Schema, llms.txt, and What Actually Makes You Citable

Two technical files get sold as AI search strategy, and only one of them does anything. The evidence on llms.txt is unusually clear: SE Ranking's November 2025 analysis of roughly 300,000 domains found no relationship between having an llms.txt file and how often a domain is cited by major LLMs, and Google's John Mueller has said flatly that "no AI system currently uses llms.txt". Schema markup is a different story — it genuinely helps, but not in the way most people implement it. Schema is not a ranking lever you pull; it is a disambiguation tool that only works when it describes facts the rest of the web can independently confirm. This article separates what is measurably useless from what is worth your afternoon.

Does llms.txt make you citable?

No, and the evidence against it is stronger than the evidence for almost anything else in this field.

SE Ranking studied roughly 300,000 domains in November 2025 and found the presence of an llms.txt file predicted nothing about citation frequency. Their own write-up is blunt about it. Google has said the same directly: Mueller compared the idea to the old keywords meta tag — a self-declared signal that is trivially gamed and therefore ignored — and Gary Illyes confirmed Google has no plans to support it. Google's position is that Search simply ignores the file.

Should you delete yours? Not necessarily. It costs nothing to host, and coding agents like Cursor and Claude Code genuinely do read it, which makes it useful developer-facing hygiene. Just do not count it as an AI search strategy, do not measure it, and do not pay anyone who sells it as one.

What is schema markup actually for?

Schema markup is structured data that tells an engine what your page *means* rather than what it says. Its job is disambiguation: turning "some text about a company" into "an Organization, named X, located at Y, which is the same entity as these profiles."

That distinction explains why most schema implementations underperform. Schema does not make a claim more true or more prominent. It makes an entity easier to resolve. Engines that lean on entity resolution — Perplexity and Gemini especially — reward it. Engines that lean on content extraction reward it less.

The corollary is the part people miss: schema that asserts things the open web cannot corroborate does nothing. Marking yourself up as the best provider in your city does not make an assistant say so. Marking yourself up with a sameAs array that points at three real, live profiles does help, because the engine can go check.

Which schema types matter for AI citation?

Five carry most of the weight. Google maintains the canonical structured data documentation if you want the full specification.

FAQPage is the most commonly missed easy win. If you have a visible FAQ section and it is not marked up, you are leaving verbatim citations on the table — assistants lift those answers almost word for word.

Why is an empty sameAs array a wasted opportunity?

Because sameAs is the one place you get to hand the engine proof, and an empty array hands it nothing. The `sameAs` property exists to link your entity to other authoritative pages describing the same entity — your LinkedIn company page, your X account, your Crunchbase profile, your directory listings.

An engine reading a populated sameAs can follow those links, find the same business described consistently by parties other than you, and resolve the entity with confidence. An engine reading sameAs: [] learns only that you declined to offer evidence.

This is one of the most common gaps we find, and one of the cheapest to close. It is roughly an hour of work, and it directly attacks the entity signal that Perplexity and Gemini weight most heavily.

One rule: only list profiles that actually exist and are live. Pointing sameAs at a dead URL or the wrong account is worse than leaving it empty, because you have now introduced a contradiction into the exact signal you were trying to strengthen.

Can schema hurt you?

Yes, in two specific ways.

Contradiction. Many WordPress sites emit several competing schema blocks — one from the theme, one from an SEO plugin, one from a reviews plugin — which disagree about what the page is or who published it. Ambiguous structured data is often worse than none, because it makes the entity harder to resolve rather than easier. Pick one source of truth and disable the rest.

Unverifiable claims. Schema asserting reviews, ratings, or awards that exist nowhere else on the web adds no citation value and creates a mismatch between what you declare and what the web confirms. Structured data should describe things that are true and checkable elsewhere.

What should you implement first?

Do the entity work before the content markup, because the entity work is what the weakest link in your profile actually needs.

For where this sits in the bigger picture, read the complete guide to AI search visibility, what AEO actually is, and how to track AI search visibility to check whether any of it landed.

Frequently Asked Questions

Should I delete my llms.txt file?

No need. It costs nothing to host and coding agents such as Cursor and Claude Code do read it, so it has value as developer-facing documentation. Just do not treat it as an AI search tactic — SE Ranking's November 2025 study of roughly 300,000 domains found it had no measurable effect on citation frequency, and Google has confirmed Search ignores it.

Does schema markup improve AI citations?

Yes, but as a disambiguation tool rather than a ranking lever. Schema helps engines resolve who you are — which matters most for entity-driven engines like Perplexity and Gemini. It does not make an unverifiable claim credible. Schema describing facts the open web can independently confirm helps; schema asserting things nothing else corroborates does not.

What is a sameAs array?

The sameAs property in schema.org is a list of URLs pointing to other authoritative pages that describe the same entity — your LinkedIn company page, X account, Crunchbase profile, or directory listings. It lets an engine follow those links and verify that independent sources describe your business the same way you do.

Is FAQPage schema still worth adding?

Yes, and it is one of the most commonly missed easy wins. AI assistants lift FAQ answers close to verbatim on FAQ-shaped queries. If you have a visible FAQ section without the corresponding markup, you are leaving citations on the table for roughly twenty minutes of work.

Can bad schema hurt my AI visibility?

It can. Multiple plugins emitting contradictory schema blocks make your entity harder to resolve rather than easier, and ambiguity is often worse than absence. Schema pointing at profiles that do not exist causes the same problem. Consolidate to a single, accurate, verifiable source of structured data.