How to Appear in Claude and Get Your Brand Cited

Appearing in Claude is neither bought nor requested. Anthropic does not sell placements inside its assistant’s answers and offers no application to get into them, so the only thing that decides is whether a company’s content is available when the assistant goes looking and whether it answers what someone asked.

For a company the practical consequence is that there is no switch to flip. What exists instead is robots that can fetch the site or cannot, pages that answer a concrete question or do not, and outside sources that describe the company the same way it describes itself or some other way.

Anthropic’s help centre, updated on 22 September 2026, puts it plainly: Claude invokes a search tool to ground its generated answers in content from the live web, and every answer includes citations so that the person asking can verify the sources themselves. That is the mode in which a company’s site can end up linked.

The other mode has a cut-off date. Anthropic publishes a knowledge cut-off for each of its models, and the eleven listed on 25 September 2026 range from August 2023 to June 2026. Whatever the assistant knows about a company in that mode comes from what was published before that date, almost always on sites the company does not control.

The consequence for the work is that there are two levers and they are not pulled the same way. A company’s own site decides what Claude can retrieve and cite when it searches; what outside sources publish decides what the assistant says about the company when it answers without searching. A change on the site only moves the first one.

The distinction matters because the usual fear is about training while the real cost sits in search. On Claude-SearchBot, Anthropic writes that disabling it “prevents our system from indexing your content for search optimization”. Blocking ClaudeBot keeps the site’s future material out of training; blocking Claude-SearchBot keeps the company out of the answers that cite.

The warning sits in the same Anthropic document: alternate methods such as blocking the IP addresses its robots operate from may not work correctly or persistently as an opt-out, because doing so stops Anthropic from reading the site’s robots.txt file. The opt-out it does recognise is the one written in that file.

The practical effect is that the decision gets taken without anyone seeing it. A bot-protection rule at the CDN returns a challenge to Claude-SearchBot, the site stops being indexed for the assistant’s search, and robots.txt still says everything is allowed. Whoever reviews the file signs off on access that is not there.

Anthropic also publishes, in a public file on its own site, the IP addresses its robots operate from, so a technical team can tell one of its visits from another that merely claims to be one. That file plus the server logs is what turns “we allowed access” into “access works”.

The figure comes with limits worth writing down before acting on it. It covers one B2B category, one tracked prompt set and 30 days, measured by a company that sells citation tracking. It shows the shape of the split in that case, not an industry number, and it promises nothing about another market.

A second measurement points the same way about the kind of source. The AI citation source index published by 5WPR on 1 May 2026, which brings together more than 680 million citations from six studies between August 2024 and April 2026, describes Claude leaning towards mastheads such as The New York Times, The Atlantic, The New Yorker and The Economist, with only 36% of its journalism citations coming from the previous twelve months.

The useful reading for a company is not to open a forum thread or get listed in any directory going. It is that on this engine the company’s own page enters the answer on its own merit, and that what sits beside it is registers and analysis rather than conversation. Which index each conversational engine uses, and where it gets the links it cites, is covered surface by surface in how AI models choose which sources to cite.

The method is technical and is set up once. A rule is created at the firewall or the CDN that blocks nothing: it logs requests with the ClaudeBot, Claude-SearchBot or Claude-User user agent and checks that the IP is on the list Anthropic publishes, to rule out anything merely claiming to be one of its robots.

Each robot counts something different. Claude-SearchBot visits show which pages are making it into the assistant’s search index. Claude-User visits are the most useful signal: that robot fetches a page when a person asks Claude to, so each visit is a conversation in which the assistant read that page. ClaudeBot visits only speak to training.

This is what DoGeo sets up for each client: the logging rule, the IP check and a monthly count by robot and by page.

What those logs do not count:

• How many people arrive from claude.ai after reading the answer: that is separated in web analytics.
• Whether the company gets named in answers that never needed to open its site. That is only visible with a fixed set of questions repeated every month.