Setting a Realistic AI Visibility Baseline in Week One
Capture where you stand before changing anything: a fixed prompt set, several runs per engine, citation frequency recorded honestly. Then lock the set.
Setting a realistic AI visibility baseline in week one means capturing where you actually stand before you change anything — a fixed prompt set, run across the engines that matter, with your citation frequency recorded honestly. Without a baseline you can never prove improvement, and you’ll be tempted to invent one later from memory. The work takes a few hours, and the discipline that makes it useful is simple: pick the prompts carefully, run each several times, record everything, and then don’t change the set.
Key takeaway
- Capture the baseline before making changes — afterwards it’s guesswork, and improvement becomes unprovable.
- Build a fixed prompt set of real buyer questions, run each several times per engine, and record citation frequency.
- Expect a low starting number, log competitor presence too, and lock the set so future comparisons are honest.
Build the prompt set first
Your baseline is only as good as the prompts behind it. Choose questions your actual buyers would ask an AI assistant — drawn from sales calls, support queries and the way people describe their problem in their own words — spanning your main topics and the stages of the buying process. Include category questions where you’d hope to be recommended (“best X for Y situation”), problem questions your content addresses, and a few comparison prompts. Twenty to fifty prompts is a workable range for most businesses: enough for signal, small enough to re-run monthly without it becoming a burden.
3–5 runs each
Because AI answers vary between runs, a single check tells you almost nothing. Running each prompt three to five times converts an unreliable yes/no into a citation frequency you can actually track.
Source — AI visibility measurement practice
Run it properly
- Cover the engines that matter to your audience. Typically ChatGPT, Perplexity, Google’s AI surfaces and Copilot. Record each separately, since your position varies by engine.
- Run each prompt several times. Three to five runs, because the same prompt produces different answers. Note how many runs cited you, not just whether one did.
- Record more than presence. Log whether you were cited, how prominently, in what terms, and which competitors appeared instead of or alongside you.
- Save the raw answers. Keep the text of at least some responses. Later you’ll want to see how the description of your brand changed, not just the count.
Expect a low number, and say so
Most businesses starting deliberate AI-visibility work find they’re cited in a small fraction of relevant prompts, sometimes almost none. That’s the normal starting position, not a failure — and recording it honestly is the entire point of a baseline. Inflating the starting number, or quietly dropping the prompts where you performed badly, destroys the comparison you’re trying to build. A low, honest baseline is far more valuable than a flattering one, because every subsequent gain is real and demonstrable. It also sets expectations properly with whoever is funding the work.
Then lock the set
The single rule that preserves the baseline’s value is that the prompt set stays fixed. If you swap prompts between periods, a rise in your citation rate might just mean you replaced hard prompts with easy ones, and the trend becomes meaningless. Keep the same prompts, the same engines and the same method each month. If you genuinely need to add prompts later — new services, new topics — track them as a separate additional set rather than mixing them into the original, so your core trend line stays comparable all the way back to week one.
Frequently asked questions
How do I set an AI visibility baseline?
Build a fixed set of twenty to fifty prompts your buyers would genuinely ask, run each three to five times across the engines that matter to your audience, and record how often you’re cited, how prominently, and which competitors appear. Save some raw answers. Do this before making any changes, then keep the prompt set unchanged so future measurements are comparable.
How many prompts should the baseline include?
Twenty to fifty works for most businesses — enough to produce meaningful signal across your main topics and buying stages, small enough to re-run monthly without becoming a burden. Include category questions where you’d hope to be recommended, problem questions your content addresses, and some comparison prompts. Quality and representativeness matter more than volume.
What if my baseline shows almost no citations?
That’s the normal starting point for most businesses beginning deliberate AI-visibility work, and recording it honestly is exactly what makes the baseline useful. A low, accurate starting number means every subsequent gain is real and demonstrable, and it sets correct expectations with whoever funds the work. Inflating it or dropping poor-performing prompts destroys the comparison you’re building.
Can I change the prompt set later?
Keep the original set fixed, because swapping prompts makes trends meaningless — a rise might just mean you replaced hard prompts with easy ones. If you need to cover new services or topics, track those as a separate additional set rather than mixing them in. That way your core trend line stays comparable all the way back to your week-one baseline.
The bottom line
A useful baseline is a fixed set of real buyer prompts, run several times across the engines that matter, with citation frequency, prominence and competitor presence recorded honestly before you change anything. Expect the number to be low and report it anyway. Then lock the set — because the value of a baseline lies entirely in the comparability of everything measured against it.
We capture a proper baseline in week one, then track the same set every month. Part of our AI Visibility service.
