How to test an AI visibility service with a single client and minimal budget before deciding
There are two bad ways to decide whether your agency offers AI visibility services. The first: diving in because "you have to be there", with a new website, defined packages and a salesperson selling something no one on your team has ever delivered. The second: dismissing it out of scepticism without having tried it, and revisiting the decision every three months when a client asks or a competitor announces it.
The good way is the boring one: an experiment. One client, ninety days, minimal cost, and —this is what almost no one does— the criteria for success and abandonment written down before starting. This article is the full design of that experiment.
The arithmetic: what it really costs to test it
Let's start by dismantling the entry barrier, because it's lower than it looks:
Tool: €55/month. A Surfeo agency account costs €20/month, and adding your first client on the Starter tier, €35/month more. Total: €55/month, €165 for the experiment's quarter. For that price, the pilot client is monitored every week across 3 AIs (ChatGPT, Gemini and Perplexity) with a battery of 40 prompts, analysis of their main competitor and 6 articles a month generated to work the sources. You need no more infrastructure than that for a pilot. Plan details are in pricing.
Hours: around 8-10 the first month, 4-6 the rest. The first month concentrates the work: initial audit, defining the prompt battery, correction plan. After that, the monthly maintenance is reviewing data, executing actions and preparing the report. It's deliberately little: the point of the pilot isn't to give the best service possible, it's to find out how much it costs to give it well.
Commercial risk: close to zero if you pick the pilot well. The ideal candidate is a current client with earned trust, in a sector where people ask the AI (professional services, health, hospitality, training), and one you can tell the truth: "we're adding this service, I want to test it with you for three months; I'll charge you [a reduced rate or nothing] and in exchange you give me honest feedback". Charging something, even a little, has an advantage: a client who pays evaluates seriously; one who doesn't pay smiles and nods.
Total sum of the experiment: €165 in tooling and around 20 hours over the quarter. It's less than attending a conference to "see if this GEO thing is serious", and it produces a better answer.
The criteria: write them before starting (or the experiment is worthless)
This is where most pilots die. Without prior criteria, at ninety days one of two things will happen: if you fancy launching the service, you'll read the data charitably; if you can't be bothered, you'll read it harshly. The experiment only protects you from yourself if the conditions for success and abandonment are written on day one.
A reasonable template, with three criteria on each side:
The pilot is a success if (at day 90):
- The measurement shows attributable movement. The frequency with which the client appears in AI answers has improved against the day-one baseline, or the specific errors detected (false data, absences in their category) have been corrected. Don't ask for miracles: from a typical starting point —in our study of 9,865 SMEs, 91% appeared in only 1 of the 4 AIs— going to appearing regularly in 2 or 3 is already a showable result.
- The client values it without you prompting them. In the day-90 meeting, asked cold ("did this help you? would you pay X a month for it?"), they say yes at a price that leaves you a margin. What price that is, you've worked out in how much to charge for AI visibility services — for reference, the range runs from €300 to €900/month depending on the client, over a tool cost of €35-79 per client.
- The real hours fit in the price. You've logged the hours (log them for real) and the service is delivered in a time that, at the price validated in point 2, leaves a margin equal to or better than your other services.
The pilot is abandoned if (at day 90):
- There's nothing to show. No measurable improvement, no corrections made, no findings the client values. Watch the opposite trap: if the initial audit revealed serious errors and you fixed them, that IS a result, even if the appearance frequency moves slowly.
- The client shrugs. They understand the reports, see the data and still wouldn't pay. If your best client, with the service given away, sees no value, the problem is one of proposal and market, not execution.
- The hours cost blows up. If delivering the pilot has systematically eaten more than double the planned hours and you can't see where to cut them, the theoretical margin was fiction.
A mixed result —two on one side, one on the other? Extend by 60 days with a single change (a different prompt battery, a different deliverable, a different price) and measure again. What doesn't count is extending indefinitely without changing anything: that's no longer an experiment, it's procrastinating on the decision while paying the subscription.
The 90-day calendar
Days 1-15 — baseline. A full initial audit: what each AI answers today about the client and their category, what errors there are, where their competitor stands. Save everything with a date: without a baseline there's no experiment, only anecdotes. The exact procedure is in the first AI visibility audit, step by step.
Days 15-75 — work and weekly measurement. Source corrections (website, Google listing, directories), content that answers the category's real questions, and the monitoring running by itself every week. Resist the temptation to check the data daily: AI answers vary and the signal is in the trend, not the specific Tuesday — why, we explain in can AI visibility be measured?.
Days 75-90 — the report and the question. A closing report with the before and after, a meeting with the client, and the two questions from criterion 2. Note the literal answers: they're the most valuable data the pilot produces.
A shortcut worth knowing: the agency account includes 3 pitch workspaces — spaces to audit a potential client and show them their visibility before signing anything. If you don't want to run the pilot with a paying client, you can do the audit phase with two or three candidates at zero client cost and pick as pilot the one with the worst (best) results: the one who's invisible is the one who'll see the value most clearly.
If at 90 days the answer is yes, the next decision is one of design: a standalone service with its own price, or a layer added to your SEO retainers? Both architectures, with their pros and cons, are in standalone service or SEO extension?.
Frequently asked questions
Why 90 days and not 30?
Because the sources feeding the AIs take time to move: a correction on the website or in directories can take weeks to show in the answers. In 30 days there's only time to do the baseline and the first corrections; measuring the effect requires the quarter. Less than 90 days produces false negatives.
Do I run the pilot for free or charging?
Charge something, even symbolic or at cost. The reason isn't the money: it's that success criterion number 2 —"would you pay for this?"— is answered with facts once there's an invoice involved. A free pilot validates that the service is liked; a paid one validates that it sells.
What if the pilot client improves but I don't know how much is down to me?
Welcome to attribution, the problem of all marketing. That's why the baseline documents specific errors and absences: "the AI said X, we corrected sources Y, now it says Z" is a defensible causal chain even if you can't sign for every percentage point. Promise verifiable work and trends, never surgical attribution.
What exactly do I do each month of the pilot? Is there a script?
Yes: baseline and plan the first month, corrections and content the rest, monthly report always. The week-by-week breakdown of month one —the trickiest, because there are no results to show yet— is in what to deliver in the first month of an AI visibility service.
The experiment starts with the baseline, and the first picture is free: run the AI visibility test on your pilot candidate and decide with their result in front of you. If they come out invisible, you've already got the first slide of the report.
