Blog

How Claude Chooses Which Brands to Mention

Robert VîjaRobert Vîja
November 16, 20266 min read
How Claude Chooses Which Brands to Mention

Claude’s choice of brands comes down to one switch: did it search the web for this answer? Without search, it goes with the names it already knows well, which favors big, established brands. With search, it pulls names from the third-party pages it finds, and that’s when niche brands get a real shot.

You ask Claude to recommend a tool in your category. Three competitors show up. Not you. This piece covers how Claude decides which brands make the cut, and what you can do about it.

How Claude chooses which brands to mention

It depends on whether Claude searches. In the app, Claude decides on its own whether to run a web search, and it shows the sources it used. Through the API, it doesn’t search unless a developer switches the search tool on.

To see how often each path happens, we looked at 2,247 Claude answers collected between 17 October and 6 November 2026. They cover 131 prompts in four projects: two B2B SaaS brands with English prompts and two B2C brands with Romanian prompts. Every answer came from claude.ai with a signed-in account, and we ran the same prompts on the same days in other engines. Four projects and 131 prompts show a direction, not a study.

55.6% of those answers (1,250 of 2,247) cited no source at all. So more than half the time, Claude answered from what it already knew. We only store pages Claude actually cites, so treat 55.6% as an upper bound. By project it ranges from about 35% to 77%, highest on the Romanian B2C prompts.

It isn’t a simple on-off switch per prompt either. Of the 84 prompts with at least five answers, Claude never searched on 27, always searched on 9, and sometimes did on the other 48. On the same prompts, Gemini answered without sources 13.9% of the time. Perplexity almost never did: 0.1%.

Why does Claude mention some brands over others

Search decides a brand’s odds more than anything else. Without search, Claude goes with established names. With search, brands with a real footprint on review and comparison pages catch up fast.

One niche brand we track appeared in 13% of Claude’s answers without search and 64% with search. On exactly the same prompts, that’s 12.5% against 71.5%.

A second niche brand makes it sharper. Across 278 answers without search, it was never named. Across 82 answers with search, it appeared in 15.9%.

A very large brand shows up either way: 94% without search, 99% with it.

Brand Without search With search
Niche brand A 13% (12.5% on matched prompts) 64% (71.5% on matched prompts)
Niche brand B 0% of 278 answers 15.9% of 82 answers
Very large brand 94% 99%

Search matters most for brands Claude doesn’t already know well. It barely matters for brands it names anyway.

Bar chart: share of Claude answers naming the brand without vs with web search. Niche brand A 13% vs 64%, niche brand B 0% vs 15.9%, very large brand 94% vs 99%.

What determines brand mentions in AI generated responses

Claude doesn’t name fewer brands than ChatGPT, and it isn’t more concentrated on a few favorites. It’s the opposite.

Claude names 5.19 brands per answer on average, against 4.55 in ChatGPT (median 5 against 4). Across 1,032 matched prompt-day pairs, Claude named more brands in 56.5% of them and ChatGPT in 30.4%. The direction held in all four projects, even though Claude’s answers are shorter.

Metric Claude ChatGPT
Average brands per answer 5.19 4.55
Median brands per answer 5 4
Prompt-day pairs where it named more brands 56.5% 30.4%
Top 5 brands’ share of all mentions 54% 61%

The top five brands take 54% of mentions in Claude, against 61% in ChatGPT. In the two SaaS projects it’s 50% against 63%. If you want to see the contrast on your own prompts, run a ChatGPT visibility tracker next to your Claude data.

One exception: Romanian fashion. Claude rarely searches there, and falls back on large international retailers.

Does Claude use web citations to pick brands

When it searches, Claude uses a different set of sources than the other engines. According to public information, Claude’s web search runs on Brave Search: Anthropic added Brave Search to its subprocessor list in March 2025, as TechCrunch reported. Our data can’t prove that, but it fits.

What our data does show is how little Claude’s sources overlap with other engines on the same prompt. Only 11.6% of the domains Claude cites also appear in ChatGPT. Against Gemini, Google AI Mode and AI Overviews it’s 17% to 20%.

Claude also skips forums and social platforms. Out of about 5,500 citations, 4 point to a forum, and none to Reddit or YouTube. It leans on a long tail of smaller sites with comparison articles. That fits an index different from Google and Bing.

What makes a brand trustworthy to Claude

To Claude, trust rarely means your own website. When it searches, it almost never cites the brand’s own site.

For a niche brand we track, Claude’s searching answers cited the brand’s own site 1.4% of the time. ChatGPT did it 48.8% of the time on the same prompts. For a large brand: 7.9% in Claude, 97% in ChatGPT.

That’s close to opposite behavior. ChatGPT treats the brand’s pages as a primary source. Claude turns to comparison articles and review sites to decide whether a brand belongs in the answer. How ChatGPT picks its sources is in how to rank in ChatGPT.

So for Claude, your own site alone won’t move much. Your visibility depends on the independent sites writing about you.

Bar chart: answers with web search that cite the brand's own site. Niche brand: Claude 1.4% vs ChatGPT 48.8%. Large brand: Claude 7.9% vs ChatGPT 97%.

Common reasons Claude skips a brand entirely

A blocked bot is the most mechanical reason. According to Anthropic’s support documentation, it runs three bots:

  • ClaudeBot: collects content that may be used for training.
  • Claude-SearchBot: indexes content for search. Blocking it may reduce your visibility in search answers.
  • Claude-User: fetches a page when a user asks Claude about it.

We checked 343 project sites with a readable robots.txt. 12 (3.5%) block ClaudeBot completely; 13 block GPTBot. None block Claude-SearchBot or Claude-User. ClaudeBot is named in 51 files, Claude-SearchBot in only 7. Most site owners who thought about Claude thought about training, not about the bot behind search answers.

One caveat: this only covers robots.txt. Blocks at the firewall or CDN level we don’t measure yet, so the real block rate may be higher.

Three cards: ClaudeBot (training, blocked by 12 of 343 sites checked), Claude-SearchBot (search index, blocked by 0 of 343) and Claude-User (user requests, blocked by 0 of 343).

How to get Claude to recommend your brand

Start by watching what Claude already cites. Don’t guess.

  1. Run your buyers’ real questions in Claude and note which pages it cites for your category.
  2. Pick out the comparison articles, review sites and roundups among them. Claude leans on those far more than on brand sites.
  3. Work to get your brand into those pages.
  4. Check robots.txt and your firewall first: Claude-SearchBot and Claude-User shouldn’t be blocked.

The gaps between Claude and ChatGPT are big once you line up real categories (anonymised):

Category Brand ChatGPT Claude
Flight compensation (Romania) Service A 0.6% 39.2%
Flight compensation (Romania) Service B 35.8% 11.9%
B2B software Platform A 37.9% 7.4%
B2B software Platform B 22.3% 62.1%
Online fashion (Romania) Three local shops 12% to 17% each 0%
Online fashion (Romania) International retailer 0% 25%

A brand that’s strong in ChatGPT and invisible in Claude isn’t necessarily doing anything wrong. It’s missing from what Claude’s search finds, or Claude doesn’t search for that type of question.

How to track your brand’s mentions inside Claude

Spot-checks won’t show a trend. Claude’s answers move with search behavior and with the day you ask.

The data in this piece comes from claude.ai with a signed-in account, not the API. So it captures the same web search and citations a real person sees.

In GEOflux, Claude is available on the Starter and Growth plans, including the 14-day trial, and in agency pitch reports. It isn’t on the Free plan. Each Claude answer costs 3 credits, against 1 for ChatGPT, Gemini and Google AI Mode. Claude is one of 8 engines you can track, and you pick at least 3.

Do this week: pick five prompts your buyers would type into Claude, run them, and write down which third-party pages get cited. That’s your baseline. If you’d rather have it run on a schedule, our Claude visibility tracker does exactly that.

Share
Robert Vîja

Robert Vîja, Co-founder & CPO @geoflux.ai

Runs product at GEOflux - what gets tracked, how citations are measured, what ships next. Ten years in SEO, on projects above 1M organic visits a month.

Ready to see your AI visibility?

Create a free account and discover how AI perceives your brand in ChatGPT, Gemini and Google AI Mode. No card, no time limit.

Start for free