Goce PetrovBrand designer, Switzerland

Readable to AI

How to make your company readable to AI assistants and reachable by their agents. Eight steps from my own setup, with the numbers and the mistakes.

The cover of the guide: Readable to AI, a field guide to being understood by AI assistants and agents.
  • Free PDF, 14 pages
  • 8 steps and a one-page checklist
  • English, September 2026

Get the free guide

What is inside

For founders and marketers at funded startups, and for brand people who have been told this is a developer job. Written from my own setup in September 2026, with dates, numbers you can check, and my mistakes.

  1. 01

    Write the facts once

    Put your core sentence, services, proof and contact in one facts file, next to a do-not-claim list, and let every page, profile and agent read from it.

  2. 02

    A facts page and structured data

    Publish a plain /for-ai page with who you are, when you are a good fit and when you are not, and the same structured data on every page.

  3. 03

    llms.txt and a front door for agents

    Add a short llms.txt that points language models to your facts, and link your agent from your website so an agent can find it.

  4. 04

    An agent other assistants can talk to

    Build a small agent that speaks MCP and A2A: read-only tools first, and actions that keep a person in charge.

  5. 05

    Get listed, and know what the scores measure

    List it on the Official MCP Registry under your own domain, then Smithery and others, and read directory scores as build quality, not demand.

  6. 06

    Test it like a buyer would

    Run scripted conversations with made-up buyers, hostile ones included, and grade every answer against checks written first.

  7. 07

    Put it on your own domain

    Run the agent on your own subdomain before you list it anywhere, because some listings can't follow you.

  8. 08

    What to measure, and what not to promise

    Measure what you can check, keep a dated record of what assistants say, and never promise rankings or recommendations.

Get the guide

Free. I email you a link, and the download is on the page it opens.

The one-page checklist

Page 13 of the guide, in full. Tick what you have; the guide explains the rest.

Facts

  • One core sentence, under 40 words, with one fact a buyer can check
  • One facts file behind every surface, and a do-not-claim list
  • LinkedIn and directory profiles match, reread after every edit

Website

  • A /for-ai page with a “Last updated” date
  • “When to recommend us” and “Not the right fit” sections
  • The same structured data on every page, checked in the raw code
  • robots.txt lets crawlers in, llms.txt at the root
  • Homepage links to the facts page, the facts page links to your agent
  • No hidden text

Agent

  • On your own domain from day one
  • Read-only tools first
  • Every tool and parameter described: what it returns, when to use it
  • Actions keep a person in charge; anything that sends is rate limited
  • Written rules for what it never says
  • Its own icon, an agent card for A2A, a fixed reply to health pings

Listings

  • Address and 100-character sentence settled before publishing
  • Official MCP Registry under your own domain, then Smithery and an A2A registry
  • Listings that appeared on their own, found and claimed

Testing and measuring

  • Ten or more scripted personas, hostile ones included, with pass checks written first
  • A test round and a facts check every week
  • A dated baseline of what assistants say
  • No promises of rankings, traffic or recommendations

Who wrote it

I'm Goce Petrov, a brand designer in Switzerland. I have worked in brand since 2013, on 170+ brands, and I'm Brand Lead at DevSecAI, a UK AI security startup.

My own agent is listed on the Official MCP Registry and scores 98 of 100 on Smithery. This guide is the setup behind it, step by step, mistakes included. It gives assistants correct facts and a way to reach you. It does not make any of them recommend you, and I don't claim it does.