Add llms.txt to your platform
llms.txt is a proposed Markdown convention for presenting a concise site overview and curated links. It is not a robots policy, licensing agreement, authentication control or universal AI indexing requirement. An H1 site name, a useful summary and organized Markdown links help readers understand the site. The proposal does not mandate a section literally called Docs or a universal 150-character summary. Adoption differs by consumer; do not claim that every major AI crawler reads it.
llms.txt is a proposed Markdown convention for presenting a concise site overview and curated links. It is not a robots policy, licensing agreement, authentication control or universal AI indexing requirement. An H1 site name, a useful summary and organized Markdown links help readers understand the site. The proposal does not mandate a section literally called Docs or a universal 150-character summary. Adoption differs by consumer; do not claim that every major AI crawler reads it.
WordPress doesn't expose root-level files by default.
guidesShopify doesn't allow static root files.
guidesWebflow doesn't host arbitrary root files.
guidesNext.js makes this trivial: drop the file in /public or expose a route handler that streams plain text.
guidesAdd a server route at src/routes/llms[.]txt.ts.
guidesGhost serves /public files via the theme, but a route file in your theme directory is cleaner and survives theme updates better.
guidesSquarespace doesn't allow root-file uploads.
guidesWix doesn't allow root files.
guidesFramer Sites can serve custom files via redirects to a hosted text endpoint, but the cleanest setup uses Cloudflare in front of the Framer domain.
guidesHubSpot supports custom file uploads via the Design Tools file manager.
guidesAstro serves anything in /public unchanged.
guidesGatsby copies /static to the site root at build time.
guides
What llms.txt does—and does not do
llms.txt is a proposed Markdown convention for presenting a concise site overview and curated links. It is not a robots policy, licensing agreement, authentication control or universal AI indexing requirement. An H1 site name, a useful summary and organized Markdown links help readers understand the site. The proposal does not mandate a section literally called Docs or a universal 150-character summary. Adoption differs by consumer; do not claim that every major AI crawler reads it.
Deploy at the public root without changing the site
Serve the file at /llms.txt with a successful response containing the intended text, not a themed HTML error page or sign-in screen. Use your CMS’s supported public-file feature, a static hosting asset or an explicitly configured text response. If the platform reserves root paths, confirm an officially supported deployment method before adding a plugin or proxy. Check the custom domain as well as the hosting preview, because redirects and caches can behave differently.
Choose a small set of useful destination pages
Link stable canonical pages: product overview, pricing, documentation, support policies and important guides. Use descriptive link labels and optional concise context so a reader knows what each page covers. Absolute HTTPS links are a practical publishing choice, not evidence that relative links automatically trigger HTTP errors. Exclude private dashboards, draft pages, tracking URLs and obsolete content. Keep linked information and translations current; do not list a page merely to inflate the file.
Verify after publishing and after important changes
Fetch the deployed file and inspect the response body, status and content type; text/plain with a UTF-8 charset is a sensible default. Follow the destination links to confirm they are public and relevant. Compare the live file with the version you intended to deploy after cache invalidation. A validator can identify formatting and link problems, but it cannot prove that an engine read the file or that a future answer will cite your site.
| Check | Evidence to retain | Action |
|---|---|---|
| File response | Successful /llms.txt response and readable Markdown | Fix HTML fallbacks and private routes |
| Site identity | A clear H1 and concise factual overview | Remove promotional or contradictory claims |
| Content links | Current canonical pages with useful labels | Replace obsolete or redirected links |
| Access policy | Separate robots.txt and authentication settings | Do not put Allow/Disallow rules in llms.txt |
| Maintenance | Published version and destination-page changes | Refresh when products or URLs change |
Implementation checklist
- Confirm your CMS supports the chosen public root-path method.
- Use Markdown, not a copied robots.txt template.
- Describe what the site actually offers.
- Include stable public URLs and clear labels.
- Test the real domain after deployment.
- Treat the file as optional context, not a citation guarantee.
A repeatable workflow
- 1Establish a baseline
Save the page URL, date, language, target market and exact questions before editing. Record what the page actually says and what each tested answer cites; a mention without a source link is not a citation.
- 2Make one coherent change
Resolve the clearest failure first, then check the published page rather than only an editor preview. Keep an implementation log so unrelated redesigns, campaigns or product changes are not mistaken for the effect of this work.
- 3Repeat and compare
Repeat the same questions and technical checks after deployment. Keep engines and measurement conditions consistent, report sample size, and inspect answer accuracy alongside citations. A single response is not a stable ranking.
Frequently asked questions
Does this guarantee an AI citation?
No. A crawlable, accurate page can still be omitted. Retrieval, source selection and generated answers vary between products and requests; these checks improve the evidence available, not guarantee placement.
How do I tell whether an improvement worked?
Compare like-for-like checks and retain the underlying evidence. Track correct brand mentions, source-linked citations and qualified referral visits separately. Do not call a technical score a traffic forecast or attribute every change to one edit.
Where are crawl restrictions configured?
Use robots.txt for preferences addressed to compliant crawlers, and authentication for private content. llms.txt does not enforce either crawl restrictions or a right to use content.
Related hubs
Industry-specific GEO starts with a real question, not a generic page containing an industry keyword. A software buyer may need compatibility, security and pricing details; a patient may need eligibility, risks and clinician qualifications. Write the answer in the language and jurisdiction of the audience, state who it applies to, and link the source of any claim. Search accessibility, useful content and trustworthy references remain important even when the final answer appears in an AI interface.
SEO, AEO and GEO overlap. SEO includes making pages accessible, relevant and useful for search; AEO emphasizes answering the question clearly; GEO concerns visibility and accurate representation in generated answers. They are not mutually exclusive replacements. Start with the business question: do you need more qualified visits, fewer misunderstood product answers, or a clearer understanding of citations? Choose measurements that match the task rather than declaring one discipline universally superior.
A search assistant can retrieve sources at answer time, use an index or answer from model knowledge without browsing. Its crawler and training controls may be separate. Test the actual feature and question you care about rather than assuming one user-agent rule controls the whole product. The engine guides here help organize that investigation; they are not claims that the vendors publish identical ranking rules.
Local AI visibility depends on accurate business facts and useful answers for the area you actually serve. A city keyword alone is not a reason to publish another page. State genuine service coverage, contact options, hours and location-specific limitations. Do not invent an office, address, local partnership, market-share statistic or citation benchmark to make a template look more substantial.
AI-search terms often describe different layers of a system. Crawling retrieves a page; indexing organizes available information; retrieval selects material for a question; generation produces an answer. A citation is an attribution in that answer, not proof of all preceding steps. Definitions are useful when they help you locate a problem, choose an appropriate check and avoid claiming more than the evidence shows.
Scan your site for AI search readiness
Get a free GEO, AEO and SEO score in under a minute — plus a ready-to-ship llms.txt, robots.txt and schema fix list.
Run a free scan