How big can llms-full.txt get before it stops being useful?

Frontier models in 2026 offer context windows from 200K to 1M tokens, but agents rarely spend a whole window on one file. Documentation sites commonly produce files of several hundred thousand tokens; beyond that, splitting by product area keeps each file loadable in a single context.

Do I need both llms.txt and llms-full.txt?

They serve different consumers. llms.txt is a link index an agent can navigate selectively; llms-full.txt is a one-shot payload for tools that ingest everything at once, like a developer pasting your docs into a chat session. Docs platforms typically generate both.

What Is llms-full.txt? The Full-Content Variant of llms.txt

Last updated: July 5, 2026 · Menra Team

llms-full.txt is the expanded companion to llms.txt: instead of a curated index of links, it concatenates the full markdown content of a site's pages into a single file. Where llms.txt says "here is what matters, go fetch it," llms-full.txt hands over the entire corpus in one request — a pattern that emerged from the llmstxt.org proposal ecosystem and was popularized by documentation platforms such as Mintlify, which began auto-generating both files in late 2024.

When does the full variant earn its keep?

The strongest use case is developer documentation. Engineers routinely paste an entire docs set into Claude or ChatGPT to get integration help, and AI coding agents fetch a vendor's llms-full.txt to load API knowledge before writing code against it. Anthropic publishes one for its own API documentation for exactly this reason. For a marketing site or blog, the case is weaker — those pages are consumed selectively, and the index-style llms.txt fits better.

How should size and freshness be handled?

Three practices keep the file usable rather than decorative. First, budget tokens deliberately: a 500-page docs site can easily exceed a million tokens flattened, so exclude changelogs, archived versions, and boilerplate, or shard into per-product files. Second, regenerate on every deploy — a stale llms-full.txt actively misleads any agent that trusts it, which is worse than not publishing one. Third, strip navigation, footers, and HTML artifacts so the file is pure content; the whole point is a higher signal-to-token ratio than crawling the rendered site.

What is a realistic expectation?

Like llms.txt itself, no major answer engine has confirmed fetching llms-full.txt during ranking or grounding as of 2026. Its proven consumers are humans and coding agents pulling docs into a context window on demand. That is still a real distribution channel — being the vendor whose documentation loads cleanly into an AI assistant shapes which tool the assistant recommends and how correctly it uses your product. Teams running a broader GEO program should treat llms-full.txt as a developer-experience asset with speculative engine upside, and measure whether AI-driven sessions actually fetch it via server logs before investing beyond automated generation.

Frequently asked questions

How big can llms-full.txt get before it stops being useful?: Frontier models in 2026 offer context windows from 200K to 1M tokens, but agents rarely spend a whole window on one file. Documentation sites commonly produce files of several hundred thousand tokens; beyond that, splitting by product area keeps each file loadable in a single context.
Do I need both llms.txt and llms-full.txt?: They serve different consumers. llms.txt is a link index an agent can navigate selectively; llms-full.txt is a one-shot payload for tools that ingest everything at once, like a developer pasting your docs into a chat session. Docs platforms typically generate both.

Keep exploring

See how AI engines talk about your brand — track mentions across ChatGPT, Perplexity, Claude, Gemini and 5 more. Start with Menra