llms.txt for agent discovery, and why it will not move your Google rankings
llms.txt is an agent-onboarding convention, not an SEO signal. What the file is actually for, how to write one, and why nobody should sell it as a ranking tactic.
If someone told you to ship an llms.txt file to rank higher on Google, they sold you a myth. The file is real and worth shipping, but not for the reason it usually gets pitched. It moves nothing in Google search results and nothing in AI Overviews. What it does is help an agent you already pointed at your product find its way around.
What is llms.txt and does it help SEO? An llms.txt is a plain-text file served at /llms.txt that tells AI agents what your product is and how to use it: its endpoints, tools, and rules. It is an agent-onboarding convention, not a search ranking signal. It will not raise your Google position or get you into AI Overviews. It helps an agent that is already looking at your product discover what it can call.
What it actually is
Think of llms.txt as a README written for machines instead of people. When a developer or an agent lands on your domain and wants to know what your product exposes, the file gives a short, readable map: here is what this is, here are the endpoints, here are the tools, here are the rules for using them.
PankoBlitz ships one at pankoblitz.com/llms.txt. It names the product, points at the API docs, and lists the tools an agent can call. An agent reads it, learns the surface, and starts making calls without a human pasting documentation into the chat.
A minimal one looks like this:
# PankoBlitz
Publish landing pages from an AI agent.
## Tools
- create-site: make a new site
- publish-site: push it live at yourname.panko.page
## Docs
- API reference: /docs
- MCP tools: /guides/mcp-server-tools-reference
That is the whole idea. Short, honest, pointing at the real thing.
What it is genuinely good for
If you run an API or an MCP server, llms.txt earns its place. An agent that hits your domain gets a fast, readable index of what to call instead of guessing from marketing copy or scraping a rendered page. It is documentation for machines. For PankoBlitz that means an agent finds create-site and publish-site, checks the MCP tools reference, and gets to work.
Keep it lean. List the endpoints and tools that exist, link the API docs, and stop. Do not stuff keywords into it. Nothing reads llms.txt for keywords, so keyword stuffing buys you nothing and makes the file worse for the one audience that does read it.
The honest limit
Here is the part the pitches skip. llms.txt is not a Google ranking signal. Google has never said it uses the file, and there is no evidence it factors into search results or AI Overviews. Shipping one does not make you rank. It does not get you cited. It does not move impressions or clicks.
The confusion comes from a category error. SEO is about being discovered and ranked by a search engine crawling the open web. An llms.txt is about being usable by an agent that has already found you and is standing at your front door. Different problem, different file. Anyone selling llms.txt as an SEO tactic is selling cargo cult: the ritual without the mechanism.
Whether a few AI engines someday read the file for anything is unsettled and unproven. Treat any claim that it helps Google today as false until someone shows the receipts.
So should you ship one?
Yes, if you have tools worth calling. If you run an API or an MCP server, a short llms.txt is cheap and makes agent onboarding faster. Skip it if you are a static marketing site with nothing for an agent to invoke, because there is nothing to describe and no ranking reward to chase.
Ship it for what it does, which is help agents drive your product, and ignore anyone who frames it as a search hack.
Next step: read PankoBlitz's own llms.txt, then copy the shape for your product. List your real endpoints, link your docs, keep it under a screen of text.