llms.txt: The Setup Guide for SMB Websites

15+ years in local marketing; Google Ads certified; Shopify Partner.

73% of Canadian SMBs are using AI tools in some form, per Microsoft's 2025 Canada report. Your customers are part of that number. They're asking ChatGPT, Claude, and Perplexity questions about your industry every single day. And most SMB websites have done absolutely nothing to tell those models what they are, what they sell, or why they should be cited.

That's what llms.txt is for. It's a small file you put on your website that gives large language models a clean, structured summary of your content. Think of it like a menu handed to a rushed reader. "Here's what this site is about, here's the important stuff, here's how to understand it."

This guide walks you through what llms.txt is, how to actually build one for a small or medium business website, what to put in it, and whether the AI crawlers are even reading it yet. No hype. Just the file, the format, and the work.

One thing up front: llms.txt is not a magic ranking file. It's not going to force ChatGPT to cite you. What it does is make your site easier for LLMs to parse when they do visit. That's the piece most guides skip.

What llms.txt Actually Is

Back in September 2024, Jeremy Howard (the guy behind fast.ai and Answer.AI) proposed a simple standard. Websites should have a file at /llms.txt that gives language models a markdown summary of the site. Not HTML. Not JavaScript-rendered junk. Just clean markdown with links to the content that matters.

The format is straightforward:

# Your Business Name

> One-sentence description of what you do.

Optional longer paragraph giving context about your business, your audience, and what makes your site worth reading.

## Services

- [Dental Implants](https://yoursite.com/services/implants): Full-arch and single-tooth implants in Regina.
- [Invisalign](https://yoursite.com/services/invisalign): Clear aligner treatment for teens and adults.

## About

- [Our Team](https://yoursite.com/team): Meet the dentists and hygienists.
- [Pricing](https://yoursite.com/pricing): Transparent fees for common procedures.

## Optional

- [Blog](https://yoursite.com/blog): Articles on oral health and dental care.

That's it. Markdown headings, bullet links, optional descriptions. The ## Optional section is the one LLMs can skip if they're budgeting tokens.

There's also a separate file called llms-full.txt which is meant to contain the actual full text of your key pages, not just links. More on that below.

Here's the thing. Neither file is an official web standard. It's not ratified by W3C. Google hasn't publicly endorsed it the way they endorsed robots.txt or sitemap.xml. It's a community proposal that picked up traction because the idea is genuinely useful.

Who's Actually Reading It (Honest Answer)

I think this is where most guides oversell. Let me give you the real picture.

As of 2026, the crawlers that matter for AI search are GPTBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot, and Google-Extended. None of them have publicly confirmed that they give special weight to llms.txt the way Google gives weight to a sitemap. For the breakdown on individual crawlers, see our posts on how GPTBot behaves and whether to allow ClaudeBot.

What we do know:

Some AI-native tools and agents (Cursor, Cline, various dev tools) explicitly look for llms.txt to understand a site's structure.
Anthropic has referenced the format in its own documentation.
Several SEO tools that track AI citations (Profound, AthenaHQ) have noted that sites with clean, parseable content tend to get cited more, and a well-built llms.txt is a strong signal of clean content.
Having the file does no harm. It's a 2KB text file. The downside is zero.

So my honest take: build it, but don't expect it to be the thing that gets you cited in ChatGPT next week. The thing that gets you cited is having actual useful content that matches the questions people are asking. For the full breakdown of that, see our AI SEO playbook and the guide to earning AI citations from ChatGPT and Perplexity.

llms.txt is a polite, structured way to say "here's the map of this site" to any LLM that wants one. Build it. Move on. Focus on content.

What to Put In Your llms.txt (SMB Edition)

Most of the templates online are written for SaaS companies or developer tools. If you run a dental practice, a law firm, an HVAC company, or a property management business, you need a different shape.

Here's what I recommend for a typical Canadian SMB site.

Section 1: Business identity. Your name. One sentence on what you do. One paragraph on who you serve and where.

Section 2: Services. Your top 5-10 services, each linked to its own page with a one-line description. This is the most important section. If an LLM is trying to answer "who does dental implants in Regina," this is the section it's matching against.

Section 3: Location and service area. If you're a local business, name the cities you serve. "Serving Regina, White City, Pilot Butte, and surrounding communities." LLMs are surprisingly good at geographic matching when you give them clean signals.

Section 4: About / trust signals. Link to your team page, your reviews page, your case studies, your credentials. This is where E-E-A-T (experience, expertise, authoritativeness, trust) lives for LLMs the same way it does for Google.

Section 5: Optional. Blog, resources, downloads. Mark this section ## Optional so models can skip it if they're budgeting.

A real example skeleton for a Saskatchewan trades company:

# Meridian Mechanical

> Residential and commercial HVAC service in Regina and Saskatoon.

We install and service furnaces, air conditioners, heat pumps, and
commercial rooftop units across southern Saskatchewan. Licensed, bonded,
and TSASK-certified since 2008.

## Services

- [Furnace Installation](https://example.com/furnace): High-efficiency gas and electric furnace installation.
- [AC & Heat Pumps](https://example.com/cooling): Central air and cold-climate heat pump installs.
- [24/7 Emergency Service](https://example.com/emergency): Same-day response in Regina and Saskatoon.

## Service Area

- [Regina](https://example.com/regina): Full service, same-day response.
- [Saskatoon](https://example.com/saskatoon): Full service, next-day response.

## About

- [Our Technicians](https://example.com/team): Red Seal certified HVAC techs.
- [Reviews](https://example.com/reviews): Google and HomeStars verified reviews.

## Optional

- [Maintenance Blog](https://example.com/blog): Seasonal tips for homeowners.

Plain. Scannable. A language model can read that in under a second and come away knowing exactly what to cite you for.

Week-by-Week: Actually Building the File

I'll walk you through how we do this for clients. If you're DIY-ing it, same process, just wearing two hats.

Week 1: Inventory. Before you write anything, list every page on your site that matters. Use Screaming Frog or just export your sitemap.xml. Separate pages into three buckets: core services, supporting content (about, contact, team), and optional (blog posts, old landing pages). Typically an SMB site has 15-40 pages. Your llms.txt will link to maybe 10-20 of them. The rest don't need to be there.

Week 2: Write the descriptions. Each bullet in llms.txt is - [Page Title](URL): One-line description. The one-line description is the thing most people phone in. Don't. Write it the way you'd explain that page to a stranger at a party. "Full-arch dental implants starting at $22,000, including consultation and 3D imaging." Not "Learn more about our implant services."

Week 3: Write the intro paragraph. The blockquote summary (the > line) and the paragraph underneath it are what an LLM grabs if it's only reading the first 500 tokens. Write them like they're the only thing anyone will ever read. Who you are. Who you serve. Where. What makes you different. No fluff.

Week 4: Publish and validate. Put the file at https://yoursite.com/llms.txt. Plain text. UTF-8 encoding. Test it by opening the URL directly in your browser, it should display as readable markdown. Then run it through one of the free validators (llmstxt.org has one). Add a link to it in your robots.txt as a courtesy signal.

After that, revisit quarterly. Every time you add a service page, update the file. Every time you drop a service, remove it. Treat it like your sitemap.

If you're running through this process and it feels like a lot, it's because writing good one-line descriptions for 15 pages is genuinely harder than it sounds. Most SMB owners spend more time on the descriptions than they expect to. That's fine. The descriptions are the part that matters.

llms.txt vs llms-full.txt vs robots.txt vs sitemap.xml

This trips people up constantly. Four files, similar names, different jobs.

robots.txt tells crawlers which URLs they can and can't access. It's the bouncer. "You, GPTBot, can come in. You, ClaudeBot, stay out."

sitemap.xml lists every URL on your site in machine-readable XML format. It's the phone book. Google and other crawlers use it to find pages.

llms.txt is a curated, human-readable markdown summary of the most important content. It's the menu. Not every page, just the ones you want LLMs to focus on.

llms-full.txt is optional and bigger. It contains the actual full text of your key pages in a single markdown document. The idea is that an LLM could ingest one file and have your whole site's content. Useful for documentation-heavy sites (software companies, knowledge bases). For most SMBs, probably overkill.

I'd recommend building llms.txt for every SMB. Skip llms-full.txt unless you have a specific reason (big knowledge base, technical product documentation, legal resource library).

A Worked Example: Cost of DIY vs Hiring It Out

Let's do the math.

Writing your own llms.txt takes a focused SMB owner roughly 4-6 hours across inventory, writing, and publishing. Call it 5 hours at whatever you pay yourself per hour. If you bill at $150/hour (typical for a professional services owner), that's $750 of your time.

An agency will charge somewhere between $500 and $2,500 for a one-time llms.txt build, depending on site size and whether it's bundled with a broader AI visibility audit. Canadian SEO consulting rates cluster around CA$125-$200/hour per 2024 industry survey data from various Canadian SEO communities.

So the math is pretty close to a wash for a small site. The real question is whether you'll actually do it. Assume 5 hours for your practice. Check your real hourly rate in your own books. Then decide.

For a site with 50+ pages, multiple service lines, and complex content, hiring it out starts to make more sense. For a single-location dental practice with 15 pages, do it yourself in an afternoon.

Pattern Observations From Building These

When I look at the llms.txt files I've built for clients across a range of industries, a few patterns hold up.

Most SMB sites that invest in a clean llms.txt also end up fixing their page titles, their meta descriptions, and their page structure in the process. The file forces you to describe each page in one sentence, and that exercise tends to surface weak pages. Usually, the sites that go through this process end up with better regular SEO too, not because llms.txt did anything magical, but because the discipline of writing tight descriptions bleeds into the rest of the site.

Across practices and trades businesses, the sites that get cited in ChatGPT and Perplexity earliest are the ones with tight topical focus. A dental practice that lists 4 services clearly gets cited more than one that lists 18 services vaguely. llms.txt is a forcing function for that clarity.

Canadian Considerations

Two quick notes for Canadian SMB owners.

CASL doesn't apply to llms.txt. CASL governs commercial electronic messages (email, SMS). A text file on your own website isn't that. Carry on.

Quebec Bill 96 might. If you operate in Quebec and your website is required to be available in French with French equal or greater prominence, your llms.txt should probably exist in both languages, or at minimum reference the French version of your site. I'd publish llms.txt in English and llms-fr.txt in French, and link both from the root. This isn't settled practice, it's my read.

PIPEDA considerations are zero. llms.txt should never contain personal information. It's a description of your site, not your customer list.

Red Flags When an Agency Pitches You llms.txt

A few things to watch for if someone's trying to sell you llms.txt services.

"We'll guarantee you show up in ChatGPT." No they won't. Nobody can. Run.
"Our proprietary llms.txt format." It's a public spec. There's no proprietary version.
$5,000+ for a one-time file build on a 20-page site. That's a shakedown. The file itself is a couple hours of work. Even bundled with an audit, you shouldn't be paying more than $2,500 for a small site.
Recurring monthly fees for "llms.txt management." If your services aren't changing, the file isn't changing. Quarterly updates are plenty. A monthly retainer just for this is nonsense.
No mention of the actual content work. llms.txt without good content behind it is a map to nothing. The file is 5% of the work. The other 95% is building pages worth citing.

For the bigger picture on what a legitimate AI marketing engagement should cost, see our AI marketing agency guide and the breakdown of AI readiness audits and what agencies actually charge for them.

Where llms.txt Fits in the Bigger Picture

llms.txt is one small piece of being visible in AI search. It's not the strategy. It's not even the most important tactic.

The important tactics, in rough order:

Having clear, useful content that answers the questions your customers actually ask. Start with AI content writing for SMBs.
Technical hygiene: fast site, clean HTML, proper schema markup. See schema markup for AI search.
Allowing (or blocking) the right crawlers in robots.txt. Different sites have different calls here.
Earning citations elsewhere on the web (directories, press, partner sites) so LLMs have corroborating signals about you.
Publishing llms.txt to make parsing easier.

If you're working on #5 before #1, you're painting a house with no walls. For the specific playbook on getting your site visible across AI answers, our posts on how to show up in AI search and how to rank in Google AI Overviews go deeper on the content and technical side. For the broader strategy on AI for marketing in Canadian SMBs, that's where the full sequence lives.

Summary: What To Do This Week

Three takeaways if you're an SMB owner reading this and wondering what's next.

One. Build a llms.txt file for your site. Even a rough one is better than nothing. It takes an afternoon. Put it at /llms.txt and forget about it until your services change.

Two. Don't expect it to move the needle on its own. It's a hygiene file, not a growth lever. The growth comes from content, citations, and technical health. llms.txt just makes those things easier to read.

Three. If someone tries to sell you a $5K llms.txt engagement, hand them this article and walk away. Or bundle it with a real AI visibility audit where the file is a line item, not the headline.

Build it. Keep it simple. Move on to the work that actually drives leads.