--- title: How AI Assistants Pick Developer Docs to Cite: 80 Answers description: 80 answers from Claude and Codex: Codex limited 78/94 searches to sites it trusts; Claude cited 133/144 links from results it never opened. How to get cited. canonical: https://spiritroom.app/blog/how-ai-assistants-cite-developer-docs/ published: 2026-10-21 updated: 2026-10-21 --- > Markdown version of https://spiritroom.app/blog/how-ai-assistants-cite-developer-docs/ for agents. Index: /llms.txt [Blog](/blog/) / [Agent experience (AX)](/guides/agent-experience-ax/) # How AI assistants pick which developer docs to cite: 80 answers from Claude and Codex If developers ask an AI assistant instead of a search engine, your docs need to be what the assistant cites. We watched two assistants answer the same 20 developer questions, and they read almost entirely different parts of the web. Leonid Bugaev · Published 21 Oct 2026 **TL;DR** - **Two different webs:** with search on, Claude cited 136 pages on 87 sites, Codex 83 pages on 47 sites. They shared 5 pages. In 11 of 20 questions, no cited site overlapped. - **Codex goes to sources it already trusts:** 78 of its 94 searches were restricted to one site (modelcontextprotocol.io alone 24 times). - **Claude cites the results page:** 133 of its 144 links came from search results it never opened. - **Fresh, specific, numeric wins:** 17 of Claude's 29 queries ended in “2026”; it picked pages with concrete measurements. - **Community sites barely appear:** 0 Hacker News, Stack Overflow or YouTube links; one Reddit result, never cited. [What we asked](#setup)[Two assistants, two webs](#two)[Codex: site-restricted search](#codex)[Claude: cite from the result list](#claude)[What gets cited](#formats)[What to do with your docs](#do)[How we measured](#how-we-measured)[FAQ](#faq) ## What we asked 20 developer questions, such as “How to test an MCP server with AI agents”, “How many tools can an MCP server have before agents get worse?” and “How do I make my API agent-ready?”. Each went to Claude (in Claude Code) and Codex, once without web access and once with web search: 80 answers, all returned. Neither assistant knew who was asking. ## Two assistants, two webs Claude: cited pages *136 pages, 87 sites* Codex: cited pages *83 pages, 47 sites* Shared by both *5 pages, 10 sites* In 11 of 20 questions the two assistants' cited sites didn't overlap at all. Of 124 cited sites, 91 were cited by only one answer. Optimising for “AI search” means optimising for at least two different behaviours. **Want to know where agents and assistants find your product?** We test how AI assistants and coding agents discover, cite and use your product, and show what to publish or fix. [Talk to us →](/#contact) ## Codex: site-restricted search - Searched in all 20 with-search answers: 94 searches, median 5 per answer. - **78 of 94** were restricted to one site: modelcontextprotocol.io 24 times, anthropic.com 7, code.claude.com 5, then framework and vendor docs. - Its 103 links: 53 vendor sites, 26 spec sites, 21 GitHub, 1 elsewhere. No directories, papers or community posts. - Queries were short (about 6 words) and only 1 had a year. To get into a Codex answer, your page has to be on a site Codex already thinks to target: the spec, the big platforms' docs, GitHub. ## Claude: cite from the result list - Searched in 18 of 20 with-search answers, usually 2 searches, about 9 words each; 17 of 29 ended in “2026”. - About 9 results per search (258 total); 43 had “2026” in the title or URL. - **133 of 144** links came straight from results it never opened; 8 from pages it opened; 3 from memory. - A second query often named tools from the first results, so a product enters the answer by being on the first result page. - For the two Claude Code questions it went straight to code.claude.com. ## What gets cited | Kind of page | Links (of 247) | | --- | --- | | Vendor sites (about half official docs) | 115 | | Spec and official project sites | 36 | | Small independent blogs | 29 | | GitHub | 27 | | MCP directories (Claude only) | 12 | | Analyst and media | 8 | | Package registries / research papers | 7 / 7 | | Community (dev.to, Medium) | 4 | | Reddit, Hacker News, Stack Overflow, YouTube | 0 | Pages with concrete numbers were picked over opinion: a server's token count, an app's tool limit, “20 of 20 runs”. On “which tool?” questions, vendor-written “best X 2026” roundups shaped the picks. ## What to do with your docs 1. **Put a number in the title and first lines.** That's what gets quoted from a result snippet. 2. **Date it.** Claude adds the year to queries; fresh titles match. 3. **Use the question's words** in the title and H1. 4. **Get onto sites Codex targets:** contribute to spec docs or examples, framework integration pages, GitHub. 5. **Serve clean Markdown** and an `llms.txt`, for assistants that do open pages. See [what agents use: llms.txt, AGENTS.md, MCP](/blog/llms-txt-agents-md-what-agents-use/). 6. **Re-run the same questions** after publishing to see whether anything changed. **Go deeper:** [Agent experience (AX)](/guides/agent-experience-ax/), [Agent-ready tools](/guides/agent-ready-tools/), [Pass rates with ranges](/blog/agent-pass-rates-confidence-intervals/). Back to [agent experience (AX)](/guides/agent-experience-ax/). ## How we measured - **Run:** 20 questions × Claude and Codex × with and without search = 80 answers, 21 Oct 2026, 18:56–19:22 (Istanbul time, UTC+3), one sample each. - **Extraction:** searches, opened pages and cited links from each assistant's own transcript; brands by an automated extractor capped at 24 names per answer. - **Limits:** one sample per question and mode; Codex doesn't expose its result lists, so “seen in results” is Claude-only; no Google AI Overviews, ChatGPT or Gemini in this pass. ## FAQ ### How do I get my developer docs cited by AI assistants? Publish specific, dated pages with concrete numbers in the title and first lines, on a site the assistant already trusts or one that ranks in fresh search results. In our test Codex mostly searched official sites by name, while Claude cited titles straight from search results. ### Do Claude and Codex cite the same sources? Rarely. Across 40 with-search answers they shared 5 cited pages, and in 11 of 20 questions their cited sites didn't overlap at all. ### Does Reddit or Stack Overflow matter for AI citations here? Not in this test: 0 Stack Overflow, Hacker News or YouTube links, and one Reddit result that was never cited. Results may differ for other topics. ### Do AI assistants read the pages they cite? Claude often didn't: 133 of its 144 links came from search results it never opened. So your title and snippet do most of the work. **Want to know where agents and assistants find your product?** We test how AI assistants and coding agents discover, cite and use your product, and show what to publish or fix. [Talk to us →](/#contact) ## Keep reading - Up[Agent experience (AX): how to measure it (guide)](/guides/agent-experience-ax/) - Related[Do AI agents read llms.txt? 18 runs in three agent apps](/blog/llms-txt-agents-md-what-agents-use/) - Related[How to make your MCP server or API agent-ready, with proof (guide hub)](/guides/agent-ready-tools/)