If you have been told that your website needs an llms.txt file to show up in ChatGPT, it is reasonable to ask what that file actually changes.
It can help AI tools navigate a website, especially technical documentation. Google says it does not improve Search visibility or rankings. For most small-business websites, I would treat it as an optional addition. developers.google.com
The useful question is who will read it, and what you expect them to do with it.
What is llms.txt?
llms.txt is a text file written in Markdown, a simple format for headings, descriptions and links. It gives an AI tool a short introduction to a website and points it towards useful content. You will often find it at example.com/llms.txt, although the proposal also allows files within a section such as /docs/llms.txt. llms-txt
Think of a coding assistant trying to find the current installation guide for a software library. A concise index can help it reach the right instructions without guessing which page to open.
The similar filenames can be confusing, so here is what each one is for:
| File | Main purpose |
|---|---|
robots.txt | Gives compliant crawlers rules about which URLs they may request. |
sitemap.xml | Helps search engines discover URLs you want them to know about. |
llms.txt | Provides an overview and selected links for AI tools to consult. |
They serve different purposes. Adding llms.txt does not replace either of the other files or create an access-control mechanism. llms-txt
Does llms.txt help SEO or Google rankings?
Google’s current guidance is explicit: it ignores llms.txt for Search, including its generative AI features. Maintaining the file neither helps nor harms your site’s visibility or rankings there. developers.google.com
That gives a business owner a straightforward decision. If someone recommends the file specifically to improve your Google rankings, ask them to explain how that recommendation fits Google’s documentation.
Do ChatGPT and other AI tools use llms.txt?
There are documented uses, but they are specific to the tool and the task. The Claude Code and Codex examples below do not establish that ChatGPT Search routinely checks a website’s file or uses its presence to choose recommendations.
Anthropic recommends providing relevant documentation when working with Claude Code and identifies llms.txt files as one source of that material. This demonstrates a practical workflow. anthropic.com
Usage across ordinary websites is a separate question. Ahrefs examined 137,210 domains and found that roughly 38,000 had an llms.txt file. Of those files, 97% received no requests during May 2026 in its dataset. The sample consisted of Ahrefs analytics customers, so it does not represent every website or every future month. ahrefs.com
Even when a file is requested, a server log only tells you it was fetched. It does not show that the file caused a citation, recommendation or customer enquiry.
Where llms.txt has a practical use
Technical documentation offers a concrete example. A developer wants an assistant to use the correct API, follow the current setup instructions or find a specific configuration option.
In July 2026, documentation platform Mintlify reported 2,400 benchmark runs with Claude Code and Codex. Serving Markdown with a link to llms.txt substantially reduced failed URL requests. The experiment measured how agents found documentation, with web search disabled. It did not measure search rankings or how often a business appeared in AI recommendations. Mintlify's benchmark, benchmark methodology
For a software company, helping an assistant find the right documentation could be worth doing even without extra search traffic. The task itself has value.
For a local business with a handful of service pages, I would want to identify an equally concrete use before spending much time on a separate index.
Why does Lighthouse check for it?
Chrome’s Lighthouse includes an llms.txt check in its agentic browsing audits. That can make the advice seem contradictory until you look at what the check measures.
The official documentation says the file is optional. A missing file returning a 404 is marked not applicable; the audit flags server errors when attempting to retrieve it. This checks an aspect of agent access, and does not establish a Google Search ranking requirement. developer.chrome.com
Should a small-business website add llms.txt?
My recommendation depends on the work involved and the intended use.
If your platform already generates a sensible file, keeping it is reasonable. Check that it contains current information and working links. There is little reason to make removing it a project of its own.
If you publish extensive documentation that AI tools use, it is worth testing. Give an agent a few realistic tasks and see whether the index helps it find the correct pages. Record wrong links, missing information and time spent searching.
If you need more enquiries from a small business website, I would start elsewhere. A file listing your service pages cannot supply the prices, examples or explanations those pages are missing.
If someone is selling it as an AI visibility package, ask what the package delivers. Generating a text file, fixing inaccessible content and improving your service pages are different amounts of work. The proposal should explain the work and the expected outcome without promising recommendations it cannot control.
What should an llms.txt file contain?
Keep it short, accurate and easy to maintain. A useful file introduces the site or project and groups links to relevant material. Documentation platforms such as Mintlify use it to point tools towards Markdown versions of their pages. Mintlify
For example, a documentation site could use this structure. The domain and paths below are illustrative:
# Example product
> Documentation for installing and using Example product.
## Documentation
- [Getting started](https://example.com/docs/start.md): Installation and setup.
- [API reference](https://example.com/docs/api.md): Endpoints and parameters.
- [Troubleshooting](https://example.com/docs/help.md): Common errors and fixes.Use URLs that actually exist, and check the destinations after publishing. Avoid creating a second version of your pricing or policies that someone has to remember to update separately. Linking to maintained information is easier to keep accurate.
Treat it as a resource you can test. If an assistant still cannot find the right answer, inspect the linked content as well as the index.
An example from this website: reviewing a generated draft
The generator at the end of this article shows why creating the file and reviewing it are separate jobs. It reads the submitted page and selected public pages linked from it, up to 30 pages in total. It copies existing titles and meta descriptions into a Markdown draft; it does not write new explanations of what the business does.
That has practical consequences. A page missing its meta description can appear without a summary. A useful guide that is only linked from a deeper page can be absent, because the generator does not recursively explore the whole site. A navigation label such as “Services” cannot tell the reader which service answers their question.
Before downloading the draft, I would check three things:
- Are the important destinations included? Add a maintained service page or guide manually if the draft missed it.
- Do the descriptions distinguish the pages? Replace repeated marketing slogans with a short explanation of what each destination contains.
- Does every link open the intended public content? Remove obsolete, duplicate or irrelevant destinations.
This is an example of the tool's implementation, not evidence that publishing its output improves AI recommendations. The draft is editable, and generating it does not publish anything to your website.
Practical llms.txt FAQs
How do I add llms.txt to a WordPress website?
First check whether your site already serves a file at https://yourdomain.com/llms.txt. If it does, identify what generates it before creating another copy. You need one maintained response at that URL; a physical text file or a CMS-generated response can provide it.
For a manually maintained file, upload it to the public website root, rather than putting it inside a theme's folder. Then open the URL in a private browser window. It should return the Markdown text without a login, an error page or your ordinary website layout. If a plugin generates the file, review its selected pages and descriptions through that plugin. Avoid having a plugin and an uploaded file compete to serve the same URL.
Should it list every page, and must the links end in .md?
Select pages that help with the tasks you expect an assistant to perform. For a service business, that might mean its services, relevant project examples and contact information. Listing every tag archive and old announcement can make the useful links harder to find.
The proposal recommends linking to clean Markdown versions where available. A .md extension is useful only when that URL really serves the content. Do not add it to ordinary WordPress URLs and assume a Markdown page will exist. llms.txt proposal
Does a bilingual website need two files?
A single root file can group links by language and label them clearly. This site's theme uses that approach: /en/llms.txt redirects to /llms.txt, keeping one document to maintain. That is an implementation choice, not a requirement for every bilingual website.
For a small site, I would start with language headings and links to the actual translated pages. If a documentation section needs its own index, the proposal also permits a file such as /docs/llms.txt. Make sure the scope and language are clear to whoever reads it. llms.txt proposal
Can llms.txt block AI crawlers or protect private content?
No. Describing a page in llms.txt does not grant access to it, and omitting a page does not prevent access. Use the relevant crawler's documented controls for crawl preferences and authentication for private material. Even robots.txt depends on crawlers obeying its rules; it is not a security boundary. Google's robots.txt guidance
If a page is intentionally blocked, adding a link to it in an index will not resolve the access problem. Decide whether that page should be available before changing the index.
How can I tell whether anyone uses the file?
Ask your hosting provider for server or CDN request logs and filter for /llms.txt. Record the date, response status and reported user agent. Exclude your own checks and monitoring requests. A reported bot name alone is not proof of who made the request, and a successful fetch is not proof that the content influenced an answer.
For a more useful test, choose questions with answers you can verify on your own site. The following are hypothetical test tasks, not results from this website:
| Test question | What to check |
|---|---|
| “Does this developer work on existing WordPress sites?” | The assistant finds the relevant service page and answers from its stated scope. |
| “Which project demonstrates a 3D product viewer?” | It reaches the correct example and does not describe a demo as paid client work. |
| “How do I request an assessment of my website?” | It finds the current contact route and explains what information to send. |
Try the same tasks with the homepage as a starting point, then with llms.txt, using the same tool and settings. Repeat them because answers can vary. Note which pages were fetched, broken links and unsupported statements. This tests whether the index helps navigation; it does not establish a ranking or traffic benefit.
How often should I update llms.txt?
Review it whenever you change an included page's URL, remove a service or substantially change its content. Keep that check in the same workflow as the page edit. If a platform generates it automatically, check the output after those changes rather than assuming every relevant page was included.
For example, suppose a service page moves from /web-design/ to /services/websites/. This is a hypothetical maintenance scenario. Update the index to the current destination and verify the old URL's redirect separately. A redirect can keep an old link working, while the file still gives a reader an outdated description of the service.
Do I also need llms-full.txt?
Some documentation platforms provide it as a separate file containing the full page content. Mintlify, for example, distinguishes its link index from a combined Markdown export. Mintlify's file documentation
For a small business website, I would keep the reviewed index first. A full copy creates another place where prices, policies and service details can become stale if it is maintained manually. Consider it when a specific workflow needs the combined content and you can generate it from the same maintained source.
What I would improve first
For a business website, I would first check whether the important pages can be accessed and understood. Google’s guidance continues to emphasise crawlable content, internal links, important information in text and a good page experience. developers.google.com
Then I would read the site as a potential customer. Does it explain what you do, who you help, where you work and what happens next? Are there useful project examples? Can someone find the information they need before contacting you?
Imagine a renovation company whose website barely explains its services. Adding a file that links to those vague pages leaves the buyer’s questions unanswered. Writing a useful page about the work, typical project scope and quotation process addresses the problem directly.
Add llms.txt when it supports a real use case and the upkeep is small. Give your main attention to content that helps someone make a decision.
Need help identifying what to fix on your website? Get in touch for a review of its technical setup, content structure and usability.
Create an llms.txt draft.
Enter a website URL, then review and download a short index of its public pages.
Check the titles, descriptions and links, then upload the file so it is available at yourdomain.com/llms.txt.
An llms.txt file does not guarantee inclusion in AI answers. About the format ↗
Turning search visits into useful next steps?
Useful content needs a clear website behind it. I can help improve page structure, performance and the path from an article to your services, products or enquiry form.
