llms.txt vs llms-full.txt vs llms-context.html: when to use which
llms.txt, llms-full.txt, llms-context.html: three files, three uses, one confusion to avoid. The guide that cleanly separates them.
1. llms.txt: the menu
llms.txt, proposed in September 2024 by Jeremy Howard, acts as a readable menu for LLMs. It contains a structured list of a site’s most important URLs with their titles and summaries. The equivalent of a sitemap.xml but designed for language model consumption.
Simple markdown format, at domain root, typical size 5 to 50 kB. Its function is not to be a manifesto but a map. LLMs that consume it know within fifteen seconds which priority pages to fetch to answer a brand question.
llms.txt = menu. llms-full.txt = full content. llms-context.html = human and machine context page.
2. llms-full.txt: the content
llms-full.txt goes further: it contains the full text of all priority pages, concatenated into a single markdown file. Useful for LLMs that cannot crawl page by page (latency) or want to ingest the whole corpus at once.
Typical size 200 kB to 2 MB. On expertcafe.be it weighs 1.4 MB, on zeroproof.one 2.1 MB. Limit: some LLMs truncate beyond 1 MB. The 2026 standard keeps llms-full.txt under 1 MB and fragments beyond.
3. llms-context.html: the context page
llms-context.html, less normalised but gaining adoption, is an HTML page designed to be read by both humans and LLMs. It describes the brand, its scope, its main entities, its authority sources. It plays the role of a hypertrophied About page geared to machines.
Its GEO value lies in being citable as a source. When an LLM cites ‘the brand X team’ it prefers pointing to a consolidated context page over a raw sitemap.
4. How the three fit together
The three files are not exclusive, they are complementary. A mature GEO policy carries all three at the root, in this implementation priority order: llms.txt first (the menu), llms-full.txt second (the content), llms-context.html third (the brand consolidation).
| File | Format | Typical size | Use case |
|---|---|---|---|
| llms.txt | Markdown | 5 to 50 kB | Priority URL menu |
| llms-full.txt | Markdown | 200 kB to 1 MB | Full content one-shot ingestion |
| llms-context.html | Structured HTML | 20 to 80 kB | Brand context page, citable |
5. Beginner mistakes
First mistake: confusing llms.txt with robots.txt. The former is an LLM menu, the latter an access control for all bots. They co-exist at root without interfering.
Second mistake: forgetting to update. llms.txt and llms-full.txt must regenerate at every content addition. On the GEO Rocket portfolio regeneration is automatic on every deploy.
Third mistake: aiming for exhaustiveness. llms.txt is not a full sitemap, it is a selection of strategic URLs (typically twenty to fifty). The more concise the better consumed.
llms.txt, llms-full.txt, llms-context.html are not three names for the same thing. They are three tools, three functions, three levels of service rendered to LLMs.
source: PROEMA LLM files standard
The PROEMA GEO diagnostic checks the presence and quality of all three files on a domain. Gaps read in under five minutes.