Why AI Cites Small Brands Over Big Ones

AI search does not rank pages the way Google does. When someone asks a buying question, the model builds an answer from whatever sources it can clearly tie to that exact question. Brand size is not a factor. Being the clearest, most consistently described answer to a specific need is. That shift is why small brands now beat giants in AI answers, and why Anthropic's newest tool matters more than it looks.

What did Anthropic just ship?

Anthropic released a browser tool that lets its AI use the web, and the twist, as The New Stack put it, is that it does not actually run a browser. There are no screenshots and no cursor moving around a rendered page. It reads the page's underlying structure and acts on a button or a field by name, not by where it sits on screen. Read that twice, because it is the whole game. The model does not care what your site looks like. It cares whether your site is legible.

Why does AI cite a small brand over a big one?

Because the small brand is easier to point at. Search Engine Land ran 20 buying questions through ChatGPT and Google AI Mode. H&M, Uniqlo and Victoria's Secret, three of the biggest names in fashion, showed up twice between them. Knix, a company a fraction of their size, showed up 29 times. Not because Knix did better SEO. Knix is known for specific things: heavy flow, leaks, postpartum, inclusive sizing. And it says the same thing everywhere the model looks, on its own pages, in reviews, in creator posts, in magazines. So when someone asks a real question, the model has something to grab. The big brands are known for selling a bit of everything, and nobody asks AI for that.

Does structured data (schema markup) get you cited by AI?

Not on its own, and this is the confusion worth clearing up. Knix did not win by publishing structured data. It won by being a specific, consistently described entity across many sources. Google states plainly that structured data is not required for its generative search features. Schema and semantic HTML help AI agents read and operate your site, which is a real and separate benefit, but being named in an answer is driven far more by clear positioning and consistent corroboration than by markup. It is also what moves your citation share, the metric that is quietly replacing share of voice.

What does being legible to AI actually mean?

It shows up on two surfaces that follow one rule. When AI answers a question, it grabs whoever is clearly tied to the exact thing that was asked. When an AI agent works your site, it grabs the structure. In both cases, big loses and pretty loses; specific wins and legible wins. For twenty years, being large and being polished was the strategy, and in front of a model both can turn into dead weight. The model starts from scratch every time someone asks. Your name counts for nothing if it cannot tie you to the thing they asked for, and your design counts for nothing if an agent cannot read the structure underneath it. Being huge won you Google. It does not win this.

How invisible are most brands right now?

More invisible than they think, and we have the numbers. Across the 109 companies we have measured with CitePulse, the median AI visibility score is 13 out of 100, and more than a third scored zero, meaning AI cited them on none of their buying questions. Plenty of those low scores belong to companies with strong brands and beautiful websites. The model still could not tie them to anything specific. If you want the method behind the number, here is how to measure AI visibility across ChatGPT, Perplexity, Gemini and Claude.

How do you get cited by AI?

Three moves, and none of them are about size.

• Be known for one specific thing, not everything. The model needs a hook, and "a bit of everything" is not a hook. "The postpartum one" is.
• Say the same thing everywhere the model looks. Your pages, third-party reviews, directories and press. When the story matches across sources, the model trusts it enough to repeat it.
• Make the structure readable, not just the surface. Semantic HTML, real headings, and content that exists before the JavaScript finishes loading. If your page is blank until four scripts run, an agent that reads structure sees blank. The same discipline is what makes your content extractable into an AI answer.

Do these, measure, then re-measure. Extraction is never a guarantee of a citation, but being unreadable is a guarantee of being skipped. The model is not impressed that you are big. It just needs to know what you are for, and it needs to be able to read you.