What are llms.txt and schema, and do they help with AI citations?

The technical layer, ranked by what actually works

There's a lot of noise about the "machine-readable site." Let me rank the pieces by how much they actually move AI citations, strongest first, so you spend effort in the right order.

1. Server-side rendering (non-negotiable)

This is the floor everything stands on. AI crawlers fetch your HTML and read what's in the server's initial response. If your page is a client-side-rendered shell that hydrates after load, the crawler sees an empty container and leaves. The test is simple: View Source (not Inspect Element) on your page. If you don't see your full content and structured data in the raw HTML, neither does the engine — and nothing else in this list matters.

Expert Take — Jason Burns: I've watched businesses invest in schema and llms.txt while their pages rendered blank to crawlers. It's spending on the roof before the foundation exists. Confirm server-side rendering first, every time. It's the single most common reason good content gets zero citations.

2. Extractable answers (write for the lift)

Lead every important page with a self-contained 40–60 word answer to the question that page targets, in plain language, near the top. Support it with short, atomic paragraphs and question-form headings that mirror how people actually ask. Answer engines extract clean, direct answers and skip pages that bury the point. This is content structure, not markup — and it's one of the highest-leverage things you can do.

3. Schema.org structured data (real, measurable help)

Structured data helps engines understand and categorize your content, improving the odds you're selected as a source. The schema types that matter most:

• Organization and Person — define your business and your authors as clear entities, with a consistent @id referenced everywhere. This is what lets an engine resolve "you" to one unambiguous thing.
• FAQPage — the highest-leverage single type, because it packages question-answer pairs the engine can lift directly. Make the schema text match the visible page text exactly.
• Article — headline, author, dates, publisher for your content pages.
• Service / Product / LocalBusiness — structured facts about what you offer and where.

Define Organization and Person once, site-wide, and reference them by @id — never redefine them per page. That consistency is the entity signal.

4. Clean crawler access (robots.txt)

Explicitly allow the AI crawlers you want to reach you — GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, PerplexityBot, Perplexity-User, Googlebot, Google-Extended, Bingbot, and others — and reference your sitemap. Don't accidentally block the bots that build the indexes you want to be cited from (a stray rule copied from an old template is a common, silent killer).

5. llms.txt (publish it, but keep expectations honest)

llms.txt is an emerging standard: a plain-text file at your site root that gives AI systems a clean, curated index of your most important content. It costs almost nothing to publish and it's a tidy machine-readable map of your best pages.

The honest caveat: as of 2026, the major AI vendors have not committed to honoring llms.txt for retrieval, and independent analysis suggests little measurable effect on whether you get cited today. Treat it as a low-cost bet on an emerging standard — worth doing, but not a lever that moves citations now. Don't let it eat the attention that belongs to server-side rendering, extractable answers, and schema.

Expert Take — Jason Burns: I publish llms.txt for clients because it's cheap and it might matter later, but I'm honest that it's not what's getting them cited. The wins come from the boring fundamentals — SSR, a clean answer up top, accurate schema, one consistent identity. If someone's selling llms.txt as a magic citation lever, be skeptical.

The order of operations

Server-side rendering first. Then extractable answers. Then schema (Organization + Person + FAQPage). Then clean robots.txt. Then llms.txt as a low-cost extra. Do them in that order and you'll have a genuinely machine-readable site that engines can find, read, understand, and cite.