How to choose the prompts you'll monitor for a client (and how many is enough)
The prompt list is to the AI visibility service what the keyword list was to SEO: the decision that shapes everything else. If you monitor the wrong questions, it doesn't matter how well you measure — you'll be measuring something nobody cares about, with great precision.
And there's an important difference from keywords: here there's no planner telling you search volumes. The list is built with judgement, not with an export. This article is that judgement: the five types of prompt any list should cover, how many of each by client size, and the mistakes we see over and over.
A prompt, to be clear, is the question exactly as the end customer would put it to an AI: "which online accountancy firm would you recommend for a self-employed worker?", not "online accountancy self-employed". That difference in phrasing is the first rule of all.
The five types of prompt
1. Discovery. The user doesn't know your client and asks for a category recommendation: "best invoicing software for SMEs", "which dental clinic would you recommend for implants?". These are the prompts that bring new clients in and should be the backbone of the list. It's also where almost everyone is absent: in our study of 9,865 Spanish SMEs, 91% only appeared in 1 of the 4 AIs (full study).
2. Comparison. The user is already weighing options: "[competitor] vs [client]", "alternatives to [sector leader]", "is X or Y better?". Lower volume but higher intent: someone comparing is close to buying. Always include one with the sector leader even if your client doesn't resemble them — it's where the AI tends to list alternatives, and there's room there.
3. Local. The category plus the geography: "emergency locksmith in Zaragoza", "where can I rent a storage unit near Chamberí?". For a local business, this type overlaps with discovery and can be half the list. For a national B2B or a SaaS, it can be zero. It's the type that varies most between clients.
4. Problem. The need expressed without naming the category: "my boiler has broken and I live in Seville, who can fix it?", "I need to invoice as a self-employed worker and don't know which software to use". These are the prompts closest to how people actually talk to an AI, and the ones that best reveal whether the client's sources explain what they solve, not just what they sell.
5. Brand. "What do you know about [client]?", "is [client] trustworthy?", "reviews of [client]". Here you're not measuring visibility — they'll almost always come up, because you've given the name — but accuracy: what the AI says when it's already being asked about them. This is where false data shows up, which is the most urgent finding to correct. Few but compulsory.
How many prompts by client size
More isn't better: every added prompt dilutes attention and bloats the report without adding a decision. The right question isn't "how many can I monitor?" but "how many can I act on?".
Local or single-service business: 20-40 prompts. Twenty is the minimum to cover a sector's commercial questions; with 40 you cover every reasonable way a prospect can reach them. Rough split at the top end: 10-12 discovery, 10-12 local, 6-8 problem, 4-5 comparison, 2-3 brand.
SME with several service lines, several cities or a B2B with several audiences: 60-75 prompts. The split multiplies per line or per city: each relevant service deserves its discovery and problem prompts, and each location its local ones. Above 75-80 you're almost always adding cosmetic variations of the same question.
It's no coincidence that those two ranges match the Surfeo plans — 40 prompts on the Starter tier, 75 on Growth: they come from the same arithmetic. Below 20 the picture has holes; above 80, noise.
A legitimate shortcut: the first audit's list is the draft of the monitoring list. If you did the initial audit step by step, you already have 15-25 tested prompts; expand from there in the types that came up short.
The typical mistakes (and what they cost)
Putting the client's name in discovery prompts. "Would you recommend [client] for renovations in Granada?" guarantees a mention and ruins the measurement: you're measuring whether the AI is polite, not whether it recommends the client. The name only goes in brand prompts and the odd comparison one.
Translating SEO keywords as-is. "employment lawyer madrid price" isn't a question anyone asks an AI. Rewrite each intent as a spoken sentence, with its context: "how much might an employment lawyer in Madrid cost me for a dismissal?".
Overloading the list with brand prompts. Coming up in "what do you know about [client]?" feels reassuring and is worth nothing commercially. If a third of your list is brand prompts, a third of your report is self-congratulation. Two or three are enough.
Monitoring prompts where the client can't compete yet. "Best hotel in Spain" for a twelve-room guesthouse will only produce months of zeros. Better the prompt where there's an actual game: "small charming hotel in [district]". What's achievable at 3, 6 and 12 months depends heavily on the sector — the realistic frameworks are here.
Changing the list every month. Every prompt you change loses its history, and the history is what lets you answer "am I better than three months ago?" with a chart. Rule of thumb: review the list quarterly and don't touch more than 10-20% at once.
Frequently asked questions
Can I use the same list for every client in the same sector?
As a starting template, yes — the types and the structure repeat. But personalise at least a third: the comparison ones (their real competitors, not the sector's in general), the local ones (their locations) and the problem ones (what their customers actually ask, which they know better than you: ask them at the kick-off).
What language do I write the prompts in?
In the language of the client's end customers. For a Spanish SME, in Spanish and naturally. If a relevant part of their audience asks in English — tourism, software — duplicate the key discovery prompts in English rather than translating the whole list.
Do I include prompts where the client already does well?
Yes, always. First because AI visibility is volatile and being there today doesn't guarantee being there next month: watching what you've won is part of the service. And second because those prompts are your proof of value in the report — defending a position is also billable.
How often do I run the list once it's defined?
Every week, always the same, across the 4 AIs. The frequency and the discipline of execution matter as much as the list; the reasoning for why (and why doing it by hand doesn't scale) is in how to monitor all your clients without screenshots.
If you want a starting point for a specific client's list, run their site through the free AI visibility test: you'll see which prompts they appear in (and which they don't) before you sit down to draft the definitive list.
