You ask ChatGPT about your category and you have no idea whether you show up, in what position, with what argument, or who it names instead of you. That blind spot is exactly what a GEO audit fixes: an orderly diagnosis of your presence in AI answers, ending in a plan of what to fix. It’s not a number you check daily —that’s monitoring—; it’s a deep snapshot you take every few months to know where you stand and what to move. This guide is the method we use, step by step, so you can run it yourself or understand what you’re being handed if you delegate it.
What a GEO audit is (and how it differs from measuring your visibility)
A GEO audit is a point-in-time analysis that answers five concrete questions about your brand across generative engines —ChatGPT, Perplexity, Google AI, Claude—: do they mention you? in what kind of questions? in what position and with what argument? which competitors show up instead? and where does the AI pull what it says from? Unlike an SEO audit, it doesn’t look at blue-link rankings: it looks at presence inside the answer the AI writes. And unlike a metrics dashboard, it doesn’t stop at the data: it ends in a prioritized action plan.
It helps not to blur three things people mix up. Measuring visibility is understanding which metrics exist and what they mean —the general framework is in the guide on how to measure AI visibility—. Monitoring is watching those metrics continuously, week over week. The audit is what comes before both: the initial snapshot that tells you your starting point, where the gaps are, and what’s worth fixing first. Without it, you monitor blind and optimize on a hunch.
The 5 phases of a GEO audit
A serious audit reproduces your customer’s real decision scenarios and turns them into data. Five phases, in this order:
- Define the prompt battery. 15-25 real questions your buyer would actually ask —not the ones you wish they’d ask—. This is the phase that decides whether the audit is worth anything.
- Run the prompts in incognito, in each engine. ChatGPT, Perplexity, Google AI, and whichever matters to you, in a session-free window so you don’t drag your history along. The manual incognito method is still the most reliable for the initial diagnosis.
- Tabulate what you see. For each prompt: whether you’re mentioned, in what position, with what argument, which competitors appear, and which sources the AI cites. A spreadsheet is enough.
- Diagnose the gap. Cross the data: do you vanish in category prompts but show up in brand ones? Does the same competitor always beat you? Is the AI citing a listicle or a directory you’re not in? That’s the real problem.
- Prioritize actions. Turn the diagnosis into a short list of what to move first by impact: get into a source the AI cites, fix your entity, reinforce a page. Not fifty tasks: the three that change the answer.
The prompt battery: without it, the audit is opinion
The costliest mistake is auditing with the wrong questions. Your customer doesn’t search for “best CRM in the world”; they search to solve something specific, often without naming you and without naming anyone. A good battery mixes three prompt types:
- Category, without your brand. How someone who doesn’t know you yet asks. Example: “what are the best AI SDR tools for B2B companies in Spain?”. This is where visibility is truly won or lost.
- Comparative. Your sector or your direct competitors against alternatives. Example: “compare [competitor A] vs [competitor B] vs cheaper alternatives for SMBs”. It reveals who the AI pairs you with.
- Problem. How your customer talks when describing the pain without knowing you exist. Example: “my team wastes hours qualifying cold leads, what solutions are there?”. It’s the most honest way in.
If building the set is hard, the GEO prompt battery and monitoring tools give you a starting point, but the final filter is yours: the prompts have to sound like your real buyer, not your brochure.
What to log for each answer
The audit lives or dies on what you record. Five columns, not one for decoration:
| What you log | Why it matters |
|---|---|
| Are you mentioned? (yes/no) | The baseline: you appear —or not— in the answer your customer sees |
| Position in the answer | Coming up first or fifth isn’t the same when the AI gives few options |
| What argument it names you with | Tells you what story the AI has about you —and whether it’s yours |
| Which competitors appear | Reveals who you actually compete with on the AI’s board |
| Which sources it cites | The map of where you need to be: listicles, Wikipedia, reviews, media |
The last column is the most actionable and the one almost everyone skips. If the AI builds its answer on a category listicle you’re not in, you already know your next move —and it isn’t “translate your website”.
How often to repeat it (and why an audit isn’t monitoring)
Models change their behavior every few weeks, so AI visibility isn’t static. But that doesn’t mean auditing daily: a deep audit is worth repeating every 4-6 months, or whenever you change something big (you launch a product, enter a market, publish pillar content). The daily thing is different —watching that you don’t drop or that your narrative doesn’t shift—, and that’s continuous monitoring, not an audit.
If you’d rather not set up the protocol, run the prompts each quarter, and interpret the variance, delegating AI visibility monitoring gives you the snapshot and its tracking without touching a spreadsheet; and when the diagnosis calls for moving content and getting into where the AI looks, GEO optimization does the work, not the report. The audit is the map; we drive.