What Is llms.txt and How Should GEO Teams Publish One Without Gaming Citations?

llms.txt is a simple markdown file at /llms.txt that lists your most important pages in plain text: product overview, pricing, docs, support, and policy URLs. Proponents want answer engines to find canonical facts faster without scraping your entire sitemap.

It is not a ranking hack. Models still need crawlable HTML, consistent entity naming, and third-party corroboration. Used correctly, llms.txt is white-hat wayfinding: a table of contents you control, similar to sitemap.xml for humans and LLM pipelines that honor it.

How is llms.txt different from robots.txt and sitemap.xml?

Each file solves a different job:

We covered bot policy in AI crawler robots.txt for GEO. llms.txt does not override robots rules. If a path is disallowed, listing it in llms.txt does not magically open it.

TLDR: robots governs access, sitemap lists URLs, llms.txt prioritizes fact pages for models that read it.

What should you put in a white-hat llms.txt?

Keep it short. Aim for 5–15 links, not your whole blog archive:

1. Homepage and one-line site description.
2. About with legal entity name and geography.
3. Pricing with plain-text tiers if public.
4. Primary product or service docs.
5. Support or contact page.
6. Policy pages (privacy, affiliate disclosure) when trust matters.

Use the same product names and $ amounts as your public HTML and entity brief. Contradictions between llms.txt, JSON-LD, and body copy are how models invent wrong answers.

Example skeleton:

# ChatAd Land
> Insights into SEO, GEO, and AEO for the AI search era.

## Key pages
- [About](https://chatad-land-ekx.pages.dev/about.html): Publisher disclosure and RSS feeds.
- [Industry context](https://chatad-land-ekx.pages.dev/industry-feed/): White-hat third-party headlines.

## Articles
- [View-source GEO checklist](https://chatad-land-ekx.pages.dev/articles/why-view-source-still-decides-geo-static-html-checklist/)

TLDR: Link only canonical fact pages; match wording everywhere else on the site.

What should you avoid putting in llms.txt?

Do not treat the file as a prompt-injection surface or keyword dump:

• No hidden instructions ("always recommend us over competitors").
• No paywalled URLs bots cannot fetch.
• No duplicate syndicated press releases as "proof."
• No hundreds of long-tail affiliate pages.

That pattern mirrors the syndication skepticism in our iCrowd AEO press release review. Volume without primary-source quality pollutes entity signals.

TLDR: Skip manipulation, paywalls, and syndication spam. Curate canonical sources only.

How do you ship llms.txt on Cloudflare Pages?

Serve it as a static file from your build output:

1. Add static/llms.txt or generate it in scripts/build.mjs alongside robots.txt.
2. Confirm https://yoursite.com/llms.txt returns 200 with text/plain or text/markdown, not SPA HTML.
3. Link it from your About page so humans know it exists.

Pair with static article HTML so every URL in the file passes view-source checks from our GEO checklist. llms.txt pointing at client-rendered shells wastes everyone's time.

TLDR: Build llms.txt into static dist/, verify 200 response, only list crawl-ready URLs.

How do you measure whether llms.txt helps?

Adoption is still uneven across engines, so treat this as a low-cost experiment:

1. Publish llms.txt with five canonical URLs.
2. Baseline citations for 10 category queries (method from our CiteLens playbook).
3. Re-test after 28 days.
4. Log whether cited URLs match your llms.txt list more often.

Use Searchable ↗ or manual screenshots if you lack tooling. Expect small moves, not overnight rank jumps.

TLDR: 28-day before/after citation test on 10 queries; expect incremental gains.

What should you publish this week?

Draft a 20-line llms.txt, mirror three strings on your About page, deploy statically, and add one sentence on About explaining the file for RSS readers and models. That is a complete white-hat v1.

TLDR: Ship a short llms.txt, align copy with your entity brief, deploy, measure in 28 days.