Claude SEO: How to Get Your Brand Mentioned in Claude

Claude answers from its training by default and searches only when a request depends on information that is current, changing, or outside its training data, per Anthropic's own documentation. Stable facts, math and coding concepts get a direct answer with no sources. So you can only win a citation on prompts that make Claude go and look.

When Claude does search, it shows its sources. Anthropic's web search tool documentation says citations are always enabled, and each one carries a URL, a title and up to 150 characters of the cited text. Simple questions use one to three searches. Comparison or multi-company questions can use ten or more.

Research mode goes further. Anthropic's Research help page says it runs multiple searches that build on each other and decides what to look into next from what it finds. It needs web search turned on and a paid plan.

The decision is made one prompt at a time, and it is not perfectly consistent. BrightEdge tested three prompts in April 2025. A purchase question and a local restaurant question triggered a search. A question about the meaning of a degree did not. That sample is tiny, so treat it as an illustration only.

Claude search runs in six steps: Claude decides to search, writes queries, gets results, opens pages, filters them and answers with citations. Anthropic's bots touch two of those steps. Claude-SearchBot helps search results, and Claude-User fetches your page when a person's question needs it. Your brand is cited only if your page survives all four gates.

Step 1 to 2. Claude judges whether the prompt needs fresh facts. If so, it rewrites the question into search-friendly queries. BrightEdge saw it turn a shopping question into "best RC cars in 2025". Simple questions use one to three searches. Comparisons use ten or more.

Step 3 to 4. Results come back from a search index. Analysts say it is Brave, and Anthropic does not confirm that. Anthropic says Claude-SearchBot analyzes online content to improve search results, and Claude-User fetches a website when a user asks a question that requires it. Block either and you may lose visibility.

Step 5 to 6. On newer models, Claude writes code that filters results before they reach its context window. It then answers, and each citation carries a URL, a title and up to 150 characters of cited text. Your brand is mentioned when your page is named in the answer, and cited when your link appears with it.

There is a second path. When Claude answers itself, no bot visits at that moment. ClaudeBot collected public content earlier for training, so your page can shape the answer without ever being linked.

Query fan-out means Claude splits one prompt into several searches, each aimed at a different angle, then reads pages that match more than one. Anthropic says simple questions use one to three searches and comparisons use ten or more. Analysts report the searches run on Brave Search, so your page must be found by several sub-queries to win.

The sub-queries in the diagram are our illustration, built from patterns BrightEdge saw, such as a shopping question rewritten into "best RC cars in 2025". Each sub-query asks something different: recency, price, reviews, fit. A page that answers only one angle gets one chance. A page that answers three gets three.

The Brave link has the best public evidence we found. Searchable analyzed about 95,000 search-result records from 3,000 Claude prompts in July 2026. Pages Claude read matched Brave's top ten 63% of the time, against 34% for Google's top ten, on the same queries. Searchable infers Brave from that overlap. Anthropic has not named a provider, and Searchable sells AI visibility tools.

Once Claude reads a page, rank matters little. Searchable saw pages cited 39.4% of the time at ranks one to three, 37.2% at four to six and 36.8% at seven to ten. The hard part is getting into the top ten for at least one sub-query. Search your category phrases on Brave Search and see who shows up.

You can win prompts where Claude has to look something up: best-of lists, comparisons, current prices and time-bound questions. You cannot win prompts it answers itself, such as definitions and boilerplate, by being cited, because no search runs. For those, your lever is training data. This framework is our judgment, built from Anthropic's docs and public tests. It is not an Anthropic rule.

Winnable now. Anthropic says Claude searches for information that is current or changing. Buying questions fit that rule: "best", "vs", "pricing", "alternatives" and "in 2026". Honeyb's July 2026 test used 20 buyer-intent prompts and every Claude answer came back with sources, averaging 8.8 per answer.

Mixed. How-to questions and "what is [brand]?" sit in the middle. Claude may answer from memory, or search if the topic looks fresh. We cannot predict which, so cover both fronts: a clear page and a consistent public description of your brand.

Claude answers itself. Definitions, explanations of established ideas and boilerplate code get a direct answer. Being cited is off the table here. The only way in is to be part of what Claude learned, which is a slow bet. Anthropic says its training crawler, ClaudeBot, collects public web content for that purpose.

Peec AI's Tomek Rudzki makes a related point about tracking: if you are invisible at the awareness stage, you miss the largest pool of buyers. Map your prompts by lane before you write a single page.

Most of what AI engines cite is earned on other sites. 84.1% of citations go to third-party authority and under 0.3% to sponsored content. Claude leans harder on press and review pages than ChatGPT does. This is ThriveStack citedby research, including our September 2026 study of about 250K citations.

The data behind it is public. Read the full ThriveStack AI visibility benchmarks, covering 100 SaaS brands and 380,000+ cited URLs across five engines including Claude, and the ThriveStack AI citation sources study.

Two more ThriveStack findings shape the plan. The top five brands in a category take 68.2% of citations, and only 16.4% of AI citation sources also sit in Google's top ten. A page that ranks well in Google is not enough, and your own site alone will not carry you. Those figures cover all engines together. We do not split them out for Claude, so treat the Claude-specific points below as directional.

Claude has its own habits. Searchable's vendor data shows Claude citing national press far more than ChatGPT does and leaning on review sites and comparison blogs. The next sections cover that in detail. A win on one engine does not carry over to the other.

Claude's memory shapes which brands it names before any search starts. A July 2026 geoSurge study found brands a model remembered were searched 55.7% of the time, against 17.4% for brands it did not remember. That is an association and proves nothing alone. Build the memory, then build the pages.

The study tracked 66 buyer prompts in nine industries, run 60 times each. It measured search on Gemini 3.5 Flash, so Claude was untested, and brand fame may explain part of the gap. Only 31% of fan-out queries named a brand, and 63% of those named one of the model's top five remembered brands. See the geoSurge report.

Claude data points the same way. Searchable found that 86% of the brands Claude proposed for unbranded questions appeared in the final answer. If Claude recalls you, you start inside the answer.

Recall also fails in odd ways. DEJAN's summary of a Google Research study says frontier models store about 98% of facts but recall only about 66% when asked directly. The test used Wikipedia facts and had no brand data. Niche details, such as pricing, integrations and leadership, are the most exposed. Thinking modes recovered 40 to 65% of the missed facts.

So say what you are, in both directions. Write that your product is a leading tool for a category, and that the category includes it. Use the same one-line description on your site, your profiles and your press. Allowing ClaudeBot keeps your new pages eligible for future training, though Anthropic does not say how it uses them.

Claude leans on outside sources more than ChatGPT does. In Searchable's July 2026 data, Claude cited 52.0% of the national press pages it read, against 19.2% for ChatGPT. Brands' own sites were used 31.6% of the time on Claude and 60.9% on ChatGPT. This is a vendor's data, so treat it as indicative.

Claude also spreads its picks wider. No single site supplied more than 9% of Claude's cited sources, while TechRadar alone supplied 22% of ChatGPT's. On the 5W index, Claude leans toward The New York Times, The Atlantic, The New Yorker and The Economist, and 36% of its journalism citations were under 12 months old, against 56% for ChatGPT. 5W is a PR agency, and its release gives no per-platform sample sizes.

The studies do not agree on everything. Oltre found none of five big outlets, including Forbes and the New York Times, among 2,170 cited URLs, while Searchable found national press cited at 52.0%. The mix depends on the prompts you test. ThriveStack's own data agrees on the bigger picture: earned authority beats owned pages.

Reddit is where they split most. Searchable saw forums cited less on Claude than on ChatGPT. Amicited reports Reddit at 6.6% of Claude's citations against 1.8% for ChatGPT, with no method given. We would not plan around Reddit for Claude.

For Claude, rank earned media like this: national press, review sites, comparison blogs, directories, then Reddit. In Searchable's data Claude cited 52.0% of national press pages it read and 49.1% of review site pages, but only 8.3% of Reddit and forum pages. For local searches, listings come first: 97.7% of local prompts pulled business-listing services.

National press. It has the biggest gap between the two engines. Pitch data and original research. Skip product news. Test it in your niche, because Oltre found no big outlets among its cited URLs for SaaS prompts.

Review sites. They work on both engines, 49.1% on Claude and 44.8% on ChatGPT. Claim your profiles, fill every field and collect fresh reviews. Amicited lists Gartner and Yelp among Claude's cited sites, and G2 for ChatGPT, with no method given.

Comparison blogs. Claude cited 42.6% of the ones it read. Get into vendor roundups and "best of" lists in your category.

Directories. About 30% on both engines. They matter most for local businesses.

Reddit and forums. Lowest priority for Claude. Take part because your buyers do, and expect few direct citations. The overlap below is where a mention comes from.

Anthropic runs three bots. ClaudeBot collects public content for training. Claude-SearchBot analyzes content to improve search results. Claude-User fetches a page when a person's question needs it. Anthropic says blocking the last two may reduce your visibility in Claude's search answers, so most brands should allow them.

The choice on ClaudeBot is different. Anthropic says blocking it signals that your future content should be excluded from model training. That is a business call. It also affects the prompts Claude answers itself, which is where training data is the only lever. We read that as our inference, because Anthropic does not describe a visibility effect for ClaudeBot.

Check your robots.txt for AI for lines like the ones below. If you find them, you are opted out. Anthropic recommends robots.txt over IP blocking, which it says may not work reliably, and it asks you to repeat the rule for each subdomain.

Your server logs show which bots visit. Anthropic's crawler help page lists all three names. Interest in the topic spiked, too: searches for "claudebot" hit 33,100 a month in February 2026, per DataForSEO, then fell to 880 by September.

Several analyses say Claude's search runs on Brave Search, but Anthropic's own pages do not name a provider. Treat Brave as a strong lead that Anthropic has not confirmed. The cheap move is to check whether your best pages appear in Brave Search. If they do not, fix crawling and indexing first.

The strongest public data point is small. A Profound analysis, covered in its Claude web search explainer, found that Claude's cited results overlapped Brave's top non-sponsored results by 86.7%. That is 13 of 15 results across three queries, so it proves little on its own. On a CRM query, only three of Brave's top five showed up.

The Signals team and Oltre both repeat the Brave claim and cite a March 2025 TechCrunch report. We could not retrieve that report ourselves, so we label it reported. The ladder below shows how much weight each claim in this article can carry.

Claude cites specific, deep pages that answer one question quickly. An Oltre test of 2,170 Claude-cited URLs found only 3% were homepages, 56% sat under /blog/, and 47% looked like list-style posts. The method is unpublished, so use it as a hint. Our ThriveStack citation sources study points the same way: specific pages with a clear job.

Build every page in layers. First it must be crawlable by the bots in the crawler section. Then it must be indexed. Then it needs something original: a number, a test or a position only you can publish. Last, other sites should say the same thing about you.

Cover several angles on each topic. Aleyda Solís, who runs Orainti, puts it this way in her AI search roadmap.

Put a 40 to 60 word answer right under each question heading, show a visible last-updated date, and add the year when the topic changes often. Oltre found 24% of cited URLs carried a year from 2024 to 2026. Our AI citation sources research covers the same pattern across engines.

Interest in Claude SEO grew about 50 times in seven months. DataForSEO shows US searches for "claude seo" rising from 20 a month in December 2025 to 1,000 in July 2026, and it held at 1,000 in August. The phrase has low difficulty, a score of 8, and a high cost per click at $16.82.

"Claude web search" moved the other way after a spring peak of 480 searches a month. It sat at 320 in July and August. People now ask how to win in Claude, as well as how its search works.

Claude is still a small channel next to Google. BrightEdge itself calls it not yet a high-volume source. The window to build an early habit is open, and few teams have claimed it.

Run each prompt several times and report the share of runs that mention you. Honeyb's July 2026 test ran 20 buyer prompts three times each on four engines. Claude changed its top pick between identical runs 28% of the time and kept 67% of its brand list. That was the steadiest of the four, and it still moved.

One run proves nothing, so track a set of prompts across the three lanes, and log which lane each belongs to. Winnable prompts should rise first. Mixed prompts follow. Prompts Claude answers itself move slowly, if at all.

ThriveStack citedby tracks prompts across ChatGPT, Gemini, Claude, Perplexity and Google AI, and ties citations to traffic. We sell it, so weigh that. See the pricing page and the AI visibility benchmarks. If clicks are hiding, our AI referral traffic guide shows how Claude visits arrive in analytics.

Use four weeks: open the door, map your prompts, publish answer pages, then earn outside mentions and measure. Start with the lane where Claude searches, because that is where a page can win in weeks. Leave the training-data lane for a longer bet.

Week one is access and indexing. Week two is the prompt map. Week three is two pages: a comparison and a best-for page, each with a 40 to 60 word answer, a date and one original number. Week four is outreach to the sites Claude reads, plus a repeat-run baseline.

Not sure where to begin? Match your situation below.