Does Schema Markup or llms.txt Actually Get You Cited by AI Search?

No. Schema markup and llms.txt make your site easier for an AI system to read. Neither one makes that system trust you enough to cite you. If you run an SME site and you've added both this year hoping to show up in ChatGPT or Perplexity answers, this is for you: what these two things actually do, why they get sold as more than that, and what the citation decision is really made on.

What does schema markup actually do for AI search?

Schema markup is a way of labelling your content so a machine can parse what it's looking at without guessing. It tells a crawler "this is a product, this is its price, this is a review" instead of leaving it to infer that from a wall of HTML. That's extraction. It is not evaluation.

An AI system that can parse your page cleanly is not the same as an AI system that has decided your page is worth repeating. Schema removes friction from the reading. It does nothing to the judgement that comes after.

I've lost count of the SME audits I've run through the Search Visibility Snapshot where schema was missing, broken, or duplicated three times across the same template. Fixing it is always worth doing. In every single one of those audits, adding correct schema on its own never once produced a citation that wasn't already earned some other way. It just made the earning visible faster.

What does llms.txt actually do?

llms.txt is a plain text file at your site's root that tells an AI crawler what you consider your most important pages. It's a signpost, not a credential. Think of it as robots.txt's polite cousin: it points, it doesn't vouch.

No major AI system has confirmed that llms.txt affects ranking or citation likelihood. What it does is save a crawler some guesswork about where your best content lives, on the sites that choose to read it at all. That's a real, if modest, convenience. It is not a trust signal, because nothing about a text file you wrote yourself can be independently verified by the system reading it.

Why do people think these two things drive citations?

They get pitched as the mechanism because they're the part of AI visibility a developer can ship in an afternoon. Everything else, the years of being mentioned favourably by people who aren't you, is slower and harder to point at, so it gets skipped in the pitch.

There's a reason "add this file and you'll show up in ChatGPT" is such an easy sell. It's the same reason a single-food diet plan sells better than "eat less, move more, be patient for six months." The brain prefers one lever it can pull to a slow accumulation of signals it can't fully see happening. Schema and llms.txt are levers. Trust isn't.

If not schema, what actually gets a brand cited?

Citation is a downstream reward for being mentioned favourably, consistently, and independently across places an AI system already trusts: review sites, forums, press, other people's content about you rather than your content about yourself. That's the raw material a retrieval system is drawing on when it decides whose sentence to repeat.

Schema and llms.txt affect whether the system can read what's already there cleanly. They have no influence on whether what's there is any good, or whether anyone independent has said so. A perfectly marked-up page with no third-party trust behind it is just a well-organised nobody. This is the retrieval layer of the wider Search Visibility Framework.

Doesn't Google say structured data helps, though?

It's a fair objection. Google's own documentation is clear that schema can produce rich results, and some AI Overviews clearly favour pages with clean structured data when they're selecting a snippet to pull from. That's real and worth doing.

But rich result eligibility and AI citation are different prizes. One is about how a single Google surface displays your existing ranking. The other is about whether a generative system, working from a much wider and messier set of trust signals, decides your brand belongs in its answer at all. Winning the first doesn't book you a seat at the second. Conflating them is where most of the "schema gets you cited" advice quietly falls apart.

So what should you actually do this month?

Fix your schema with a free schema markup generator, because it's cheap, it's correct practice, and it removes a genuine barrier to being read properly. Then generate an llms.txt file for the same reason: low cost, plausible small upside, no downside. Then stop treating either as the strategy and start on the thing that actually moves citation, which is getting mentioned by people who aren't on your payroll.

That means answering questions properly in the communities your buyers already use, being the source journalists and reviewers quote, and building a clear entity definition so that when someone does mention you, the AI system can tell it's the same brand every time. None of that ships in an afternoon. All of it is the actual mechanism.

Call this gap what it is: the Extraction Gap. It's the space between a page a machine can parse and a source a machine is willing to cite, and technical fixes only ever close the first half of it.

The objection that's actually true

The honest counterargument here isn't "schema doesn't matter." It does. A site an AI crawler can't parse is genuinely invisible, no matter how well trusted it is elsewhere. Getting the technical layer right is a precondition, not a nice-to-have.

The mistake is treating a precondition as a cause. Clean plumbing doesn't make a restaurant good. It just means the good food, once it exists, reaches the table.

Before the algorithm changed, a citation was still a form of trust extended by one party to another. Structured data didn't earn that then and a text file doesn't earn it now. What earns it is still, mostly, other people.

Common questions