If you've read about GEO anywhere, you've seen llms.txt pitched as the magic file that gets you into AI answers. We add it to client sites ourselves — and we think you deserve the honest version of what it does.
What it is
"llms.txt" is a proposed convention (September 2024, from Jeremy Howard of Answer.AI): a Markdown-formatted text file at your site root that gives language models a curated summary of who you are and where your important pages live. A companion "llms-full.txt" can carry the expanded version. Think of it as a reception desk for machines: name, one-line positioning, key offerings, key pages, contact.
What it is not
Here is the part the hype skips: as of today, no major AI engine has committed to reading llms.txt. It is not a ranking signal. There is no evidence that ChatGPT, Gemini or Perplexity consult it when composing answers. When we run citation tests, what engines actually quote is your HTML pages — titles, descriptions, visible body text. A fact that exists only in llms.txt effectively doesn't exist for AI retrieval.
That's why in our own 24-check audit framework, the llms.txt checks are explicitly marked as informational, non-scoring items. A site without one is not defective. A site with a perfect one and an unreadable homepage is still invisible.
Why we add it anyway
Three reasons, all modest:
- It's nearly free. One static text file, written once from facts you already have.
- The convention may gain adoption. If an engine starts reading it next year, sites that have it benefit retroactively. That's the "insurance" logic.
- Writing it is a useful discipline. Forcing yourself to state, in plain text, what you are / where you operate / what you offer with real numbers often exposes that your actual website doesn't say these things clearly either. Fix the website first.
The one rule if you add it
Every number in your llms.txt must also appear in your visible page content. We check this explicitly in audits: if your llms.txt claims "500+ clients" but no page on your site says so, you've put your best evidence in the one file machines don't read — and created an inconsistency for the day they do. The file should be a faithful index of your site, not a parallel reality.
Bottom line
Add llms.txt the way you'd add a "humans.txt" or a well-formed favicon: quickly, correctly, once. Then spend your real effort where retrieval actually happens — server-rendered pages, answer-shaped descriptions, hard numbers above the fold, and a genuine corpus of content. If a vendor's GEO proposal leads with llms.txt, ask what's in the other nine-tenths of the plan.