How to know if AI recommends your brand (and what to do if it doesn't)

Most brands don't know whether AI recommends them. And you can't improve what you don't measure.

While you check your Google rankings, a customer asks ChatGPT for "the best agency for my case" and gets three names. If yours isn't there, you just lost an opportunity that doesn't even show up in your analytics. This is a practical guide to audit your visibility in AI assistants — no paid tools to start — and to know exactly what to do with the result.

What AI visibility is

It's how often and in what way your brand appears (cited, recommended or linked) in conversational engine answers when someone asks for a solution like yours. It's the AI-era equivalent of being on Google's first page: if you're not there, you don't exist for that user.

The difference is that here there aren't ten results. An AI answer names two or three brands and closes. The competition for that space is harder and less visible, and almost nobody is measuring it yet. That's your edge: what you measure before your competition, you win before them.

Why measure before optimizing

Optimizing without a baseline is spending blind. An audit gives you three things no intuition can: where you appear today (your real starting point), who takes your place when you're not there, and what source would make you citable (the specific content AI rewards in your category). Without those three data points, any investment in content or GEO is a bet.

How to audit it in 5 steps

1. Define your key prompts. Write 10 to 20 questions a real potential customer would ask, not the ones you'd ask: "best SEO agency for SaaS", "how to automate customer support for a clinic", "GEO agency in Spain". Mix category and comparison questions. They're the exam you'll put your visibility through.
2. Ask in the four models. Run each prompt in ChatGPT, Perplexity, Gemini and Claude. Use a clean window, no history or account that biases the answers. Repeat each prompt once or twice: you want the pattern, not the lucky hit.
3. Record the result. For each prompt and model note: do you appear? in what position within the answer? with a link to your site? which competitors show up? A spreadsheet is enough; what matters is being systematic.
4. Calculate your share of voice. In what percentage of total answers your brand appears. It's your baseline, the number you'll measure every gain against. Note it with a date.
5. Analyze the why. When AI recommends a competitor, look at which source it cites: an article of theirs, a G2 review, a Reddit thread, a media mention. That source is the exact clue to what you're missing. Don't guess: copy the mechanism, not the text.

How to read the results

Three scenarios, three different diagnoses:

• You don't appear at all. It's an entity and authority problem: AI isn't sure who you are or that you're relevant, so it doesn't risk you. The foundation (structured data, consistent description, presence in sources the model already reads) is unbuilt.
• You appear in passing, at the end or without a link. It's a citable-content problem: the model recognizes you but can't find a fragment of yours clear enough to put you first.
• Your competitor shows up and you don't. It's a specific-gap problem: there's a source that makes them citable and not you. You already know which one (you saw it in step 5) and you can do something better and better structured.

What to do if you don't appear

• If you're not cited at all: start with entity. Structured data (Organization, Product), a description of you identical on and off the site, and mentions in the sources the model consults.
• If you appear in passing: create direct answers to your key prompts, with first-party data and self-contained sentences. Content with original data is cited up to three times more than content that repeats third-party stats.
• If your competitor beats you: study their cited source and publish something more complete, more current and better structured on that same topic.
• In every case: measure again in 30 days with the same prompts. Without re-measuring you don't know whether what you did worked.

When to move from manual audit to monitoring

The manual audit gives you a photo. For video you need tools that track your prompts daily across the four models and alert you when you rise, fall or a competitor enters. It's worth it when AI visibility is already a business channel and every point of share of voice translates into leads. Before that, the monthly spreadsheet is enough and honest.

Example: a real audit, summarized

A brand runs twenty prompts from its category across the four models. It appears in 15% of answers, almost always in passing and without a link; its main competitor shows up in 55%, cited from a very structured guide article. Diagnosis: acceptable entity, nonexistent citable content. Action: five guide-answers with first-party data on the prompts where it was losing. At 90 days, share of voice from 15% to 45%. There was no magic: there was measurement, diagnosis and the specific missing piece.

What you can do today

Pick five questions your ideal customer would ask, run them in the four models with a clean window and note whether you appear and who shows up in your place. In twenty minutes you have your first baseline and, with it, the exact list of what's missing. That's already more than 90% of your competitors measure.

Frequently asked questions