On this page
What Claude does with a question
Claude can search the web during a conversation, and when it does it turns your question into one or more searches, reads the results, and writes an answer with sources.
Which search index it draws on is not something you can verify or influence directly, and it can change. What you can rely on is the shape of the result: pages that are indexed somewhere, readable without scripts, and written so a sentence can be lifted out cleanly.
One practical difference from classic search: an assistant answering a question does not show ten blue links and let the user choose. It picks a few sources and writes. Being the third-best answer on a topic can mean being absent from the response, so the standard is higher than a first-page ranking.
Let it read the page
Anthropic documents separate crawlers for different jobs, so look at your robots.txt and your CDN or firewall rules. A blanket block on unknown bots, or a bot-protection challenge, can lock out an assistant fetching a page for a user. Decide on purpose which ones you allow.
Then check that the content is in the HTML. If the answer only appears after JavaScript runs, many fetchers see an empty shell. Fast, plain pages with real headings win. The technical SEO checklist covers indexing, canonicals, and rendering, and the same fixes serve every search tool.
Does ChatGPT mention you yet?
Checks whether ChatGPT, Gemini and Perplexity mention your brand. Tracks it every week.
Write pages Claude can quote
Open each page with the direct answer in two or three sentences, then support it. Use headings that match real questions, put prices, specs, dates, and limits in plain text, and name the entity clearly: who you are, where you operate, and what you sell.
Cite your own sources by name when you rely on them, and keep claims you can stand behind. Models cross-check, and a page that says something the rest of the web contradicts tends to get skipped. Answer engine optimization goes through the page shapes that get quoted.
Outside mentions matter as much as your own pages. Honest reviews, comparisons that include you, and useful threads in communities all give a model more than one place to find the same fact.
Avoid hiding the answer behind a form, a modal, or a login. A page a fetcher can read in one request is a page that can be cited. Long preambles before the point work against you for the same reason, since a model scanning for the answer may take it from a page that gets there faster.
Make the brand easy to identify
A model that has never heard of you needs a quick way to understand who you are. Put a plain About page on the site that states what the business does, where it operates, who runs it, and when it started. Use the same description on your social profiles and directory listings.
Consistency is what lets a model connect mentions. If your product has three names across the web, or your address differs between listings, its picture of you is blurry. Pick one name, one description, and one set of core facts, and repeat them.
Then give people something to quote. Original data, a clear methodology note, a pricing table, a named author with credentials: these are the details a model treats as signs that a page was written by someone who knows the subject.
If you publish documentation or developer content, a clean Markdown-friendly structure helps tools that read it, including coding assistants. Short pages, stable URLs, and headings that say what each section covers make it easier for any model to find the relevant passage.
What to expect and how to measure
Expect slow, uneven results. Retrieval changes, pages are re-read on their own schedule, and a source that was cited last month may not be cited this month. A single run proves nothing, so repeat your prompts across several weeks.
Look at three things: how often you are named, whether the description is accurate, and whether the cited page is the one you want. If a different page of yours is cited than the one you built for the question, that is useful information about what Claude found easier to read.
Compare with other engines. If you are cited in one and not another, the cause is often the retrieval source, not your content, so do not rewrite pages to chase a single result.
Test it and be wary of “Claude SEO” tools
Searches for a Claude SEO skill or plugin are growing: people package SEO checklists as prompts or add-ons that run inside Claude. They can speed up an audit of your own pages. They do not change how Claude ranks you, and no plugin can.
The real test is simple. Write 15 to 30 prompts a buyer would ask, run them with web search on, and note which brands and sources appear. Track it monthly. Sprites can run buying prompts through ChatGPT, Gemini and Perplexity weekly and report which brands they name, which is a useful proxy while Claude-specific measurement stays thin. See also how to track your brand in ChatGPT.
Questions
Does Claude have its own search engine?
Claude has a web search feature that retrieves pages during a conversation. The index behind it is not something site owners can submit to, so the practical lever is being crawlable and quotable.
Is there a Claude SEO plugin that will rank me?
No. Plugins and skills labeled Claude SEO are checklists and automations that help you audit your own pages. They cannot change how Claude chooses sources.
Should I block Anthropic’s crawlers?
Only if you do not want to be read or cited. Blocking removes you from answers that fetch pages live. If you want visibility, allow them and review your robots.txt and firewall rules.
How do I check whether Claude mentions my brand?
Ask it the questions your buyers ask, with web search on, and record which brands and sources appear. Repeat on a schedule, because answers vary between runs.