- [ Home ](/) / - [ Blog ](/blog/) / - How large is a real llms.txt file? # How large is a real `llms.txt` file? The proposal sets no universal ideal length. Production data gives a more useful baseline, provided percentiles are not turned into arbitrary rules. Last updated: September 1, 2026 On this page - [ What does a typical file contain? ](#what-does-a-typical-file-contain) - [ Why the distribution is wide ](#why-the-distribution-is-wide) - [ How to set a practical content budget ](#how-to-set-a-practical-content-budget) - [ Limits of size metrics ](#limits-of-size-metrics) Among 113 production responses meeting the H1-first inclusion rule, the median was 14,347 bytes with 7 H2 sections and 76 Markdown list links. Those numbers describe the sample. They are not recommended limits, because a focused product site and a large documentation corpus solve different problems. **Key takeaways** - The middle 50% ranged from 5,765 to 42,850 bytes. - Link counts ranged from 0 to 3,280, with a median of 76. - Curation quality cannot be inferred from size alone. ## What does a typical file contain? The 3 August directory snapshot contained 113 files. Size ranged from 648 to 741,350 bytes. The 25th percentile was 5,765 bytes, the median 14,347, and the 75th percentile 42,850. Section counts ranged from 0 to 84, with a median of 7. Parsed list links ranged from 0 to 3,280, with a median of 76. Ninety files included a blockquote summary. The complete aggregate is available in [`directory-insights.json`](/directory-insights.json), derived from the same files listed in the [verified directory](/examples/). ## Why the distribution is wide The proposal can describe a small business, a personal site, or a large developer platform. Documentation publishers often generate long catalogs automatically; a focused site may curate only a few destinations. Redirects, formatting style, link notes, and generated navigation all affect byte size. Category medians illustrate the difference. The seven e-commerce files measured 4,361 median bytes and 28 median links. The eight fintech files measured 93,207 median bytes and 471 median links. Those samples are small and descriptive, not sector standards. ## How to set a practical content budget Start from tasks, not a target word count. List the smallest set of pages an agent needs to answer the intended questions. Separate specialist material into path-level files where the boundary is real. You may label secondary resources `Optional` for editorial clarity, but v2 gives that heading no special client behavior. Review every description for disambiguation value. Remove marketing language, duplicates, redirects, and pages that merely repeat another source. If the map grows with every sitemap URL, it is no longer curated. The [generator](/generator/) can create a starting draft, but the [best-practice guide](/best-practices/) explains why manual selection remains necessary. Validate syntax with the [validator](/validator/). ## Limits of size metrics Bytes are not tokens, and tokens vary by tokenizer and language. A small file can be vague; a large file can be useful to a specialist client. The proposal does not define a ranking threshold or a universal maximum. Use file size as an operational signal: monitor sudden growth, build latency, stale links, and duplication. Evaluate usefulness with task-based tests, not a score derived only from length. ### Is 14 KB the ideal size? No. It is the median of this sample, not a recommendation. ### Are 76 links too many? It depends on scope. A path-specific documentation map can justify more links than a company-wide overview. ### Does a larger file improve AI visibility? There is no evidence for that causal claim. Larger files consume more context and can add noise. ## Sources - [ llmstxt.org, v2 proposal ](https://llmstxt.org/) - [ llmtxt.info, directory insights ](https://llmtxt.info/directory-insights.json)