只需填写【中文企业/品牌名】+【微信号 / 手机号 / 工作邮箱】，或直接扫码添加顾问微信「美链数智-Joy」，1–3 个工作日免费获取未删减 AI 推荐截图与同行对标清单。
Make beauty evidence accessible to Claude search
Claude can search the web and visit sources when needed. Anthropic documents separate search, user-request and training bots. An audit checks the relevant access path and the actual response before reporting a citation.
Access and evidence
| Check | What the evidence establishes |
|---|---|
| Bot roles | Claude-SearchBot supports search; Claude-User handles user requests; ClaudeBot is used for potential training data. Official role definitions. |
| Publisher controls | Anthropic says its bots respect robots.txt. Check each relevant subdomain and retain the intended training preference. |
| Identity check | Anthropic publishes crawler IP ranges. Verify the IP and use the user agent to distinguish roles; the shared list alone does not identify a search request. Official IP list. |
| Search state | Record whether search actually ran. The current Claude interface may not expose the earlier manual web-search switch. Consumer search guidance. |
| llms.txt | Anthropic's tools article mentions supplying developer documentation to Claude Code. That limited context does not establish a website search-ranking signal. |
Practical priorities for beauty businesses
- Review Claude-SearchBot and Claude-User access on the public pages you want buyers to retrieve; keep ClaudeBot policy separate.
- Provide core company and service facts in readable HTML. Anthropic's API web-fetch tool does not support JavaScript-rendered content; do not assume every browsing tool has the same limitation.
- Use descriptive headings and concise summaries to help readers locate specifications, evidence and limitations.
- Keep llms.txt optional and accurate if maintained; prioritize public pages and evidence over a presumed ranking benefit.
- Save the exact model, interface, search activity, answer and citations for each independent test.
How to measure this surface
Separate consumer Claude answers from API web-search and direct-fetch experiments. A requested URL fetch establishes only that access path. Record unavailable tools and tool errors separately from valid answers that did not mention the company.
Follow the measurement methodology: keep retrieval, citation, factual accuracy and qualified enquiries as separate outcomes.
Published evidence
Illustrative buyer questions
Frequently asked questions
Does blocking ClaudeBot also block Claude search?
Does Anthropic publish crawler IPs?
Does llms.txt make Claude cite a website?
When people ask AI what to buy, be the beauty name it recommends.
Tell us what you sell, make or supply. We test frontier AI engines with real shopper and buyer prompts and send you the raw, unedited answers within 3 business days.
微信扫一扫 · 领诊断报告
美链数智-Joy
💡 中国源头工厂、生物活性原料商（重组胶原蛋白/多肽）、检测认证实验室与出海品牌，可直接用微信扫左侧二维码或右侧表格直接填【中文公司名 + 微信号/手机号】免费申领诊断报告。
- 4 Frontier AI Platforms Tested3–5 representative buyer prompts tested on ChatGPT, Perplexity, Claude, and Gemini.
- 100% Unedited Raw CitationsReceive actual unedited AI response text and see which competitors are currently recommended.
- Commercial Privacy & Data ProtectionClient confidentiality strictly protected. Mutual NDA available upon enterprise request prior to data exchange.