Skip to content
Zaphia AI

Home  /  How AI Search Works  /  Claude

Per-engine guide

How Claude decides which businesses to recommend

Claude is the most careful of the major assistants. It searches when it needs to, cites what it used, and would rather hedge than name a business it cannot verify. That makes it a smaller channel for local search, and an unusually clear test of whether your information is solid.

Updated July 19, 2026   We keep this page current as Claude changes.

The short version

If you only read one box

Claude answers from training data plus a live web search it chooses to run per question, and it cites the pages it used. It is deliberately cautious, so thin evidence usually produces a hedge rather than a recommendation.

  • Allow Claude-SearchBot and Claude-User. Those two affect visibility. ClaudeBot is training only and can be blocked.
  • Verifiability is the currency. Clear, specific, well-structured pages are what it is willing to stand behind.
  • It is a smaller local-search channel. Do not build your plan around it, but you get it for free by doing the rest well.

How Claude sources its answers

Answer firstClaude draws on two things: what it learned in training, and a live web search tool it can choose to use. For anything current or local it will typically search, then answer with clickable citations to the pages it drew from.

Web search reached paying users in the US in early 2025, arrived in the API shortly after, and expanded to free users globally later that year. In the app it is a setting the user controls, and within that, Claude decides per question whether searching is warranted. Ask it something stable and general and it may simply answer. Ask it for the best clinic in your city and it will go and look.

Anthropic also offers a Citations capability that lets Claude quote exact sentences from documents it has been given. That is mostly relevant to businesses using the API, but it tells you something about the design philosophy: this is a system built to point at where an answer came from.

Sources: Anthropic on web search, Anthropic on the Citations API.

What Claude favors

Answer firstDeep, well-organized, genuinely extractable pages from sources it considers reliable. Independent testing suggests its consumer web search leans on Brave Search results, so pages that rank there are more likely to be seen.

Two caveats before you act on that. First, the Brave connection comes from independent analysis and a subprocessor listing rather than a detailed Anthropic disclosure. Second, one study found Claude's citations overlapped Brave's top organic results around 87% of the time, which is a striking figure from a single test and should be treated as directional rather than a law of nature.

What is consistent across observations is the character of what it picks: substantial pages with a clear structure, unambiguous facts, and no overselling. Claude is not looking for the most enthusiastic page, it is looking for the one it can safely summarize.

Sources: Analysis of Claude web search, study of Claude source selection (both third-party, directional).

Why it hedges instead of recommending

Answer firstClaude is tuned to avoid stating things it cannot support. When the evidence about a small business is thin, it will describe the category, list what to look for, or mention established names, rather than confidently recommending you.

This is worth understanding because it reframes what you are working toward. With some engines the goal is to be popular. With Claude, the goal is to be verifiable. A business with a clear website, matching details everywhere, named practitioners with real credentials, and specific recent reviews gives Claude enough to stand behind a recommendation. A business with a vague site and a thin trail gives it a reason to stay general.

Observers describe it as leaning toward established, authoritative sources and away from confidently naming a small local business. Treat that as a tendency rather than a rule, since it comes from third-party testing rather than published behavior.

A useful side effect: Claude works well as a self-test. Ask it about your own business and read the answer honestly. If it hedges, describes you vaguely, or gets a detail wrong, that is a fair signal that the public evidence about you is thin or inconsistent, and that will be costing you on other engines too.

Source: Third-party observation of Claude visibility behavior (directional).

The three Anthropic crawlers

Answer firstAnthropic runs three separate crawlers and you control each independently. Claude-SearchBot and Claude-User affect whether Claude can find and read you. ClaudeBot is training only and can be blocked without cost to visibility.
CrawlerWhat it doesKeep it allowed?
Claude-SearchBotIndexes pages to improve Claude's search answers. Blocking it can reduce your visibility in those answers.Yes, allow
Claude-UserFetches a page in real time when a user's question calls for it.Yes, allow
ClaudeBotCollects content for model training. Does not affect whether Claude can cite you.Optional to block
The practical setup: allow Claude-SearchBot and Claude-User, and block ClaudeBot only if you want to opt out of training. All three follow robots.txt directives including Disallow and Crawl-delay.

Source: Anthropic support: crawling and how to block it.

How to earn a mention

Answer firstGive Claude enough verifiable detail that recommending you is a safe call: precise service descriptions, named and credentialed people, consistent business details, and specific recent reviews.
  • Name the humans. Practitioners, credentials, and who is responsible. Claude weighs identifiable expertise heavily when deciding whether to vouch for something.
  • Be precise instead of enthusiastic. "Board-certified nurse injector, 9 years in aesthetics, Botox from $12 per unit" is usable. "Award-winning luxury experience" is not.
  • Keep details identical everywhere. Contradictions between your site, your listings, and directories are exactly what makes a cautious system retreat to generalities.
  • Structure the page cleanly. Real headings, plain HTML text, and one idea per section. Claude reads structure well and rewards it.
  • Collect specific reviews. Reviews that name the service performed are evidence. Star ratings alone are not.
This is exactly what a Zaphia AI Kit does for you. The Med Spa AI Visibility Kit walks you through the credential, service, and review detail that makes a cautious engine willing to name you.

How much should Claude matter to you

Answer firstLess than Google, ChatGPT, or Perplexity if you are a local service business. Claude is used more heavily for professional and technical work than for finding a nearby practice. Do not build a separate strategy for it.

We would rather be straight with you than pad this page. For a med spa or a salon, Claude is a secondary channel today. It is also worth noting that much of what the industry believes about Claude's source selection comes from third-party testing rather than Anthropic disclosures, so confidence here should be lower than on the Google or Bing pages, where the documentation is direct.

The reassuring part: there is nothing separate to do. Clear pages, named experts, consistent details, and real reviews are the same things that win everywhere else. Claude simply has a higher bar for the same evidence, so if you satisfy Claude you are almost certainly in good shape elsewhere.

Source: Comparative usage of AI assistants.

Frequently asked

Does Claude search the web?

Yes. Claude has a web search tool that reached paying users in early 2025 and expanded to free users globally later that year. It decides per question whether a search is needed, then answers with clickable citations to the pages it used. For stable general knowledge it often answers from training data instead, so not every question triggers a search.

Should I block ClaudeBot?

You can block ClaudeBot without losing visibility in Claude's answers, because ClaudeBot is the training crawler. The ones that affect whether Claude can find and cite you are Claude-SearchBot, which builds the search index, and Claude-User, which fetches a page when someone asks about it. Keep those two allowed.

Why does Claude not recommend my local business?

Claude is noticeably cautious. When evidence is thin it tends to hedge, describe a category, or point to well-established names rather than confidently recommending a specific small business. If your business has limited third-party presence, few detailed reviews, or inconsistent details across the web, Claude will usually decline to name you rather than risk being wrong.

Is Claude worth optimizing for as a small business?

It is a smaller channel than Google, ChatGPT, or Perplexity for local recommendations, with more of its use in professional and technical work. The good news is that you do not need a separate Claude strategy. The clear, well-structured, verifiable pages that earn citations everywhere else are exactly what Claude rewards, so it comes along with the rest of the work.

Do it, do not just read it

Make your business easy to verify

The Med Spa AI Visibility Kit gives you the service, credential, and review detail that makes even a cautious engine willing to name you.

See the Med Spa Kit

← Back to How AI Search Works