How to Build a Prompt Set That Actually Measures AI Visibility
Teams beginning AI visibility work usually start the same way: they open ChatGPT, type their own brand name, see a reasonable answer, and conclude things are fine. That test tells you almost nothing. Nobody discovers a new vendor by typing a brand they already know.
Measure Discovery, Not Recognition
The queries that matter are the ones a prospect types before they have heard of you. “Best analytics tool for small ecommerce teams” is a discovery query. “What is [YourBrand]” is a recognition query. Only the first reveals whether you are winning new consideration.
A serious measurement programme is built almost entirely from discovery prompts, which is a core principle behind any credible AI search visibility platform.
Cover the Full Range of Intent
A useful prompt set spans several categories. Category queries ask for the best tool in a space. Comparison queries name two or more vendors directly. Problem queries describe a situation without naming a category at all. Attribute queries filter by price, integration, compliance, or region.
Each category surfaces a different competitive set. Brands frequently perform well on category prompts and disappear entirely on problem-led ones, which is exactly the sort of gap worth knowing about.
Scale Beats Precision
AI responses are non-deterministic. The same question asked twice can produce different names. This is why five carefully chosen prompts produce misleading data.
A hundred or more prompts, run repeatedly across several platforms, converts noise into a stable signal. You stop asking “were we mentioned?” and start asking “in what percentage of relevant answers were we mentioned, and is that improving?” The how it works page explains how that measurement is structured.
Test Across Platforms Separately
ChatGPT, Perplexity, Gemini, Copilot, and Google’s AI Overviews draw on different sources and behave differently. Strong visibility in one says little about the others.
Perplexity leans heavily on live retrieval and citation. Others rely more on training data and reason differently about authority. Aggregating them into one number hides the differences that would actually direct your effort.
Track Competitors in the Same Run
Your own mention rate is only half the picture. Knowing which competitors are consistently named — and for which prompt types — tells you where the gap is and often why it exists.
It is common to discover a smaller competitor dominating a specific intent category because they published genuinely useful content on that narrow topic while everyone else wrote generic overviews.
Write Prompts the Way Buyers Talk
Prompts drawn from keyword tools tend to read like keywords. Real users write in full sentences, include constraints, and mention their situation. “We’re a 12-person agency, what project tool handles client billing well?” is far more representative than “best project management software”.
Sales calls and support tickets are the best source for this phrasing, because that is buyers describing their problem in their own words.
Re-run on a Fixed Cadence
A single measurement is a snapshot with no context. Running the same prompt set monthly turns it into a trend, which is the only way to attribute movement to the work you have done.
Keep the core set stable so results stay comparable, and add new prompts as a separate cohort rather than swapping them in. Consistency is what makes the numbers mean anything over time.
Roderick Smith is a writer, blogger, and business owner. He has been writing for over 5 years and his blog naouelmoha.net offers valuable information about the business, health, law, and the latest technology. Roderick lives in Nashville with his wife and three children.
