Claude is noticeably more cautious about naming specific businesses than the other assistants, particularly in regulated categories. Absence is often not a crawling problem at all but a corroboration problem: it will describe the category confidently and decline to recommend anyone it cannot verify from established sources.

The short version

Why this happens

Claude behaves differently enough from the other assistants that firms often assume something is broken when nothing is.

Caution about recommendations. Ask several assistants to recommend a supplier and you will see the difference. Some will name three firms readily. Claude is more likely to explain what to look for in a supplier, list the criteria, and stop short of naming anyone, particularly where a wrong recommendation could cause harm. In financial services, legal, healthcare and anything touching money or wellbeing, this caution is most pronounced.

That is a design choice, not a fault, and it changes what works. You do not overcome it by publishing more of your own content, because the reticence is not about how much it knows about you. It is about how confident it is that naming you is appropriate.

What does move it is verifiable, independent corroboration: professional body membership, regulator registration, established directory presence, genuine third-party coverage. Claude responds to the kind of evidence that would satisfy a careful person checking whether a firm is legitimate, because that is broadly the standard it is applying.

Source preference. Claude leans toward established and stable sources over fresh ones. This is the mirror image of Perplexity. A recent blog post moves Perplexity faster; a long-standing entry in a recognised register moves Claude more.

Crawlers. ClaudeBot gathers training data. Claude-User fetches pages when a user’s question requires it. As with the others, these are governed separately, and a rule written for one does not cover the other.

Training lag. Like ChatGPT, Claude may answer from training rather than retrieval. If it holds an outdated view of your business, that corrects on a training cycle rather than on yours.

How to check it yourself

  1. Ask for a recommendation in your category and note whether it names anyone at all. If it names nobody, that is category caution rather than a problem with you specifically.
  2. Ask about your firm directly. If it can describe you accurately, you are identifiable, and the gap is willingness to recommend rather than knowledge.
  3. Ask what it would want to know before recommending a firm in your category. The answer is a usable checklist of the corroboration it is looking for.
  4. Check robots.txt for ClaudeBot and Claude-User separately.
  5. Ask it to search explicitly, then compare. That separates the retrieval question from the training question.

What to do about it

  1. Allow both agents.
  2. Prioritise verifiable credentials: regulator registration, professional bodies, accreditations, and make them machine-readable in text rather than as a logo image.
  3. Publish a machine-facing statement of fact that a cautious system can check you against. Ours is at /llm-info.html.
  4. Build presence in established, stable sources rather than chasing recency.
  5. Set expectations internally. In regulated categories, Claude naming you at all is a higher bar than the other platforms, and it is reasonable for it to move last.

The full method is in how to get cited by Claude and how Anthropic’s model approaches citations.

Platform behaviour changes. This page reflects how Claude worked in August 2026 and is reviewed quarterly.

Back to all GEO questions →