Do AI chatbots favor big brands?

AI chatbots cite and name big brands far more often than small ones. The evidence for that pattern is strong. The evidence for why it happens is not.

Two stages explain most of it:

• Parametric knowledge. With no web search, a model answers from its training data. Big brands appear in that data far more often, so the model knows more about them. Nothing you publish this quarter changes a model already trained.
• Retrieval. With search on, the engine pulls a small set of pages into its context. Those pages are concentrated among large, frequently cited sources.

No AI vendor documents a rule that boosts famous brands. The authors of a 252,000-trial study of AI citation (arXiv 2605.25517) anonymised brands on purpose and warned that real systems may favour established domains. So treat brand prominence as a strong association with no lever attached. The useful question is what still blocks you, regardless of size.

The clearest number comes from news citations. A study of AI Search Arena data (arXiv 2507.05301) found the top 20 most frequent news sources accounted for 67.3% of all citations for OpenAI models. The same study measured concentration with a Gini coefficient, where 1 means one source gets everything: 0.83 for OpenAI, 0.77 for Perplexity and 0.69 for Google.

Read it with two limits. The data window was March to May 2025, so it is old. It also covers news sources from a self-selected evaluation population, not commercial categories. Our digest of the news citing study covers the method in more detail.

Semrush's index of 126 million prompts (January to April 2026) points the same way. Only 36 brands held visibility across all four surfaces it tracked every month, including YouTube, Google, Reddit, Amazon, Facebook, Apple, Walmart, Disney and Nintendo. That finding is brand size. It is not a GEO technique anyone can copy.

What this means: if a big competitor shows up more than you, that alone tells you nothing about your site. It is the expected result.

Brand prominence is the dominant uncontrolled confounder in correlational GEO research. A study that finds cited pages share some trait usually can't rule out that the trait simply tracks bigger brands.

The usual fixes don't work either:

• Domain Authority and Domain Rating are vendor metrics built from backlinks and traffic. No AI engine computes them, and they are proxies for brand size itself.
• Google publishes no site-level authority score. There is nothing to build toward.
• E-E-A-T isn't a specific ranking factor, by Google's own statement. There is no E-E-A-T score to improve.

So advice to "build authority for AI" sells a number no engine uses. A critical survey of 45 GEO studies (arXiv 2607.14035) found no technique with a stable, cross-platform causal effect on discoverability.

The practical consequence is about measurement. Any benchmark that compares you with a household name is reporting brand size wearing a GEO label. Compare against brands of similar size in your category.

Concentration is not exclusion. Three conditions give smaller brands room.

Low-competition topics

Otterly's experiment with two new AI-generated sites about capybaras is a useful edge case. Otterly reported 3,934 tracked AI citations for one site, 2,718 of them from Microsoft Copilot, even after Google deindexed it. The topic was informational with little competition, and it was one pair of sites. It says little about commercial categories, and mass publishing got both sites removed from Google. Do not copy it.

Specific, answerable facts

Topical relevance to the query is one of the few factors that several studies agree on. A page that plainly states a price, a number or a one-line definition gives the engine a sentence it can quote.

Different engines, different indexes

In the same experiment, Copilot cited the sites heavily while ChatGPT and Google's AI surfaces barely did. Each engine draws on its own index, so measure each one separately.

You can't change your size this quarter. You can remove the blocks that keep you out of answers. Work through these in order:

1. Check crawler access per bot. robots.txt rules are independent per crawler. Per OpenAI's crawler page, GPTBot governs training and OAI-SearchBot governs ChatGPT search. Check PerplexityBot and Claude-SearchBot too.
2. Check your firewall or CDN. It can return 403 to a crawler that robots.txt allows. The AI Crawler Checker requests a page as each AI crawler and shows which ones are turned away.
3. Put key facts in the HTML. Vercel found ChatGPT and Claude crawlers fetch JavaScript but don't execute it. Pricing that appears only after JavaScript runs may be invisible to them.
4. Remove accidental suppression. Look for stray noindex, nosnippet, max-snippet:0 or data-nosnippet on pages you want quoted.
5. Fix contradictions on your own site. One price on the pricing page and another in the footer gives an engine two answers.
6. Get named where engines already look. When answers cite other sites, being mentioned there is associated with appearing. See how third-party mentions feed AI answers.

The AI SEO checklist covers each item, and the one-afternoon visibility check fits a small team. To see where you stand today, the free checker on this site shows whether engines name you for up to three buyer questions. None of this guarantees a citation. It removes the reasons you can't be cited at all. If you're weighing outside help, read whether a small business needs an AI search service.

Does domain authority matter for getting cited by AI? Not as a lever. Domain Authority is a Moz metric and Domain Rating is an Ahrefs metric. No AI engine computes either, and Google publishes no site-level authority score. Both are built from backlinks and traffic, which largely track brand size. Chasing those numbers doesn't address anything an engine documents using. Spend the effort on crawler access, rendering and clear facts instead.

Can a brand-new website get cited by AI chatbots? It can, especially on low-competition topics and on engines that draw from an index still holding the site. In Otterly's test, new sites were cited by Microsoft Copilot thousands of times. A model answering without search won't know a site launched after its training cutoff, though. As of OpenAI's model docs, GPT-5.6 has a February 16, 2026 cutoff.

Should I benchmark my AI visibility against the market leader? Only if you want to measure the size gap. Brand prominence drives so much of who gets named that a comparison with a household name tells you little about your own site. Pick competitors of similar size in your category. Then compare engine by engine, because one engine can cite a site often while another barely does.

Does allowing GPTBot help a small brand appear in ChatGPT? No. OpenAI documents GPTBot as its training crawler. ChatGPT search uses OAI-SearchBot, and the two settings are independent of each other. Allowing GPTBot has no documented retrieval benefit. If you want to be eligible for ChatGPT search, allow OAI-SearchBot and confirm your host or CDN doesn't block OpenAI's published search crawler IP addresses.