How to appear in Claude
Claude is the engine where the memory path weighs most and where there is no paid lane. Those two together make it the slowest to move and the most honest about who has substance.
Diego MoraesCTO
Open in ChatGPT
Open in ClaudeClaude is the engine where the memory path weighs most and where there is no paid lane. Those two things together make it the slowest engine to change and the most honest about who has substance.
How Claude decides
In base mode, from memory. The model answers with what stayed from training: entities, and what many independent sources say about them. A brand only ever asserted by itself is an assertion without confirmation, and memory does not keep it.
With search, from institutional sources. When it searches, it prefers clear documentation and cites reference journalism more often than the other engines (5W Citation Source Index, June 2026; vendor data, treat as directional). In the app, search is always on; in the API it is an optional tool.
No advertising. By public decision of Anthropic. There is no paid column to make up for the earned one, which makes reading what appears cleaner, and the presence work more demanding.
What to do, in order
Allow the three Anthropic agents, which do different things
Anthropic splits the crawling into three: ClaudeBot collects for future training, Claude-SearchBot feeds the search index, and Claude-User fetches a page at the moment somebody asks for it in a conversation. Blocking the first out of caution and leaving the other two is a legitimate decision; blocking all three is being out of the search path and the memory path at the same time.
Do: in robots.txt, allow at least Claude-SearchBot and Claude-User. Decide on ClaudeBot with a cool head: whoever lives on being found tends to gain by allowing it.
Be a stable entity, with the same description everywhere
In base mode Claude answers mostly from memory, and memory stores entities, not pages. A brand described three different ways in three profiles is three weak entities instead of one strong one. The house playbook says the same for every engine, and here it weighs more because the memory path weighs more.
Do: one name, a one-line description, the same sameAs in the organisation schema pointing at profiles that actually exist. And the Wikidata item only after there are public references: an item without references gets deleted, and a sameAs to a deleted item is worse than none.
Write directly, factually, without inflated promises
With search on, Claude prefers institutional sources and clear documentation, and cites reference journalism more often than the other engines (5W Citation Source Index, June 2026; vendor data, treat as directional). Marketing copy with superlatives is precisely what this engine tends to leave out.
Do: pages that explain what the brand does, for whom, and with what verifiable result. Technical documentation, original data and verifiable comparisons are worth more here than in any other engine.
Go after reference press, not clipping
It is the source profile Claude cites most. It is not any mention: it is the reference titles of the category and of the country. One interview in an outlet the engine cites is worth more than ten press releases in portals it does not read.
Do: find out which sources Claude cites in the questions of your category and work on presence in those, one by one. It is relationship work, and it is slow: it feeds the training memory of the next cycle.
Measure in the app, not only in the API
The Claude API is not the Claude the buyer opens. In the app search is always on, and the interface, the context and the way results enter the answer change what it cites. We measured both paths on the same questions in August 2026: a different number of sources per answer (7 on one path, 16 on the other) and only 24% of the named brands overlapped. And Claude Code is yet another engine: it uses web search in 13% of answers against 93% in claude.ai, with about 20% overlapping brands for the same question (Profound, 24 August 2026; Search Engine Journal, 31 August 2026).
Do: if you measure Claude, measure the app, in a real session, and not only the API. It is what we do for the half with search; the memory half stays on the API on purpose, because in the app you cannot turn search off.
The check, in one command
So that this does not stay theory, the part that is verified with a command. Swap the domain for yours.
The Anthropic agents have a way in
$ curl -s https://example.com/robots.txt | grep -iA1 'ClaudeBot\|Claude-SearchBot\|Claude-User' User-Agent: ClaudeBot Allow: / -- User-Agent: Claude-SearchBot Allow: / -- User-Agent: Claude-User Allow: /
Three agents, three blocks. If you only have a generic rule with Disallow, the three are left out and the rest of the work never gets to count. The full check on whether the site is legible is in six commands to find out whether the AI can read your site.
What does not work
There is no submission form, no brand panel and no advertising. Marketing copy with superlatives is what this engine leaves out most. And measuring only through the API is measuring a path the buyer does not use: the two paths agree on fewer than a quarter of the brands they name.
- 93%of answers in claude.ai use web search
- 13%of answers in Claude Code use web search
- 24%of the named brands overlapped between the API and the app, on the same questions
Vendor data, treat as directional. Sources: Profound, 24 August 2026 (web search); our measurement of the two paths, August 2026 (overlap).
Frequently asked questions
No. Anthropic decided publicly not to sell advertising in Claude. Earned presence is the only path, and that is why entity work, documentation and reference press weigh more here than in engines with a paid lane.
In the app, yes: web search is on by default and is used in 93% of answers, according to Profound (August 2026). In the API, search is a tool the caller turns on or not. In Claude Code, it is used in 13% of answers. Those are three different behaviours, and what you measure has to be what the buyer uses.
The foundation is the same: a consistent entity, content that answers the questions of the category, presence in the sources the engine reads. The difference is in the weight of the sources. Claude cites more reference journalism and institutional documentation; ChatGPT reads the Bing index and weighs lists and comparisons more. You measure them one by one.
Through the search path, weeks: as soon as the content exists, is findable and answers the question, it can be cited. Through the model's memory, months, at the pace of Anthropic's training cycles. Nobody serious promises a deadline; what is promised is weekly measurement.
Read next
- How to appear in ChatGPT, the engine with the other balance between memory and search.
- How to appear in Perplexity, the engine that always searches.
- Knowledge vs augmented, the two paths explained.