There is an llms.txt at this site’s root — why we put it there.
robots.txt speaks to the crawler; llms.txt speaks to the model. A single Markdown file at the root answers “what is this site, what does it offer?” in the plainest form a model can read.
It is not a standard yet, only a proposal — and it costs about as much as a text file. For an engine that ignores it, nothing changes; for an engine that reads it, you have described your own site — a statement, not a guess. There is no risk in publishing it on day one and no return in waiting.
Do not confuse it with robots.txt
The two answer different questions. robots.txt is a permission file: "what may you crawl". llms.txt is an introduction: "what is here, where should you look". One opens the door, the other describes the room. Being allowed to crawl a site is not the same as understanding it — this file fills that gap.
The robots.txt side still deserves care. This site welcomes sixteen answer-engine groups explicitly and blocks no path at all. "User-agent: *" already covered them; the groups were written out anyway, because some crawlers read only their own group, and putting intent in writing separates a decision from an oversight.
What goes inside
The format is plain: an H1 title, a one-paragraph summary, then linked sections. Each line gives a page by name, address and a single sentence. Not long marketing copy — short lines clear enough for a model to quote.
- # murattunalı.
- > Makinelerin de okuduğu web tasarımı —
- > İstanbul merkezli stüdyo: web tasarım +
- ## Sayfalar
- - [Hizmetler](https://…/hizmetler/): Üç hizmet, tek disiplin. Tasarım v…
- ## Doğrulanabilir gerçekler
- - Kurucu: Murat Tunalı — 20 yıl kurumsal BT altyapısı + web geliştirme.
An llms.txt without links is half a file
The first version of this site’s file was 247 bytes and contained NOT A SINGLE ADDRESS. It described the identity but offered no navigation: a model could answer "what is this site" but not "where should I look". The audit caught it and it was fixed — the index now carries 44 real addresses, the full directory 219.
More importantly, the file is not hand-written. It is generated from the route map, the published posts and the case files; when a new post lands, llms.txt grows by itself. A hand-maintained index starts lying at the first line someone forgets.
The last section: verifiable facts
The file ends with a short list of facts: languages, how it is built, the performance budget, the accessibility baseline, the contact address — and one more line: "no ranking guarantees are given". If a model ever summarises this site, the raw material for that summary sits right there, and all of it can be checked. No marketing adjectives; an adjective is the one thing that never survives being quoted.
Where the proof is
The terminal block on the home page is not a prettified copy of this file but its real OPENING: open /llms.txt and compare the first lines. The file grew, the block stayed — and the verification suite measures, before every release, that what is on screen is still the start of the file. If the two ever diverge, we see it first.
Do not leave it to the model’s guess: describe your site in your own sentences.