Claude SEO, when the goal is appearing in Claude's answers, starts with accessible pages that answer the buyer's question and support their claims. Allow Claude-SearchBot and Claude-User to reach your public content. Make product facts easy to verify, then measure citations separately from recommendations and enquiries. Training access is a separate decision. You can restrict ClaudeBot without treating every Anthropic crawler the same way.
This guide covers getting found in Claude, not using Claude Code to run SEO audits.
I'm Brenden House. I run Searchmaxxed, an SEO and AI SEO agency.
In summary
Claude visibility needs accessible evidence, relevant pages and measurement that distinguishes being quoted from being recommended.
- Give search and user-requested retrieval access to approved public pages; decide training permissions separately.
- Test an unbranded buying question as well as a direct URL. They test different things.
- Put specifications, integrations and eligibility conditions in readable text, with evidence beside the claim.
- Count mentions, citations and recommendations separately. Then connect actual visits to enquiries.
- Treat Claude as its own measurement surface, with Australian and US results kept apart.
What does ranking in Claude actually mean?
Ranking in Claude means appearing in the generated answer to a particular question. The useful distinction is whether Claude names your company, cites your website or recommends your product for that situation.
They mean different things.
| What happens | What it establishes | What it doesn't establish |
|---|---|---|
| Your brand is mentioned | The answer names you | That Claude recommends you |
| Claude cites your page | A source link points to your content | That the answer recommends your product |
| Your product is recommended | Claude presents it as a suitable option | That someone visited or bought |
| A visitor enquires | Your website produced an enquiry | That every earlier touchpoint is known |
A manufacturing specification can earn a citation while another supplier earns the recommendation, so read the surrounding sentence before celebrating the link itself.
Searchmaxxed's AI search optimisation work combines technical SEO, on-page content and consistent company facts. Suitable customers need to recognise the fit.
How does Claude web search find and cite sources?
Claude can search for current information, read the sources it finds and write an answer that lets the reader check its citations. It can also retrieve a specific URL supplied by the user, which is a different route to the page.
Anthropic's web search documentation describes the sequence: Claude decides to search, retrieves results and produces a cited response. It can search again.
Search is conditional. Anthropic distinguishes changing information, such as product details, from stable knowledge that Claude can answer without searching.
Check before measuring. An answer written without web search doesn't test whether your new integration page is discoverable through web search.
Does Claude use Brave, Google or Bing?
Brave is a documented search provider for Claude. Google Cloud's Claude documentation lists Brave Search among the third-party services Anthropic's web search can call.
Check Brave too. Confirm the search provider for the specific Claude product or application you test.
There are other distinctions. Anthropic's consumer help page says Claude's image results use Bing. A developer can also restrict API searches to selected domains.
The application matters. Record which Claude surface you tested rather than treating every answer carrying its name as the same experiment.
Is a citation proof that Claude agrees with your claim?
A citation identifies the source supporting part of an answer. Check the claim against its source.
The API documentation describes citations containing a URL and title, plus a cited-text field holding up to 150 characters from the source.
That's a field limit. It isn't an instruction to reduce every product explanation to 150 characters, nor evidence that Claude reads only that much.
At Searchmaxxed, we build answer-first pages with supporting evidence underneath, keeping any condition that changes the claim beside the claim itself.
Which Anthropic crawlers should your website allow?
Allow Claude-SearchBot and Claude-User on public pages you want available for search and user-requested retrieval. Your company can make its training-data decision independently from the decision to let customers retrieve approved public information through Claude search and questions.
Anthropic's crawler documentation distinguishes these agents:
| User agent | Documented purpose | Your decision |
|---|---|---|
ClaudeBot | Collects web content that could contribute to model training | Decide whether to permit that collection |
Claude-SearchBot | Analyses content to improve search results | Permit access to appropriate public search content |
Claude-User | Retrieves content in response to user questions | Permit access to appropriate public pages users may ask about |
Check each agent by name. A blanket block on AI crawlers can contradict your search objective even when the original intention was only to restrict training.
Anthropic says its bots honour robots.txt and do not bypass CAPTCHAs. Check public pages behind browser challenges.
Our technical SEO work covers crawl access, rendering and indexing faults that stop useful pages reaching searchers, including faults outside the website code. For this task, the check needs to reach beyond robots.txt into your CDN and firewall rules.
Ask your technical team to inspect:
- robots.txt on each relevant hostname, including documentation subdomains;
- bot-management settings that challenge requests before they reach your application;
- response bodies, not just a successful HTTP status;
- logs for the actual page request, including redirects and blocked responses;
- request authenticity against Anthropic's published crawler information, rather than the user-agent name alone.
Keep private material private. Search visibility is a reason to improve public evidence, not to expose customer records or internal procurement documents.
What pages can Claude read?
Claude's API web fetch tool retrieves web pages and PDF documents, but its documentation says it doesn't support dynamically rendered JavaScript websites. Browser-based tools are separate capabilities.
That distinction is explicit in Anthropic's web fetch documentation. Test whether Claude can retrieve the important text, even when the page looks complete in your browser.
Read the delivered text yourself. A product name, navigation menu and loading spinner aren't a product page.
| Page format or condition | What to check | Practical action |
|---|---|---|
| Public HTML page | The response contains the important facts | Keep specifications and answers in delivered HTML |
| JavaScript-dependent product page | The initial response omits the product details | Server-render or pre-render the important content |
| Public PDF | Text, version and source are clear | Provide a readable document and a useful HTML summary |
| Login-only documentation | The relevant facts require authentication | Publish an approved public summary where appropriate |
| CAPTCHA or firewall challenge | The requester gets a challenge instead of content | Review access rules without weakening private-area security |
| Image-only specification | Facts are visible only inside a graphic | Add the same approved facts as text |
For a manufacturer, put compatibility conditions beside the downloadable datasheet; for SaaS, explain integration limits outside the help centre that requires a login.
We build those product, industry and integration pages as part of our on-page work, with a useful next step for whoever clicks through.
How do you test whether Claude can read a page?
Give Claude the public URL and ask a question whose answer exists on that page. Check the supporting passage yourself.
Choose something specific. Use an integration limitation or approved specification, rather than asking whether Claude recognises the company name you supplied in your prompt.
Next, start a fresh conversation and ask the underlying buying question without your name or URL. Request web search and retain the answer with its sources.
The first test checks access. The second checks discovery and selection against competing sources, using a question that gives Claude no instruction to start with your own website. Access alone proves very little.
What should you publish to earn useful Claude citations?
Publish evidence for the actual requirement. A clear product page, supported comparison or technical explanation gives the answer something specific to cite.
Start with the decision. For an enterprise buyer, “works with your existing system” isn't enough when the integration only supports part of the workflow.
Here is an illustrative SaaS example, not a client result:
| Buyer question | Page evidence worth publishing | What Searchmaxxed would build or improve |
|---|---|---|
| “Which tools support our identity provider?” | Named integration, supported functions and configuration limits | Integration page linked from the product page |
| “Can Australian customer data remain in Australia?” | Approved hosting and processing details, with exceptions | Clear public data-handling explanation, reviewed by your team |
| “What changes when we roll this out internationally?” | Market availability, support coverage and deployment requirements | Industry or market pages where the differences warrant them |
| “When is this product a poor fit?” | Unsupported workflows and a precise scope boundary | Comparison and fit sections that answer directly |
A public-sector organisation might publish service eligibility and the exceptions that change an applicant's next step. Same principle: answer the consequential detail.
Use real authors where expertise matters, date evidence when its age changes the answer, and update the facts when the product changes.
Keep schema consistent with the page. Prioritise accessible content and supported claims over speculative Claude-specific schema or llms.txt changes.
Keep the foundations sound. Relevant titles, internal links, clear site structure and crawlable pages remain part of the work, rather than being replaced by an AI-only writing format.
Look beyond your own website too. Check which publications or technical sources appear in your actual category answers before deciding what external coverage to pursue.
Searchmaxxed's plans include link work, with digital PR in Agentic Search Engineering, chosen around the publications and sources relevant to your market.
Our guide to how modern models choose sources covers the broader selection problem. Compare Claude's sources with your missing evidence.
How do you measure Claude SEO without confusing activity with results?
Track repeated answers to the same buying questions, keeping records of mentions and citations alongside visits and enquiries rather than combining unlike results. Preserve the answers so your team can check any reported increase.
For Australia and the US, keep separate question sets where product availability, terminology or legal requirements differ. Don't average away the market you're trying to win.
Anthropic says Claude may use an IP-derived location for localised results; its API also supports a location parameter, which belongs in your test record.
For example, test 20 relevant questions three times per market. These are illustrative sampling choices, not platform rules.
Keep the setup consistent:
| Record | Why it matters |
|---|---|
| Exact question and market | Changed wording can change the answer |
| Date, model and product surface | A different setup is a different test |
| Search activity and settings | A non-search answer doesn't test live discovery |
| Brand mentioned and recommended | Being named isn't necessarily being selected |
| Cited URL and surrounding statement | Shows which page supports which claim |
| Incorrect or outdated details | Identifies the facts needing correction |
| Landing page, enquiry and sales feedback | Connects visibility with business outcomes |
Define citation rate as the share of valid test answers that cite your domain, counting an answer once even when it links repeatedly. Report failed requests separately.
Track recommendation rate separately. Also distinguish your own website citations from third-party citations supporting a recommendation of your brand.
A company can win the recommendation without winning the link.
Why shouldn't you trust the AI traffic chart on its own?
An AI-labelled traffic increase can include automated visits. Searchmaxxed found this in an enterprise B2B client account.
Window: August to October 2026, GA4 Data API.
A client's GA4 showed AI-assistant visits jump from 354 to 6,050 in a month. Almost none of it was real. About 5,700 came in three one-hour bursts: same 1920x1080 desktop Chrome, all new users, no referrer, about 20 seconds each, landing on two pages. Real AI traffic was 385 sessions against 354 the month before, roughly flat.
That was an AI-traffic measurement finding, not a Claude citation result, and shows why the traffic line needs checking before becoming a growth claim.
Inspect the actual visits. Compare traffic bursts with available server evidence, then keep bot requests separate from people visiting after reading an answer.
Our managed work includes Search Console and GA4 checks, competitor measurement and a monthly decision memo. Name Claude in the agreed scope.
What should you fix when Claude finds competitors but not you?
Find the failed stage first. A blocked page, missing product fact and misleading citation each need different work, so another article may leave the actual problem untouched.
Use this diagnosis before briefing the next release:
| Observation | What it suggests checking | Useful next action |
|---|---|---|
| Direct-URL test cannot retrieve the page | Access rules, response content or rendering | Fix the public page's accessibility |
| Direct URL works, but unbranded searches miss you | Discovery and relevance for the question | Check search results, internal links and page targeting |
| You appear, but the answer omits a key capability | Missing, unclear or conflicting evidence | Publish the capability with its conditions |
| Claude describes an obsolete feature | Stale cited source or inconsistent product pages | Update the source you control and correct the record elsewhere |
| Claude cites your page but recommends another product | Your content informs the answer without proving fit | Improve comparison and product-fit evidence |
| Citations rise, enquiries don't | Query relevance or the landing page's next step | Check whether the questions attract suitable customers |
These are diagnostic suggestions, not conclusions drawn from an unseen website. Keep the answer and source URL beside the fix so you can test the same question again.
Searchmaxxed handles implementation as well as the audit: pages, technical fixes, schema and internal links. A dashboard alone cannot change those pages.
For the OpenAI-specific access rules, use our separate guide to ranking in ChatGPT. Their crawler settings differ.
Is Searchmaxxed right for your Claude SEO work?
Searchmaxxed fits companies that need SEO and AI search improvements implemented on their website, not just a visibility dashboard. That includes product pages, technical fixes and evidence that helps buyers compare options.
The scope matters.
| Plan | Relevant work |
|---|---|
| Growth System | Ongoing page releases, technical SEO, schema, internal links and AI visibility tracking |
| Agentic Search Engineering | Adds comparison, industry and integration page families, digital PR and qualified link campaigns |
| Custom & Enterprise | Can include multiple markets, dedicated engineering, custom reporting and original first-party research |
Our managed plans include AEO and GEO. You own the site, repository, content and data, and speak directly with the people doing the work.
For a Claude brief, agree the exact questions, markets and testing surface before work starts. That prevents general AI tracking from being mistaken for Claude coverage.
| Searchmaxxed is a good fit if | Look elsewhere if |
|---|---|
| You need SEO and AI search improvements implemented on your website | You want strategy advice with no changes to your site |
| You need product pages, technical fixes and evidence to support buying decisions | Your team handles implementation and only needs monitoring software |
| You want customers to find your public product and service information | You need a private Claude application for confidential internal documents |
Frequently asked questions
Is Claude SEO the same as using Claude for SEO?
No. Using Claude for SEO means using it to assist SEO tasks. Appearing in Claude means getting your company or content included in its answers.
Can I block ClaudeBot and still allow Claude search access?
Yes. Anthropic documents ClaudeBot, Claude-SearchBot and Claude-User separately, with distinct purposes. Check the access rules too.
Does a Google ranking mean Claude will cite my page?
No. A Google position and a Claude citation are separate observations. Check Claude directly, then Brave.
Does Claude cite every page it fetches?
No. Retrieval and citation are different observations. Anthropic's API web search has citations enabled, while web fetch citations are optional and disabled by default.
Do I need to publish a PDF for Claude?
No. Claude's API fetch tool supports web pages as well as PDFs. Keep important facts readable.
Is SEO dead because Claude answers questions directly?
No. Accessible pages and useful search content still matter when an answer needs web sources. Keep measuring rankings and enquiries too.
How long will it take for a changed page to appear?
No fixed inclusion timetable is documented. Check the changed page's accessibility, then repeat the same discovery tests and record which version gets cited.
Bring the product or service you want recommended, the markets you serve and the questions that precede an enquiry. We'll work out what needs changing.