Top 100 websites cited by AI (2026)

The 10 websites AI cites most

Across all four combined, in order: 1. wikipedia.org, 2. reddit.com, 3. youtube.com, 4. github.com, 5. nih.gov, 6. ncbi.nlm.nih.gov, 7. quora.com, 8. developer.mozilla.org, 9. nytimes.com, 10. learn.microsoft.com. The head is encyclopedic, institutional, and community-driven rather than brand-led, and it is flatter than a search ranking: the leader takes 3.44% and the 100th row still takes 0.11%. By category, News is the largest slice of the combined hundred at 19% of its citation share, followed by Software & docs at 15%.

The top 5 for each assistant

ChatGPT: wikipedia.org, g2.com, reddit.com, britannica.com, openai.com. Claude: wikipedia.org, github.com, developer.mozilla.org, ncbi.nlm.nih.gov, nih.gov. Gemini: youtube.com, reddit.com, quora.com, linkedin.com, amazon.com. Perplexity: reddit.com, youtube.com, nytimes.com, wikipedia.org, yelp.com. The overlap between any two of those lists is thinner than the shared phrase "cited by AI" suggests. Software & docs accounts for 38% of Claude's hundred; on Gemini the largest category is News at 17%, and on Perplexity it is News at 32%.

What gets cited by all four

55 of the 100 domains here appear in every assistant's top hundred. They have a type: reference works, government and health institutions, standards and documentation, and the wire services. What they share is not authority in the marketing sense but resolvability, a stable URL that answers one question completely and is unlikely to be contradicted. That is a far narrower target than ranking on Google, and it is the practical reading of this page.

What this data does not tell you

Every ranking hides its own assumptions. These are the ones worth knowing before you quote a number from this page.

• These are modelled estimates, not a measured crawl. None of the four companies publishes a citation log, and nobody outside them can count citations directly. Quote the shape of this data, which domains lead and which are missing, rather than a decimal place.
• The studies this is calibrated against disagree with each other. Profound puts Wikipedia first for ChatGPT at 7.8% with Reddit far behind on 1.8%; other published rankings have Reddit at the top of the same list. They sample different questions on different days, which is the entire explanation. Where they conflict, this page follows Profound.
• Claude is the least constrained of the four. It appears in none of the public studies, so its ranking here comes from the way it retrieves rather than from anyone's measurement of it, and it should carry the widest error bars on the page.
• A citation is not a visit. Appearing in an answer puts a domain in front of a reader who, more often than not, never clicks. Read the ranking as share of attention inside the answer, not as traffic.
• The ranking follows the question mix. Ask a hundred shopping questions and you get Gemini's list; ask a hundred coding questions and you get Claude's. Every AI-citation study is partly a ranking of the questions its authors asked, this one included.
• Averaging the four is the weakest number on the page. wikipedia.org is the only domain in all four top tens, and 4 of the 100 appear on a single assistant's list alone. The combined view is a starting point rather than a target list.
• Assistants over-cite their own documentation. anthropic.com sits 10th on Claude and outside every other hundred, which is a property of who is answering rather than of the domain.
• The estimate is pinned to July 2026 and retrieval stacks change month to month. The OpenRush MCP server returns live measurements of the search results these systems retrieve from, through inspect_serp and inspect_domain.