The big question with LLMs right now isn't whether llms.txt is worth doing. It's simpler and more uncomfortable: what do these tools actually know about your company? And how consistent is that across all of them?
For almost every brand, the honest answer is that nobody knows — and there's always something the model is missing, because no one remembered to explain it.
That's not a new problem. It's the oldest problem in SEO, with a bigger audience.
Years ago, when I worked SEO for complicated products, I figured out something that shaped how I've worked ever since: if I cared about SEO, Google was one of the audiences I was marketing to. Not a channel. An audience. And if I didn't market to Google directly, I wasn't going to get the SEO I wanted.
So I treated it like one. The products I worked on were technical enough that Google couldn't make sense of them on its own — it couldn't see how the technology related to other technology, which is the part that actually matters. So I went into the schema and made sure it understood. That's not the standard marketer's instinct. It takes a technical digital marketer comfortable enough to get into the structured data and confirm the machine understood the relationships, not just the words.
That was the whole job. Give the machine context.
Here's the only thing that's actually different now.
For years, that audience was effectively singular: Google. Today it's Google, Gemini, Claude, ChatGPT, and hundreds of other AI tools and systems answering questions about your company. The audience I always marketed to didn't go away. It grew — from one to many.
llms.txt is just one way to speak to that audience. It's a plain-language definition of who you are, who you are not, what you do, and how it all relates — written so a model can find the signal without crawling your whole site. Same instinct as schema. New, much larger room to address.
This is the shift in thinking worth sitting with: a bot is now one of your ideal customer profiles.
You already obsess over whether your messaging lands with the people you want to reach. The crawlers and models deserve the same treatment, because they're increasingly the thing standing between your brand and the human asking about it. If your messaging doesn't work for them, it doesn't reach the person behind them.
Things are changing fast, and I won't pretend the adoption picture is settled. But the direction doesn't depend on any single file or format. Whether it's traditional SEO or an LLM answering a question, the crawler has to understand what you do. That requirement isn't going anywhere. Context is the constant.
Make context something you manage on purpose, not something you hope a model infers. A few practical places to start:
The agencies that win the next few years will be the ones already treating machines as an audience worth marketing to — because they've been doing it since the only bot in the room was Google.
Context was everything then. With a hundred tools answering questions about your brand, it's everything now.
You don't need a tool budget to start. Open the major models and ask them about your brand directly. Run the same prompts across each one — the disagreements are where the work is.
Where to look:
Prompts worth running:
Read the answers the way a prospect would. Where a model hedges, guesses, or gets your category wrong, that's a gap in the context you've given it — not a flaw in the model.