llms.txt: the file everyone argues about. We ship it anyway.

The proposal is simple: place a Markdown file at /llms.txt that gives language models a curated map of your site — what you are, which pages matter, where the substance lives. Born in late 2024, it spread through 2025 the way robots.txt once did: no committee, just adoption. By 2026 it sits in an odd place — widely shipped, genuinely debated, and officially ignored by the biggest player.

The honest state of play

Let us not oversell it. Google’s John Mueller reiterated this August that Google does not use llms.txt — its systems read your actual pages, structured data, and the open web, and Google’s own 2026 documentation on generative AI features says plainly that optimising for AI search is still SEO. No major assistant has committed to honouring the file as a standard. Anyone selling llms.txt as a ranking lever is selling smoke, and we will not.

Why we ship it regardless

Three reasons, all modest and real. First, cost: it is one small text file — if it does nothing, you lost five minutes. Second, the landscape is wider than Google: an ecosystem of assistants, agents, and retrieval tools is growing fast, some of it reads curated context when offered, and the agentic web — software visiting sites on a person’s behalf — only expands the audience of machine readers. Third, discipline: writing the file forces you to state, in one page, what your site is and which pages carry the substance. If that exercise is hard, the problem is not the file.

What a good one contains

A one-line description of who you are; your core pages with a short annotation each; your best evidence — case studies, data, key articles; and languages and contact. Curation is the point: ten links with context beat a hundred without. It complements, never replaces, the fundamentals — clean structure, structured data, and content worth citing.

Practice what you publish

As of today, fib-consulting.com/llms.txt is live — and we shipped the same file on a client build in our healthcare work before we wrote this article. Our position in one sentence: ship it, don’t worship it — and spend the saved energy on the things all machines demonstrably read. If you want your site legible to every kind of reader, get in touch.

Sources

llmstxt.org (proposal, 2024) · John Mueller via ROI Revolution recap, August 2026 · Google, Optimizing your website for generative AI features on Google Search, 2026 · arXiv, A Critical Survey of Generative Engine Optimization, July 2026.