llms.txt explained, and whether your site needs one

Somebody probably told you your site needs an llms.txt file. Maybe an agency pitched it as the thing that gets you into ChatGPT. Maybe a tool flagged it as a missing item on an audit.

Here is the straight version. llms.txt is a reasonable idea, it costs almost nothing to add, and it is not a ranking factor, a citation lever, or a shortcut into AI answers. Anybody selling it as one is selling you something else.

This post covers what the file actually is, where adoption really stands, what belongs in it, and whether you should spend an afternoon on it.

What llms.txt is

llms.txt is a proposed convention: a single markdown file at the root of your domain, at yoursite.com/llms.txt, that gives a large language model a clean, human-written map of your site.

The reasoning behind it is sound. A modern web page is mostly not content. It is navigation, cookie banners, scripts, tracking, chat widgets, footers, and layout markup wrapped around a few hundred words that matter. When a model fetches that page at inference time, it has a limited context window, and a large share of it gets spent on the wrapper instead of the substance.

llms.txt proposes you skip the parsing problem. Instead of making a model reverse-engineer your site structure from HTML, you hand it a short markdown document that says: here is what this business is, here is the terminology we use, and here are the specific pages that matter, with one line each explaining what is in them.

The format is deliberately plain markdown:

• An H1 with the name of the site or business
• A blockquote with a short summary
• Optional prose giving context, definitions, or disclaimers
• H2 sections containing lists of links, each link followed by a colon and a brief description
• An optional "Optional" section at the end holding lower-priority links that can be skipped when context is tight

That is the whole specification. It is not complicated, which is part of why it caught on as an idea.

A related convention travels with it. Some sites also publish clean markdown versions of individual pages at the same URL with .md appended, so a model following a link from llms.txt gets the content without the wrapper. Some documentation platforms generate both automatically.

The honest state of adoption

This is the part most articles about llms.txt skip.

As of late 2026, no major AI provider has publicly documented that its crawlers or models read llms.txt as an input to how they retrieve, rank, or cite content. The convention is a community proposal, not an adopted standard. There is no equivalent of the robots.txt situation, where every serious crawler has honored the file for decades.

What has happened is narrower. Developer documentation sites adopted it quickly, partly because the audience genuinely is developers pasting docs into AI tools, and partly because several documentation platforms added automatic generation so publishers got it for free. In that context it has real utility: a developer can point an AI coding assistant at llms.txt and get a usable map of an API.

What has not happened is evidence that publishing one improves your visibility in AI search for an ordinary business website. If someone shows you a traffic chart and credits llms.txt, ask what else changed that month. Agencies that added llms.txt usually added a dozen other things at the same time.

So treat it as what it is: cheap, harmless, possibly useful later, occasionally useful right now when a human or a tool is pointed at it directly. Not a growth channel.

How it differs from robots.txt and sitemap.xml

These three files get lumped together and they do completely different jobs.

robots.txt is permission. It tells crawlers what they may and may not fetch. It is an established convention that essentially every legitimate crawler respects, and it is where you control AI crawler access by user agent if you want to allow or block specific bots. This file has teeth.

sitemap.xml is inventory. It is a machine-readable list of every URL you want indexed, with metadata about when each changed. Search engines genuinely use it for discovery, especially on large sites. If you have to pick one file to get right, it is this one.

llms.txt is curation. It does not grant or deny access, and it is not a complete inventory. It is an editorial document that says which pages matter most and what each one contains. It is the only one of the three that is written for a reader rather than a parser.

The practical difference: a sitemap lists four hundred URLs with no opinion about any of them. An llms.txt lists the twelve that answer the questions people actually ask, and explains each one.

Note also that llms.txt does nothing for AI crawler policy. If you want to control whether AI companies can train on or retrieve your content, that happens in robots.txt and in your terms, not here.

What to actually put in it

Keep it short and genuinely informative. A bloated llms.txt defeats the entire point, which is to save context.

Include:

1. Your business name, what you do, and where you do it
2. A one-paragraph plain summary a model could quote accurately
3. Service pages, one line each describing what the page covers
4. Service area, if you are a local business
5. Pricing or pricing structure, if it is published
6. Your highest-value answer content, the pages that resolve real questions
7. Contact details and hours
8. Anything commonly misunderstood about your business, stated plainly

Leave out:

• Marketing adjectives. "Industry-leading" tells a model nothing and wastes tokens.
• Every blog post you have ever written. Curate.
• Anything you would not want quoted back to a prospect, because that is the use case.
• Duplicated content. Link to pages, do not paste them.

Here is a realistic example for a service business:

# Hill Country Comfort Systems

> Residential HVAC repair, replacement and maintenance serving Boerne,
> Fair Oaks Ranch and northwest San Antonio, Texas. Family owned,
> licensed in Texas, 24 hour emergency service.

Founded 2011. We service all residential brands and install Trane and
Daikin systems. Emergency calls are answered by a human at all hours.
We do not service commercial rooftop units or mobile home systems.

## Services

- [AC repair](https://example.com/ac-repair): Diagnostic process, common
  failures, same-day repair availability, what a service call costs.
- [AC replacement](https://example.com/ac-replacement): Sizing, SEER2
  ratings explained, equipment options, financing, typical price ranges.
- [Heating repair](https://example.com/heating-repair): Gas furnace and
  heat pump repair, no-heat emergency response.
- [Maintenance plans](https://example.com/maintenance): What two visits
  per year include, plan pricing, member discounts.
- [Indoor air quality](https://example.com/air-quality): Filtration,
  UV, humidity control, duct sealing.

## Service area

- [Service area](https://example.com/service-area): Boerne, Fair Oaks
  Ranch, Leon Springs, Helotes, Bulverde and northwest San Antonio,
  with response times by area.

## Common questions

- [How much does AC replacement cost](https://example.com/ac-cost):
  Price ranges by system size and efficiency tier, what drives cost.
- [Repair or replace](https://example.com/repair-or-replace): Age,
  refrigerant type and repair cost thresholds for making the call.
- [Why is my AC freezing up](https://example.com/ac-freezing): Airflow
  and refrigerant causes, what to check before calling.

## Contact

- [Contact](https://example.com/contact): Phone, text, hours, emergency
  line, office address.

## Optional

- [About us](https://example.com/about): Company history, team, licensing.
- [Reviews](https://example.com/reviews): Customer reviews by service type.

That is maybe twenty minutes of work and it reads like something a person wrote on purpose.

Who should bother

Worth doing now:

• Documentation sites, software products and anything with an API, where the audience is already feeding your content to AI tools
• Sites on a platform that generates it automatically, since the cost is zero
• Large sites where structure is genuinely hard to infer from navigation
• Anyone who wants a single clean reference document to point a tool or a new employee at

Do it when you have a spare afternoon:

• Ordinary local service businesses. It is cheap insurance against a future where this matters, and the exercise of writing it forces useful clarity about what your important pages actually are.

Do not prioritize it over:

• Having real pages that answer real questions
• A correct and current Google Business Profile
• Page speed and mobile usability
• Clean structured data on your key pages
• Reviews
• A working sitemap and sane robots.txt

Here is the test I would use. If an AI tool read only your llms.txt, could it accurately tell a stranger what you do, where, for whom, and roughly what it costs? If not, the problem is not the file. It is that your site has never said those things clearly anywhere, and no file at the root of your domain fixes that.

Write the llms.txt last, after the content it points at is worth pointing at.

The short version

llms.txt is a small, sensible convention with limited adoption and no proven effect on AI visibility for most business sites. Add it because it is nearly free and it costs you nothing to be early. Do not let anyone bill you for it as a strategy.

If you want a plain read on which AI search work is actually moving the needle for businesses like yours and which is noise, Heights Digital does this every day and will give you a straight answer. Call or text 210-729-4008.