How Claude Chooses Which Sources and Brands to Cite (2026)

Claude picks sources in two ways. When it answers from memory, it uses what it learned in training and cites nothing. When it decides a question needs current information, it runs a web search, reads the results and cites the pages it used. Independent tests show those search results line up closely with Brave Search.

For brands, this makes Claude different from ChatGPT and Perplexity. Claude searches less often, so training data carries more weight. When it does search, ranking well in Brave’s results is the clearest lever, and the pages it cites lean heavily towards listicles, reviews and opinion content rather than forums.

Anthropic does not publish ranking factors, and nobody can promise a Claude mention. What you can do is make sure Claude can reach your pages, and that your brand appears in the kinds of pages it tends to cite. The 2025 and 2026 data below shows what those are.

• The short answer: what decides a Claude citation
• Training data vs live search: when Claude searches
• How Claude’s web search finds and cites pages
• How citations appear in Claude’s answers
• Anthropic’s crawlers, Brave and robots.txt
• Which sources Claude cites most
• Listicles, reviews and “best X” articles
• Freshness: does Claude prefer new content?
• Which brand signals matter for Claude mentions
• What a brand can do: a 9-step plan
• Limits: what no one can guarantee
• FAQs
• Get your brand into the sources Claude reads

The short answer: what decides a Claude citation

Here is the whole process in one table. Each row is covered in detail below.

Training data vs live search: when Claude searches

Claude can talk about your brand without searching at all. In that case it relies on what the model learned during training. There is no citation and no way to see which pages shaped the answer.

Live search is where citations come from. According to Anthropic’s help page on web search, Claude uses a search tool when a question would benefit from current information. In the newer Claude experience there is no on/off switch. Claude searches when it judges that it helps, and users can type “Search the web” to make it search. On Team and Enterprise plans, an owner has to enable web search first.

Claude searches less often than ChatGPT. In data from Profound presented at its Zero Click conference in June 2026 and reported by Search Engine Land, Claude searched on 36.6% of prompts, against roughly 90% for ChatGPT. The type of prompt made a big difference:

The practical split for brands:

Buying questions (“best project management tool for agencies”, “X vs Y”) are exactly the kind that trigger a search. That is why most of this guide is about the search path. But because Claude answers more questions from memory, long-term presence across the web matters more here than in ChatGPT.

How Claude’s web search finds and cites pages

1. Claude rewrites the prompt into search queries

When Claude searches, it writes its own queries. The Profound data reported by Search Engine Land in June 2026 found that 94% of Claude’s fan-out queries included the current year, against 17% for ChatGPT. Claude’s fan-out queries were also stable: the same query appeared about 65% of the time when a prompt was repeated.

For brands, that suggests a clear target. Pages that rank for “[category] [current year]” style searches are the ones Claude is most likely to retrieve.

2. Results come largely from Brave Search

Anthropic has not published how its search works. The evidence that Brave Search supplies most results comes from three directions:

• The subprocessor list. In March 2025, Simon Willison noticed that Anthropic had added Brave Search to its list of subprocessors as of March 19, 2025, the day before web search launched. A test search he ran returned ten citations that exactly matched Brave’s results.
• Early overlap tests. Profound tested three “best X” queries in March 2025 and found that 13 of Claude’s 15 citations (86.7%) matched Brave’s top non-sponsored results (Profound). That was a very small sample.
• A larger 2026 test. Profound repeated the test in June 2026 on about 400 queries and found 79.2% of Claude’s citations came from Brave’s top 10 (Search Engine Land). Notes from the same session suggest Claude takes Brave’s top results with little re-ranking, though others dispute that.

There is one open question. An analysis by Xponent21 (July 2026) reports that Brave was still on Anthropic’s subprocessor list, and that Anthropic’s Trust Center added turbopuffer, a search infrastructure company, as a second web search subprocessor on May 6, 2026. Its role has not been made public. Brave remains the best-supported explanation for where Claude’s results come from.

3. Google rankings still correlate

Brave and Google often agree on the best pages. In the same Profound data, 64% of the sites Claude cited also appeared in Google’s top 50 for the same query, against 37% for ChatGPT. Strong Google rankings help, but checking your position in Brave is the more direct test.

4. Claude reads, filters and cites

Claude does not cite everything it retrieves. On Anthropic’s developer platform, the web search and fetch tools now write and run code to filter results before they reach the model. Anthropic says this dynamic filtering, released with Claude Opus 4.6 and Sonnet 4.6 in February 2026, improved accuracy by 11% on average while using 24% fewer input tokens on its search benchmarks.

Anthropic has not said how the consumer Claude app decides which retrieved pages to cite. The safe assumption is the same as for other engines: pages that answer the query directly and clearly are more likely to be used.

5. Claude and ChatGPT cite different pages

Do not assume that being cited in ChatGPT means being cited in Claude. Profound found Claude and ChatGPT shared only about 8% of citations in its 2026 data. Each engine needs its own check.

How citations appear in Claude’s answers

According to Anthropic, answers that use web search include citations so users can check the sources. These come as direct citations, links to the sources and relevant quotes.

A few details matter for brands:

• Location affects results. Claude infers a user’s approximate location from their IP address to localise results.
• Users can fetch your page directly. If a user pastes a URL, Claude’s web fetch retrieves that page. This is how Claude reads a product page someone is already looking at.
• Image search uses Bing. Anthropic says image search in Claude is powered by Bing, even though text search results match Brave.

Web search launched in March 2025 as a preview for paid US users and became available on all plans worldwide on May 27, 2025.

Anthropic’s crawlers, Brave and robots.txt

This is the part you control. Anthropic lists three bots in its crawler help page.

Key facts from Anthropic’s documentation:

• The three settings are independent. You can block ClaudeBot for training and still allow Claude-SearchBot and Claude-User.
• Anthropic says blocking Claude-SearchBot “may reduce your site’s visibility and accuracy in user search results”. Blocking Claude-User “may reduce your site’s visibility for user-directed web search”.
• The bots support the non-standard Crawl-delay rule, and Anthropic says they do not bypass CAPTCHAs or other anti-circumvention measures.
• Anthropic says blocking by IP address may not work reliably. Use robots.txt, set for each subdomain. Anthropic publishes its bots’ IP ranges for verification.

Brave’s crawler

Because Claude’s results match Brave so closely, Brave’s crawler matters too. According to Brave’s help page, its crawler does not use a separate user agent. If a page is not crawlable by Googlebot, Brave will not crawl it either. Brave also says robots.txt alone does not stop a page from being indexed; a noindex tag does. Brave offers a form to request a re-crawl of a URL.

In short: if Googlebot can reach your pages, Brave usually can. If you block Googlebot from a section, assume Claude will not see it through Brave.

Many sites have never made a decision about AI bots. Our study of AI bots in SaaS robots.txt files found that 73% of top SaaS sites name no AI bot at all. You can check your own file with the free robots.txt AI bot checker on our tools page.

What about llms.txt? Anthropic’s crawler page does not mention it as a ranking signal. Adoption is growing anyway: 74% of 47 leading B2B SaaS sites serve an llms.txt file. Treat it as optional housekeeping, not a ranking lever.

Which sources Claude cites most

There is no public top-50 list of Claude’s most-cited domains like the ones Ahrefs publishes for other assistants. What exists is a set of studies on the types of sources Claude prefers. They point the same way.

Three things stand out:

• Claude barely cites forums. Reddit dominates citation lists for ChatGPT and Perplexity. In the Profound data, community sites made up less than 1% of Claude’s citations.
• Reviews matter a lot. In Yext’s data, the share of Claude citations from reviews and social sources ranged from 6.3% for organisations to 24.4% for food and beverage. In food and beverage, that was about one in four Claude citations, against about one in forty for Gemini. In business services it was 15.9%, more than double any other model.
• Hard news is less central. Muck Rack found Claude cited Reuters 20 times less often than Gemini and 50 times less often than ChatGPT. The outlets it did cite lean towards buying guides and business advice.

The lesson: for Claude, review sites and well-ranked comparison articles in your category are more important than forum threads or breaking news.

Listicles, reviews and “best X” articles

“Best X” articles are the most direct route to a brand mention in any AI engine. In Claude, the data says they are even more central. Listicles made up 36% of Claude’s citations in Profound’s 2026 data, and “best X” prompts with a time element triggered a search 81% of the time.

Which lists are most likely to be cited

We found no Claude-specific study of which listicles get cited and which do not. Based on how Claude’s search works, these traits should help:

1. The list ranks in Brave. About four in five Claude citations came from Brave’s top 10 in Profound’s test. A list that does not rank there is unlikely to be read.
2. The title carries the current year. Claude adds the year to almost all of its search queries, so “best X in 2026” lists match its searches.
3. The list is maintained. A dated title with outdated content is a weak source, and it will drop as soon as the year changes.
4. The list gives reasons. Claude cited opinion content almost twice as often as ChatGPT. Clear verdicts and “best for” lines give it something to quote.

So the goal is not “get on any list”. It is to be included, with accurate detail, in independent, current comparisons that rank for the searches Claude runs. That is what our brand mentions service focuses on.

Review sites

Yext’s data makes review platforms a priority for Claude, especially for local and service businesses. Keep profiles on Google Business Profile, Yelp, Trustpilot, G2 or the sites that matter in your category up to date, with recent reviews and accurate details.

Freshness: does Claude prefer new content?

The signals are mixed.

• Its searches are dated. 94% of Claude’s fan-out queries included the current year in the Profound data. That pushes results towards pages written or updated for this year.
• Its journalism citations are older. Muck Rack found that 36% of Claude’s citations to news articles were from the past 12 months, against 56% for ChatGPT.

One way to read this: Claude favours current buying guides and comparisons, while news coverage counts more for the brand’s long-term profile than for any single answer. We found no study that measures the age of all pages Claude cites.

In practice:

• Update comparison and pricing pages when facts change, not just the date.
• Make sure the lists that mention you are refreshed for the current year.
• Do not expect a new page to beat an established, well-ranked page just because it is new.

Which brand signals matter for Claude mentions

Citations are about pages. Mentions are about brands. We found no correlation study of brand signals for Claude specifically. The largest such study, by Ahrefs on 75,000 brands, covers ChatGPT and Google’s AI features; we summarise it in our guide to how ChatGPT chooses sources.

Two things make brand signals especially important for Claude:

• It answers more from memory. Claude searched on about a third of prompts in Profound’s data. For the rest, what it knows about your brand comes from training. A brand that is widely and consistently described across the web has a better chance of being remembered.
• It reads third-party verdicts. Listicles, reviews and opinion pieces make up a large share of what it cites when it does search.

This is why digital PR and earned mentions sit at the centre of AI visibility work. Links help the pages that mention you rank in Brave and Google, so link building supports the retrieval step. But the mention itself is what gets your brand named.

What a brand can do: a 9-step plan

1. Allow Claude-SearchBot and Claude-User. Check robots.txt for each subdomain. Decide separately whether to allow ClaudeBot for training.
2. Do not block Googlebot from key pages. Brave follows Googlebot’s rules, so a Googlebot block also hides pages from Brave.
3. Check your Brave rankings. Search your main buyer queries on Brave Search, with the current year added. This is the closest public view of what Claude retrieves.
4. Build a prompt list. Write 20 to 50 prompts a buyer would ask, focused on “best X”, rankings and comparisons, since those trigger searches most often.
5. Record who gets cited. Run the prompts in Claude, note the cited sources and named brands, and repeat monthly.
6. Get into the lists Claude already cites. Pitch the independent “best X [year]” articles from step 5 with accurate, useful information.
7. Strengthen review profiles. Collect recent reviews on the platforms that matter in your category. Claude leans on reviews more than other models.
8. Earn press and expert mentions. Original data gives writers a reason to name you. Data-driven content and digital PR build the long-term presence Claude draws on when it answers from memory.
9. Keep your story consistent. Use the same name, category and positioning on your site, review profiles, LinkedIn and directories.

Limits: what no one can guarantee

Be careful with anyone who promises Claude placements.

• Anthropic has not published how search ranking works. The Brave link rests on its subprocessor list and on outside testing, not on an Anthropic statement about ranking.
• The setup can change. A second web search subprocessor was reportedly added in May 2026. Results could shift without notice.
• Answers vary. Whether Claude searches at all depends on the prompt. Location and model also change the results.
• Training data is slow. Work that changes how the model itself describes you only shows up after future model updates.
• Most studies are small or correlational. The first Brave overlap test used 15 citations. The later ones are larger but come from a single vendor.

A sensible goal is to raise the share of relevant prompts where your brand appears, track it over months and compare it with competitors.

FAQs

Does Claude use Brave Search?

Anthropic lists Brave Search as a subprocessor, and independent tests show strong overlap. Profound found 79.2% of Claude’s citations came from Brave’s top 10 in a June 2026 test of about 400 queries. Anthropic has not explained its ranking, and a second web search subprocessor was reportedly added in May 2026.

If I block ClaudeBot, will Claude stop mentioning my brand?

No. ClaudeBot collects content that could be used for training. Claude-SearchBot and Claude-User control search and user-requested visits. You can block ClaudeBot and still allow the other two. Claude can also name your brand based on other sites that mention you, or on earlier training data.

Why doesn’t Claude search for every question?

Claude decides when current information would help. In Profound’s 2026 data, it searched on 36.6% of prompts. “Best X”, ranking and comparison prompts triggered searches most often. “What is” and “how does” questions rarely did, so Claude answered those from what it learned in training.

How is Claude different from ChatGPT for brand visibility?

The two cite very different pages. Profound found they shared only about 8% of citations. Claude barely cites Reddit and other forums, cites listicles and opinion content more often, and adds the current year to most searches. Results line up with Brave Search rather than the providers ChatGPT uses.

Does Claude cite Reddit?

Rarely. In Profound’s 2026 data, forums and community sites such as Reddit and Quora made up less than 1% of Claude’s citations, against about 16% for ChatGPT. For Claude, review sites and well-ranked comparison articles are a better use of effort than forum activity.

How can I track whether Claude mentions my brand?

Run a fixed list of buyer prompts every month, focused on “best X” and comparison questions, and log the cited sources and named brands. Check the same queries on Brave Search. Also watch referral traffic from claude.ai in your analytics, since search answers link to their sources.

Get your brand into the sources Claude reads

Claude names the brands it finds in current lists, reviews and opinion pieces, and the brands it already knows from training. Create & Grow places brands in relevant “best X” articles, reviews and press pieces, and runs digital PR campaigns that earn mentions on sites AI assistants cite. Brand mention plans start at €1,500 a month.

We cannot guarantee that Claude will mention you, and nobody honestly can. We can show you which sources it cites for your category and get you into the ones that matter. Contact us to start with a short audit of your prompts.

This guide is part of our series on AI search sources, which also covers ChatGPT, Perplexity and Microsoft Copilot.

More guides in this series on how AI search picks sources: ChatGPT, Google AI Overviews, Gemini, Perplexity, Copilot.