Free Check of What Claude Says About Your Brand

A Claude brand check answers three questions. Does Claude name you when a buyer asks about your category? Are the facts it states about you right? And when it searches the web, which pages does it quote?

The catch is that Claude answers in two modes. Anthropic's Claude.ai makes web search a user toggle, and as of October 2026 no default is published. With search off, Claude answers from what it learned in training. With search on, Claude decides per question whether to search, then quotes the pages it retrieved.

So one test is really two tests. Mixing them gives you an average of two different systems, which tells you nothing you can act on. For the bigger picture of ongoing monitoring, see tracking Claude's answers about your brand.

You need a Claude.ai account, a terminal, and 20 minutes.

1. Write three buyer questions. Use questions a prospect would ask, not your brand name alone. Say you run a dental clinic in Austin: "best clinic for Invisalign in Austin", "how much does teeth whitening cost in Austin", and "is [your clinic] good for kids".
2. Turn web search off. Open a new chat at claude.ai and switch off web search in the chat's tools menu. Ask each question in its own new chat. Copy every answer into a sheet, with the date and the model name shown in the app.
3. Turn web search on. Open fresh chats, switch web search on, and ask the same three questions word for word. Save each answer and every citation Claude shows.
4. Check your robots.txt. Open https://yourdomain.com/robots.txt and look for rules naming ClaudeBot, Claude-SearchBot or Claude-User, plus any blanket User-agent: * block.
5. Check the firewall. Request a key page with a Claude crawler name in the user-agent:

curl -I -A "Claude-SearchBot" https://yourdomain.com/pricing

A 200 is a pass. A 403, a 429 or a challenge page means your CDN may be turning the crawler away. This is a rough test, since real bot verification can also check IP ranges. 6. Check the raw HTML. Fetch the page without running JavaScript and search for a fact that matters:

curl -s https://yourdomain.com/pricing | grep -i "per month"

If the price only appears after JavaScript runs, Claude's crawlers may never see it.

The search-off answer

This comes from training data frozen at the model's knowledge cutoff. If it misses a product you launched last spring, the cause is the cutoff, not your site. You can't change this answer this quarter. The knowledge cutoff check shows which of your news falls after the line.

The search-on answer

Anthropic's web search documentation shows that Claude cites text spans, quoting up to 150 characters per citation, rather than whole pages. That makes Claude unusually easy to audit. For each citation, ask:

• Is the quoted sentence from your site? If it's wrong, find the page and the contradiction. The same fact stated differently on your pricing and about pages is a common cause.
• Is it from someone else's site? When engines quote other sites instead of yours, being named on those sites is the route into the answer. That's an association in the evidence, not a promised effect.
• Is a stated fact uncited? It may come from training data even with search on.

One caution: a citation is not proof of influence. Research on how language models cite (arXiv:2606.28358) repaired over 90% of missed citations by adjusting internal model components without changing the sources. Treat citations as a useful but noisy signal. The free AI citation checker covers this across engines.

Anthropic runs three agents, and each one is a separate robots.txt decision.

Blocking ClaudeBot is a training choice. It is a different token from Claude-SearchBot, so a site can opt out of training and still allow search crawling:

User-agent: ClaudeBot
Disallow: /

User-agent: Claude-SearchBot
Allow: /

Three access problems hide behind a clean robots.txt. A firewall can return 403 to a crawler that robots.txt allows. Anthropic warns that IP blocking impedes its ability to read your robots.txt file at all. And Vercel's crawler study found Claude's crawlers fetch JavaScript files but don't execute them, so client-rendered content may be invisible. For what blocking does and doesn't stop, read whether AI can still mention a brand that blocks crawlers.

Be plain about what three questions can show.

• It's a sample. Claude's answers vary between runs. One wrong answer is a lead to investigate, not a verdict.
• Search is discretionary. Even with the toggle on, Claude decides whether to search for each question.
• Apps differ from the API. Products built on Claude add their own system prompts, which steer when Claude searches and what it says.
• There is no rank. Claude doesn't publish a position for brands in answers, so any "Claude rank" is invented.
• Content tweaks have weak evidence. A 2026 critical survey of 45 GEO studies (arXiv:2607.14035) found no technique with a stable, cross-platform effect on discoverability. The fixes with solid evidence are access fixes: crawler permission, firewall reachability and server-rendered facts.

What you control is whether Claude can reach and read your pages, and whether your own facts agree. What Claude writes is up to Claude.

The free checker on this site runs up to three buyer questions, yours or written from your homepage, and scores the answers with the 99Visibility score. Its engine list follows the 99Visibility Free plan, so the hand test above stays the direct way to test Claude with search on and off. It's a snapshot, not continuous tracking.

If you need Claude tracked over time, 99Visibility's paid plans audit Claude alongside ChatGPT, Google AI Overviews, Gemini and Perplexity. You can start on the Free plan with no card. To compare engines by hand, the Perplexity brand check uses the same method.