Loading
Anthropic runs one bot for search, one for user requests and one for training, and you set each on its own. A site that blocks the wrong one can drop out of Claude's answers. This sheet sets out what each does.
Updated
From Anthropic's pages, read
Claude-SearchBot
Navigates the web to improve search result quality. Anthropic says blocking it stops your content being indexed for search optimization and may reduce your visibility and accuracy in answers.
For you: Allow it if you want Claude's search to find you.
Claude-User
Retrieves pages when a person asks Claude something. Anthropic says blocking it prevents your content being retrieved in response to a user query.
For you: Block this and a buyer who asks Claude to read your pricing page can't get it read.
ClaudeBot
Collects web content that could go towards model training. Anthropic lists it apart from the other two so you can limit it alone.
For you: You can refuse training and keep the other two allowed. Older robots.txt rules may name only this one, so read yours.
When a question calls for it
Claude invokes a search tool to inform and ground responses when a question benefits from current information. When it does, Anthropic says every response includes citations.
For you: Some answers come from what the model already knows, with no page read that day. We record which is which.
Owners decide
On Team and Enterprise, an Owner has to enable web search for the workspace before members can use it.
For you: A buyer at a company with search off gets an answer that never touched your site.
Your own analytics
These Anthropic articles give no tracking method. Clicks on Claude's citations reach your site as referral visits.
For you: Clicks are a partial count, since many readers stop at the answer. We read the answers as well.
Seen in practice, not documented by Anthropic
Based on the pages linked in the sheet above, read .
All five engines run on one retainer, so this work is never sold alone. The plan you pick sets how many questions we track and how much ships each term.
We write the questions your buyers ask, with you: 100 or 150 depending on the plan, mostly unbranded. You approve them before anything runs.
We read your robots.txt for all three bots, since older rules may name only one, and check your firewall lets them through. Fixes go into your plan's technical cycles.
Two weeks of answers across all five engines. For Claude, we record who gets named, how it describes you, and whether the answer cited pages or came from the model alone. On day 14 you get a written estimate of what the scope can move by day 90, as ranges per cluster and engine.
Comparison and alternatives pages, revamps of pages you already have, corrected review profiles, placements on the pages Claude cites, and community work. Each month end is measured against the forecast.
Nobody controls what the engines say. You get a measured baseline, a written forecast and a monthly check that shows a flat month as flat.
We don't take your direct competitor, so we are never chasing the same citations for two of you.
We interview your product lead at the start, then for 30 minutes a month, and you approve every brief.
Each monthly check reads the same questions again and sets the result against the forecast. A flat month shows as flat. This is the shape of it, filled in for a made-up company.
Answers naming each, out of 100 questions.
Below range
Range set on day 14 for this cluster: 10 to 18 answers. This month: 8.
Right: Category and who the product suits.
Wrong: Describes a plan the product no longer sells.
No citation behind it, so it comes from the model. Current pages and placements are next month's work.
Visits referred from Claude: 9
It is the work of getting your product named, and described correctly, when buyers ask Claude about your category. It covers access for Anthropic's search and user bots, the pages Claude cites, and what those pages say about you.
Send your website, your category in a sentence and up to five competitors. If we think the free audit will tell you something, it arrives 48 business hours after we confirm.