01Notes · 26 August 2026
Getting Cited by Claude, and Why It Differs
Quick answer
Claude has a smaller audience than ChatGPT and a differently shaped one: it skews toward developers, analysts and technical evaluators. That changes which of your pages get used, not just how many.
The practical consequence for most B2B companies: your documentation matters more here than your marketing site, and the person using it is often the one who gates the purchase rather than signs it.
Key takeaways
- Weight it by buyer, not by audience size. If a technical evaluator screens vendors before anyone talks to sales, this surface reaches them.
- Documentation outperforms marketing copy in our sampling — consistent with an audience that is evaluating rather than browsing.
- Pages that state limitations do unusually well. Honest constraints read as credible to a technical reader.
- ClaudeBot handles web content. Allow or block it deliberately, with the reason recorded.
- ChatGPT results do not predict Claude results. The two most similar platforms shared only 24.1% of cited pages.
- Sample it separately or not at all. A blended AI visibility number tells you nothing about this surface.
Who is actually on the other end
Every AI platform comparison ranks by monthly users, and on that measure Claude sits well below ChatGPT and Google’s surfaces. For most consumer categories that settles it.
For B2B it does not, because the question is not how many people use a platform — it is whether your buyer does. In technical categories the person who screens vendors before anyone speaks to sales is frequently an engineer or analyst, and that is the population this surface over-indexes on.
A committee of six where one person does the shortlist research is common. If that person uses Claude and you are absent from it, you were removed from consideration by a surface you never measured.
Which pages get used
| Page type | How it performs | What that tells you |
|---|---|---|
| Technical documentation | Strong | Specific, factual, structured, and written to answer rather than sell |
| Explainers with worked detail | Strong | Reasoning is valued over conclusions |
| Pages stating limitations | Unusually strong | Honest constraints read as credible, not as weakness |
| Comparison content with real trade-offs | Good | A table where you win every row is not usable as a recommendation |
| Marketing and landing pages | Weak | Your story about yourself is the least useful input available |
| Thin SEO content | Very weak | Nothing extractable, and it shows |
Labelled Observedthis is our own sampling, not published research. Treat it as directional and re-check it yourself.
The documentation point, again
We keep arriving at this from different directions, so it is worth stating once more.
Technical documentation has exactly the profile that gets cited: specific, factual, structured, updated regularly, and written to answer a question rather than to persuade. On a surface used by technical evaluators, that advantage compounds.
And yet most B2B companies treat docs as a post-sale support cost. They sit outside the content strategy, nobody measures them, and a surprising number are blocked from crawlers entirely by a robots rule someone added for a staging environment years ago.
Paying for content marketing while blocking your best citation candidate is an expensive contradiction, and it takes an hour to check. Crawler policy work is here.
How to sample it
Same procedure as any other platform, and separately from all of them. Fresh conversation per run, fixed locale, three runs per prompt, results coded as cited, named or absent, with accuracy recorded when you appear.
What not to do is infer Claude from ChatGPT. The two most similar platforms in 2026 research shared only 24.1% of cited pages, and every other pair overlapped less. A strong ChatGPT position tells you almost nothing here.
The full method is published, including the parts it cannot see.
What this means for you
- Decide whether a technical evaluator gates your sale. If yes, this surface is worth the sampling time regardless of its user numbers.
- Check your documentation is crawlable before commissioning any new content.
- Never infer one platform from another. 24.1% overlap at the highest measured.
02Common questions
Questions this raises
Is Claude worth measuring given the audience size?
It depends entirely on who buys from you. Claude skews toward developers, analysts and technical evaluators. If your buying committee includes someone technical who does the pre-vendor research, that person may well be using it.
Which crawler does Claude use?
ClaudeBot for web content. As with every agent, allowing or blocking it should be a recorded decision with a reason rather than a default copied from a template.
Does Claude cite sources?
When it retrieves, yes, with links. Like every assistant it also answers from model knowledge at times, and those answers carry no citation to trace.
What kind of pages does it seem to favour?
In our sampling, documentation, technical explainers and pages that state limitations. Marketing copy performs poorly, which is consistent with an audience that is evaluating rather than browsing.
Can I assume ChatGPT results predict Claude results?
No. In 2026 research the two most similar platforms shared only 24.1% of cited pages, and every other pair overlapped less. Each platform needs its own sampling.
Where should Claude sit in our priorities?
Below the Google surfaces on reach for most businesses. Higher if you sell developer tools, infrastructure or anything where a technical evaluator gates the purchase.
Sources
Every figure on this page comes from one of the following. Where a claim is ours rather than a source’s, it carries an evidence label instead.
- Rankability — State of AI Search, 1,645 citation observations across 31 topics, 2026. Source for the 24.1% cross-platform overlap and the 73.5% ChatGPT exclusive citation rate. Independent study
- Anthropic — crawler documentation covering ClaudeBot and related agents. Primary
- Searlixa sampling across client engagements, labelled Observed with sample and date. Our own
Ask an assistant
Hand this page to an assistant and ask it to summarise or challenge the argument. If it gets something wrong about us, that is itself a useful finding.
→Next step