A client's first AI visibility audit, step by step (with template)

The first AI visibility audit is the document that decides whether a client hires the service, understands it and values it — or whether it all stays a meeting-room curiosity. And unlike the SEO audit, here there's no fifteen years of inherited templates: most agencies are improvising their own.

This article is the template. It defines which prompts to test, what to note from each answer, how to look at competitors and sources, and how to present the result so it's understood. All done by hand: for the first audit you don't need a tool, you need a method. (If the context is pre-sales —auditing a prospect before signing— the commercial approach is in the audit as a hook to close clients; here goes the technical procedure, which is the same.)

A definition before we start, in case the client asks: this discipline is called GEO (Generative Engine Optimization) — measuring and improving how a brand appears in the answers of AI search engines like ChatGPT, Gemini, Perplexity or Claude.

Step 0: what you need before opening any AI

• Who the client's end customer is and what they'd ask an AI (not which keywords they search: what phrases they'd say).
• Their 2-3 real competitors — the ones they name, not just the ones ranking in Google.
• Their basic verified data: address, opening hours, services, indicative prices. You'll need them to spot when an AI gives false facts.
• A blank spreadsheet with these columns: prompt · AI · appears? · position/context · what's said · data correct? · competitors mentioned · sources cited · date.

Step 1: build the prompt list (15-25)

The quality of the audit is the quality of the prompts. Cover these five types:

• Generic category (5-8): "best [service] in [city]", "what [type of business] do you recommend for [need]?". These are the ones that bring new clients.
• Problem (4-6): the need without naming the category: "my boiler's broken, who fixes it in Seville?", "I need to invoice as a freelancer, what software do I use?".
• Comparative (3-4): "[competitor] vs alternatives", "is X or Y better?", "alternatives to [sector leader]".
• Brand (2-3): "what do you know about [client]?", "is [client] trustworthy?". Here you're not measuring visibility but accuracy: what the AI says when it's already being asked about them.
• Imminent purchase (2-3): "how much does [service] cost in [city]?", "where do I hire [service] today?".

Writing rules: natural language (people write to AIs as they speak), in Spanish, with the city when the business is local, and without putting the client's name in the generics — that would be cheating at solitaire.

Step 2: run each prompt across the 4 AIs and note it

The 4: ChatGPT, Gemini, Perplexity and Claude. One won't do — in our study of 9,865 Spanish SMEs across 30 sectors, 91% appeared in only 1 of the 4 (full study) — so a single AI's picture almost always misleads.

From each answer, note in the sheet:

• Does the client appear? Yes / no / only if named.
• In what position and context? First recommendation, one option among five, a passing mention.
• What exactly does it say about them? Copy the literal sentence. This is where the gems and the scares show up.
• Is the data correct? Address, hours, services, prices. Mark every error in red: they're the most actionable findings of the audit.
• Which competitors appear? All of them, including the ones the client didn't have on the radar.
• Which sources does the AI cite? Perplexity and Gemini usually link; note them down — they're the map of where to work next.
• Dated screenshot, filed by prompt and AI.

Execution discipline: a clean session (no history to contaminate it), all queries the same day, and resist the temptation to re-ask until what you want comes out. The audit photographs what's there, not what you'd like there to be.

Step 3: run the generic prompts on the competitors

Using the same sheet, measure the 2-3 competitors on the generic and comparative prompts. You're after two numbers: in how many prompts each one appears against the client, and with what description. If a competitor shows up systematically and the client doesn't, that contrast will be the most important slide in the presentation — and the conversation that follows has its own script.

Step 4: audit the sources the AIs draw from

AIs don't invent answers out of nothing: they compose them from sources. Review the client's:

• Their website: does it clearly and extractably say what they do, where and for whom? Does it have structured data (the markup that helps machines understand the page)?
• Listings and directories: Google Business Profile, sector directories — do they exist, are they up to date, do they match each other?
• Reviews: volume, rating and freshness on the platforms in their sector.
• Third-party mentions: press, rankings, forums, comparisons. It's the material AIs cite most and where almost every SME is at zero.

Cross this list with the sources you noted in step 2: where the AI cites sources the client isn't on, you've got the service's task list.

Step 5: present the results (the document's structure)

The classic mistake is handing over the spreadsheet. The sheet is yours; the client gets a 6-10 page document with this structure:

1. One-page summary: in how many of the X prompts they appear, in which AIs, the 2-3 most serious findings (false data first) and the picture against competitors. If they only read one page, make it this one.
2. The screenshots that hurt: 4-6 annotated examples — where they don't appear, where the competitor does, where there's a false fact.
3. The why: which sources are missing or failing (step 4), explained without jargon.
4. The plan: what would be done in the first 90 days, in order. Without promising appearances — promising measurable work.
5. Annex: the full table of prompts × AIs, for anyone who wants the detail.

In the meeting, open with page 1 and leave the screenshots on screen as long as possible: this audit sells itself when it's seen.

A note of honesty to close the document: this audit is a dated photo, and AI answers change from one week to the next. That's why the audit is the beginning of the service and not the service: the continuous version —the same prompts, the 4 AIs, every week, with comparable history— is what we built in Surfeo for agencies, where each client has their space with weekly tracking and a PDF report for your reporting. The hours arithmetic of doing it by hand every week is done for you here.

Frequently asked questions

How long does this audit take by hand?

Between 3 and 5 hours: one preparing prompts, one or two running and noting (20 prompts × 4 AIs is 80 queries), and one or two on analysis and the document. Don't compress it by dropping AIs or prompts: an audit of 5 prompts in one AI won't survive the client's second question.

Don't the answers change depending on who asks? Is the picture worth anything?

They vary between sessions and phrasings, yes, and the audit must say so explicitly. That's why it's all run the same day, in a clean session and noting the date: you're not measuring an eternal truth, you're fixing an honest baseline to compare against a month from now.

Do I charge for the audit or give it away?

Depends what you use it for. As a pre-sales hook with a prospect, free and reduced (8-12 prompts). As the start of a signed service, it's included in the first month — that's what it's for, the centrepiece of what you deliver in the first weeks. As a standalone product, €300-600 is a defensible range for the hours it takes.

What do I do if the audit comes out "well" and the client appears in almost everything?

First, congratulations: it's statistically rare. Second, the audit still has value: it documents the position (which no one watches), the incorrect data that almost always shows up in the brand prompts, and the volatility — being there today doesn't guarantee being there next month. The service shifts from "getting them to appear" to "watch and defend", which is also billed.

---

If you want to see how a first pass looks before investing the 4 hours, drop the client's website into the free AI visibility test: you'll have the initial picture in minutes to build the full audit on.