Guide
Is llms.txt worth it? The honest 2026 answer
Updated · July 7, 2026 — Joffrey
What is GEO?→You've probably read that llms.txt is useless. It's partly true — and worth knowing before you spend time on it. Here's what the 2026 data actually shows, the one place the file still matters, and above all what really makes a site citable by AI answer engines.
Is llms.txt actually worth anything?
Yes, but far less than it was originally sold as — and not in the way you'd think. On the promise “drop an llms.txt and AI will cite you”, the critics are right: in 2026 the answer-engine crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended) fetch your pages' HTML directly and almost never read the /llms.txt file.
Where it still helps: Perplexity sometimes retrieves it to prioritize, and above all coding assistants (Cursor, Copilot, agents) use it against technical docs. llms.txt has shifted from “SEO weapon” to “convenience for agents and developers”: useful, free to add, but it is not what gets you cited in a ChatGPT answer.
Why does everyone say llms.txt is useless?
Because the 2026 numbers are harsh:
- ~10% adoption only (SE Ranking study of 300,000 domains), eighteen months after launch.
- AI crawlers ignore it: across hundreds of millions of analyzed AI bot visits, a negligible share actually hit
/llms.txt. - No correlation measured between having an llms.txt and getting cited.
- Google put it in writing (May 2026): llms.txt is not needed for AI Overviews, AI Mode or any generative answer feature.
So it isn't a myth: the “llms.txt = citations” promise doesn't hold up.
Should you still add an llms.txt in 2026?
Yes, as long as you know why. llms.txt keeps two real uses. Perplexity is one of the few engines that retrieve it to prioritize pages. And above all it has become a developer standard: Stripe, Vercel, Cloudflare and Anthropic publish one so coding assistants generate correct code against their API instead of hallucinating.
Since the file costs almost nothing to produce and deploy, there's no reason to skip it — as long as you don't believe it replaces the real citability lever. Add it, yes; bet on it, no.
So what really makes a site citable?
What gets a site cited is the machine-readability of your pages — not a file at the root. Concretely:
- structured data (schema.org markup, especially question-answer pairs in JSON-LD) an engine can lift word for word;
- content that directly answers specific questions, with a clear heading structure;
- off-site authority (real mentions and links).
That's exactly what Google recommends instead of llms.txt: robots.txt, sitemap and structured data. Q&A markup is the most underused lever — it turns “readable” into “quotable”. Go deeper: how AI engines pick who to cite.
How do you cover the real lever — and llms.txt — in one go?
Citeable generates bothfrom a single crawl of your public pages: the Q&A schema.org markup (the citability lever Google favors) and a clean llms.txt (for Perplexity and agents). Because they come out of the same analysis, they tell the same story, with no contradiction.
You also get a before/after citability score, and the files stay regenerable for life for a one-time payment. Honest scope: nobody controls the answer engines, so no tool can guarantee a citation — Citeable guarantees your site is as readable and citable as possible, which is exactly where the game is played in 2026.
Sources
- SE Ranking — Why brands rely on llms.txt and why it doesn't workseranking.com
- Presenc AI — State of llms.txt 2026: adoption and practicepresenc.ai
- Google's 2026 guidance on llms.txt, explainedgetpassionfruit.com