A do-it-yourself guide
Audit your AI visibility in an afternoon
You don't need a tool or a budget to get a real first read on what AI says about you. Here's the afternoon version, and exactly where its limits are.
The promise
What an afternoon can actually tell you
You can get an honest first read on what AI says about your company today, yourself, for nothing. It won't be a rigorous baseline — that takes more, and the last section is straight with you about the difference — but in an afternoon you can find out whether assistants recognise you, whether what they say is true, and who they recommend instead of you. That's often enough to know whether you have a problem.
Here's how to do it properly, in five steps.
Step 1 — the questions
Ask what a buyer asks, not your own name
The most common mistake is typing your company name. That tests whether the model can read your homepage, not whether you're visible. Instead, ask the questions a real buyer asks before they've heard of you — "what's the best tool for [your job]?", "how do I solve [the problem you solve]?", "[a competitor] alternatives". Write down five to ten of these. They're the queries where being named actually matters.
Step 2 — a clean room
Sign out, search on, same words
Two settings decide whether your read is real. Sign out, or turn memory off — logged in, an assistant will flatter you with your own history and you'll see a version of yourself no buyer gets. And turn web search on through the assistant's own control — never by typing "search the web" into the question, because that changes the question and breaks the comparison across engines. Then ask each question word-for-word the same everywhere.
Step 3 — across the engines
The same questions, six assistants
Ask your list across the major assistants — ChatGPT, Perplexity, Claude, Gemini, Copilot, and Google's AI Mode — because they don't agree with each other. For each answer, also ask it to "list every source you used, with full URLs," and save the result (printing to PDF is the simplest way to keep a copy). You're building a small record you can actually read, not a vibe.
Step 4 — read it
Four questions for every answer
For each answer, ask four things — the same four a proper audit is built around:
- Were you named, without being prompted? If not, you're not yet recognised in the category.
- Is what it says true? Confident and wrong is the costly kind.
- Whose sources is it using? Often not yours — that's who owns your narrative.
- Who shows up instead of you? That's the comparison the buyer sees.
Step 5 — more than once
One answer is an anecdote
Assistants drift between sessions — ask the same question twice and you can get two answers. So don't trust a single response. Check each engine, and check on more than one day. The pattern across engines and days is the finding; any one answer is just an anecdote.
The honest limits
What the afternoon version isn't
An afternoon gets you a real snapshot — enough to spot the obvious problems and decide whether to go further. It is not a reliable baseline, and it's worth being clear why, so you don't over-trust it. You're one reader scoring your own answers, on one day, by feel — no fixed rubric, no second rater to check your judgement, no way to reproduce the exact run later or prove a change moved the number. Those are the things that turn a snapshot into a measurement you can stand behind over time, and they're the difference between checking and auditing.
But that's the next problem. The first one is just finding out — and now you can, this afternoon.
This guide is the short version of the open course, which goes deeper on each step. Written in the open; no tool to buy.