch 180 (20).
- By country, with question-country pairs in parentheses: Belgium 243 answers (27), United States 153 (17), United Kingdom 144 (16), France 126 (14), Netherlands 90 (10), Luxembourg 88 (11).
The questions include “What are the best alternatives to Profound for AI visibility tracking?”, “How do I get my brand cited by ChatGPT?”, “AEO agency Belgium” and “Is there an AEO tool with an MCP server I can connect to my AI assistant?”. We also ask “Combien coûte un audit GEO ?”, “Agence GEO à Paris” and “Beste GEO-bureau in Nederland”. These are buyer questions written by us, not a sample of all ChatGPT usage.
Which domains does ChatGPT cite most?
The interface shows 5,613 citations from 810 distinct domains. An answer has a mean of 6.65 citations (median 6) and 5.8 distinct cited domains. Twenty answers (2.4%) cite nothing. The top 10 domains hold 1,521 citations, or 27% of the total. A citation count and a count of answers citing a domain are different measures: one answer can cite a domain more than once.
| Domain | Type | Citations | Answers citing it, of 844 | Homepage share of its citations |
|---|---|---|---|---|
| developers.google.com | Platform documentation | 341 | 149 | 0% |
| semrush.com | Software vendor | 199 | 163 | 62% |
| peec.ai | Software vendor | 158 | 137 | 84% |
| otterly.ai | Software vendor | 145 | 138 | 94% |
| ahrefs.com | Software vendor | 132 | 126 | 20% |
| blog.hubspot.com | Media and blogs | 125 | 102 | 0% |
| help.openai.com | Platform documentation | 124 | 89 | 0% |
| tryprofound.com | Software vendor | 122 | 108 | 87% |
| flyeo.be | Agency | 89 | 68 | 20% |
| developers.openai.com | Platform documentation | 86 | 86 | 0% |
| arxiv.org | Research | 80 | 52 | 0% |
| agences-generative-engine-optimization.fr | Ranking or directory | 63 | 41 | 54% |
| firstpagesage.com | Agency | 55 | 35 | 55% |
| geoagencyindex.com | Ranking or directory | 55 | 37 | 0% |
| depends.be | Agency | 52 | 42 | 12% |
The most cited domains differ by language
- French: developers.google.com 125 citations, agences-generative-engine-optimization.fr 63, semrush.com 60, peec.ai 59, help.openai.com 52, tryprofound.com 50.
- Dutch: developers.google.com 75 citations, goldlemon.nl 33, depends.be 32, flyeo.be 31, ahrefs.com 31, searchlab.nl 30.
- English: developers.google.com 141 citations, semrush.com 113, otterly.ai 81, peec.ai 76, blog.hubspot.com 64, ahrefs.com 59, firstpagesage.com 55.
How often are homepages cited?
A homepage here is a URL whose path is “/”, with no deeper page. Homepages receive 1,711 of 5,613 citations (30%). The language split is 1,028 of 2,829 citations in English (36%), 490 of 1,951 in French (25%) and 193 of 833 in Dutch (23%). These percentages describe the citations in each language, not the share of answers.
The pattern varies by domain: homepages account for 94% of otterly.ai citations, 87% of tryprofound.com citations and 84% of peec.ai citations. The three documentation domains in the table have no homepage citations. This is an observation about this set of questions, not evidence that replacing detailed pages with a homepage would improve visibility.
What kinds of sources are cited?
We classified the 150 most cited domains by hand, using what each site's own homepage says it is. They account for 4,044 citations, or 72% of the total. The shares below use only those classified citations as their denominator. The other 660 domains, accounting for 28% of citations, were not classified.
| Source type | Domains | Share of classified citations | Homepage share within this type |
|---|---|---|---|
| Agencies and consultants | 79 | 33.6% | 37% |
| Software vendors (AI visibility, SEO, marketing software) | 36 | 32.1% | 56% |
| Search engines and AI platforms (documentation, help centres, product pages) | 17 | 20.8% | 9% |
| Media and blogs | 7 | 5.3% | 0% |
| Rankings and directories of agencies or tools | 7 | 4.7% | 18% |
| Research (papers, code) | 4 | 3.5% | 0% |
Does ChatGPT cite what it reads?
“Read” means the web search results that ChatGPT's interface reports for an answer. It does not prove a page was fully read. The mean is 15.1 search results per answer (median 12), from 11.7 distinct domains. Fourteen answers have no reported search results.
Only 34% of the domains read are cited in the same answer: 3,395 of 9,884 answer-domain pairs. Conversely, in the 830 answers with reported search results, 31% of cited domains are absent from those results: 1,501 of 4,896 cited answer-domain pairs. These are per-answer comparisons, not a comparison of two whole-corpus domain lists. Absence from reported results does not establish how a citation was produced.
| Domain | Answers with domain in search results | Answers citing domain |
|---|---|---|
| otterly.ai | 13 | 138 |
| peec.ai | 20 | 137 |
| tryprofound.com | 19 | 108 |
| scrunch.com | 3 | 49 |
| semrush.com | 66 | 163 |
| ahrefs.com | 41 | 126 |
| developers.google.com | 149 | 149 |
| help.openai.com | 97 | 89 |
| developers.openai.com | 87 | 86 |
| marketercoffee.com | 73 | 28 |
| loudface.co | 68 | 17 |
| digitaltatva.com | 67 | 12 |
| aeogeolabs.com | 64 | 19 |
| answerenginesoptimization.ai | 49 | 0 |
The largest gap concerns software vendors, whose homepages are cited without the domain appearing in that answer's reported search results. Documentation is cited about as often as it appears in search results. Some agency sites appear in many more search results than citations. answerenginesoptimization.ai was in the results of 49 answers and cited in none.
Which brands does ChatGPT name?
We count mentions only among the 51 brands tracked in these projects. A brand outside that list is not counted. At least one tracked brand appears in 542 of 844 answers (64%). A brand mention is separate from a source citation: naming a company does not necessarily mean citing its website.
| Brand | Answers naming it | Share of 844 |
|---|---|---|
| Semrush | 197 | 23.3% |
| Ahrefs | 154 | 18.2% |
| Peec AI | 154 | 18.2% |
| Otterly.AI | 145 | 17.2% |
| Profound | 144 | 17.1% |
| HubSpot | 144 | 17.1% |
| Flyeo | 73 | 8.6% |
| Scrunch AI | 55 | 6.5% |
| Depends | 51 | 6.0% |
| Omniscient Digital | 47 | 5.6% |
| ClickForest | 46 | 5.5% |
| iPullRank | 40 | 4.7% |
| GeckoBoost | 0 | 0% |
GeckoBoost was named in 0 of 844 answers. We publish that result alongside the other brands. Our case study follows our own visibility, and our Profound alternatives comparison looks at one of the tool questions in this corpus.
How we collected and counted the answers
- We used ChatGPT with web search through DataForSEO's LLM Scraper for ChatGPT, which collects the ChatGPT web interface. The model reported by the interface was gpt-6.
- The answers are dated 10 to 21 November 2026, a 12-day window with 9 collection days. We collected one answer per question per day in each question-country pair, through the scraper's own sessions.
- We counted the citations shown under each answer and the search results reported by the interface. For the read-versus-cited comparison, each domain is counted once per answer.
- We identified homepages by the path “/”, classified the top 150 domains from their own homepages and counted brand mentions against the 51 brands we track.
What this study cannot establish
- One engine and one interface: ChatGPT as the scraper collects it. The figures do not describe every assistant, and a personalised ChatGPT session (signed in, with memory) may answer differently.
- Twelve days, with nine collection days and one answer per question per day. Answers vary from day to day.
- Questions written by us about our own market, in English, French and Dutch. Countries and languages have different numbers of answers and question-country pairs.
- Citations are what the interface shows. Search results are what it reports, not a complete account of how an answer was produced.
- The source classification is our judgement on the top 150 domains. The remaining 28% of citations are not classified.
- Brand counts cover only our 51 tracked brands. This is a snapshot, not a ranking factor study or a test of causation.
How to read these observations in practice
- Observation: tool questions suggest brand knowledge. Vendors are named and linked to their homepage even when their site is absent from the answer's reported search results. This is consistent with a known name mattering beyond any one page. It does not prove the mechanism or a ranking rule.
- Observation: agency questions lean on search results. Agencies, rankings and directories enter through the results ChatGPT reports reading. Presence there can be examined separately from whether the answer cites the site.
- Observation: official documentation leads on how-to questions. Google and OpenAI documentation is the most cited source on these questions in our corpus. That describes the observed sources, not a guarantee for another topic.
- Observation: reading is not citing. About two in three domains read are not cited in the same answer. Keep reported search presence, citations and brand mentions separate when assessing visibility.
Questions
How many sources does ChatGPT cite per answer?
In this corpus, the mean is 6.65 citations per answer, the median is 6 and the mean number of distinct cited domains is 5.8. Twenty of 844 answers (2.4%) cite nothing. These figures describe our AEO and GEO questions, not all ChatGPT answers.
Does ChatGPT cite the pages it reads?
Not systematically. At the domain level, only 34% of domains in reported search results are cited in the same answer. In answers with search results, 31% of cited domains are absent from those results. Reported search results do not reveal every step used to produce an answer.
Which brands does ChatGPT name for AEO tools?
Among the brands we track, Semrush appears in 197 answers, Ahrefs and Peec AI in 154 each, Otterly.AI in 145 and Profound in 144. These counts cover the whole corpus, not only tool questions. See the Profound alternatives comparison for the exact alternatives question.
Is GeckoBoost cited?
The brand-mention measure is 0 of 844 answers for GeckoBoost. That count concerns the brand being named, which is distinct from a website citation. Our case study documents our own visibility.
Can these results be used as ranking rules?
No. This is a dated observation of our own questions on one engine, in three languages. The questions, countries and days shape the results. It is not a ranking factor study and cannot show that a particular change will cause a citation.