This index covers the five AI assistants tracked on this site: ChatGPT, Claude, Gemini, Grok and Perplexity. For each one it shows which websites and platforms appear in its answers, measured on the same set of weekly questions, so you can see how differently the models source information and which of them matter for your visibility.
Is your site cited by AI?
Type a domain or a brand to see whether ChatGPT, Claude, Gemini, Grok and Perplexity cite it among the most cited sources of this dataset.
Key takeaways
- First source of each model: Wikipedia for ChatGPT, Trustpilot for Gemini, YouTube for Perplexity and sortlist.fr for Claude and Grok.
- Gemini cites the 37 tracked platforms 6.5 times more than Claude (28.54 vs 4.37 citations per 100 answers in total).
- ChatGPT has the most analyzed answers (114'466), Grok the fewest (22'038): compare rates, not raw counts.
Summarize this page with ChatGPTClaudePerplexity
The five AI models side by side
| Model | Answers | French / English | First domain | First platform | Platforms per 100 answers | Government share of top 20 |
|---|---|---|---|---|---|---|
 ChatGPT | 114'466 | 100'139 / 13'502 |  Wikipedia wikipedia.org | Wikipedia | 16.06 | 48% |
 Gemini | 114'458 | 100'131 / 13'502 |  Trustpilot trustpilot.com |  YouTube | 28.54 | 11% |
 Perplexity | 113'962 | 99'685 / 13'452 | YouTube youtube.com | YouTube | 9.85 | 2% |
 Claude | 22'547 | 21'591 / 956 |  sortlist.fr |  Amazon | 4.37 | 0% |
 Grok | 22'038 | 21'132 / 906 | sortlist.fr | Trustpilot | 23.17 | 6% |
Platforms per 100 answers: sum of the rates of the 37 tracked platforms. Government share: share of government websites in the citations of the model's 20 most cited domains.
All 5 AI models
No match. Try another spelling.
Why compare AI models on their sources?
The five assistants do not answer a question with the same sources. Some search the web for nearly every prompt, others lean more on what they learned during training, and they do not query the same search index. A site can therefore be cited regularly by one model and ignored by another.
For GEO (Generative Engine Optimization), the practice of getting a brand mentioned in AI answers, this gap is the starting point. Your audience spreads across several assistants, and a strategy tuned to the habits of one model can leave you invisible in the others.
How do the five models find their sources?
- ChatGPT answers from its training data and switches to web search when a question calls for recent or specific information, then lists the pages it used.
- Gemini can ground its answers in Google Search results, which ties its sources closely to classic search rankings.
- Perplexity runs a web search for every question and displays numbered sources next to the answer.
- Grok combines web search with direct access to posts on X.
- Claude cites sources less often by default and relies on its web search tool when that tool is active.
These behaviours shift as the products evolve, one reason why the figures are refreshed every month rather than frozen.
How to read the model pages
The number under each model is the volume of answers analysed. It differs from one model to another, so compare them with the rate per 100 answers, which neutralises volume, and not with raw citation counts.
Each model page lists the domains and platforms that model cites most, with their monthly evolution. Look for gaps: a platform that one model cites far more than the others points to a source you can work on for that assistant in particular.
How to use this index
- Start with the assistants your customers actually use, then open their pages.
- Compare the top platforms of two models to spot sources that only one of them favours.
- Cross-check with the platform pages, which show the same split seen from each platform.
- Come back after major model updates: a change in search behaviour shows up in the monthly curves.
Frequently asked questions
Which AI assistant cites the most sources?
Assistants that search the web for every question, such as Perplexity, show sources systematically, while a model answering from memory may cite none at all. The model pages show how often each assistant cites the tracked platforms and domains, month by month, so you can compare their habits on the same questions.
Are the same questions asked to every AI model?
The questions come from the same weekly tracking that Sorank runs for its clients, and every answer is analysed with the same method whatever the model. Volumes can still differ from one model to another, which is why all figures are expressed per 100 answers rather than as raw totals.
Is Google part of this comparison?
Gemini, Google's AI assistant, is one of the five models compared. Google Search itself, however, is not counted as a cited platform: it is the search infrastructure Gemini relies on, not a source of content in its own right, so it is excluded from the platform figures.
If one AI cites my site, will the others cite it too?
Not necessarily. Each model has its own way of searching and selecting pages, so a citation in Perplexity does not guarantee one in ChatGPT or Claude. Work on the sources each model favours, and check your presence model by model rather than assuming a single result applies to all five.
Method and sources
387'471 answers from ChatGPT, Claude, Gemini, Grok and Perplexity, from August 2025 to September 2026, to the questions Sorank asks every week to follow its clients' visibility (88% of them in French). A citation is a source domain given in an answer; the rate is the number of citations per 100 answers. Data snapshot of 2 October 2026.
Full methodology · Raw data (JSON)
How to cite this page: Sorank Data, “AI models compared: sources each one cites”, data.sorank.com, data snapshot of 2 October 2026.
Track the citations of your own site with Sorank's AI mention tracking.
