Every question your buyer asks before they call you.
One prompt tells you almost nothing. A set tells you where the gaps are. Pick your category and city and get the full run, grouped by where the buyer is in the decision.
A prompt set is the fixed list of questions you test your AI visibility against, run the same way every time so the results can be compared. It is the difference between a spot check and a baseline.
Most owners test one question, usually the hardest one: who is the best in town. They are not named, they conclude AI does not know them, and they stop. The more useful finding is almost always further down the decision. A business can be absent from the discovery question and named on the price question, or the reverse, and those two gaps call for completely different work.
Run the whole set, write down what you see, and you have a baseline you can re-run in 90 days.
Not seeing your category? Type it anyway. The set still builds, with generic wording for the trust and price stages.
Edit the wording to match how your customers actually talk. What matters is that the set stays fixed once you start, so the run you do today can be compared to the one you do in 90 days.
Specialization Teams and practices only
If individual people at your business carry specialties, a real estate team, a therapy practice, a law firm, a dental office, the set above only tells you whether the business gets named. Add one prompt per specialty, phrased as the buyer with that exact need, without naming the person. The engine has to make the association itself.
“I’m a first-time buyer with almost nothing saved for a down payment. Which agent in your city actually works with buyers like me?”
“I need a therapist in your city who specializes in postpartum anxiety and takes new clients.”
Read it in three layers: does the business get named, does anyone on the team get named, and does the right person get named for their specialty. If a teammate wins the prompt instead, that’s a differentiation problem, and the fix is sharpening each person’s page so the specialty attaches to one name. If nobody gets named, it is the standard gap, and the stages above locate it.
An hour, once. Then every quarter.
Size the set to how you’ll run it
Running by hand, start with two per stage, twelve total. That is an hour per baseline and enough to find which stage is broken. Using a tracking tool that runs prompts daily, take the full set and more, because engines answer differently run to run, and frequency across a bigger set is what turns noise into a trend. Either way, hold the set steady: a changed set breaks the comparison.
Start in a logged-out window
A signed-in session personalizes results based on your history, and you are the one person who searches for your own business. Open a private window so you see what a stranger sees. If an engine makes you sign in partway through, use an account with no history around your business, or finish that engine on another day.
Run all three engines
ChatGPT, Perplexity, and Google AI Mode weight sources differently. Being named on one and missing on another is the most useful signal in the whole exercise, because it tells you which surfaces are thin.
Record everyone who gets named
Write down every business the engine mentions. That list is your real competitive set, and it is frequently not the one you assumed.
Note the sources it cites
Click through. The pages being cited tell you what kind of content earns a citation in your category, which is more actionable than the ranking itself.
Screenshot everything
Answers change between runs and between days. A screenshot is your evidence, and in 90 days it is your before.
What each pattern actually means.
Named on branded, absent everywhere else
The engines know you exist but do not consider you a candidate. This is the most common result and the one the audit is built for. It is a consideration problem: they know you exist and don’t yet pick you.
Absent on the broad question, named on the specific one
This looks like failure and is often the opposite. The big query has the most volume and the most competition, and being eighth in a list of ten is not being named. The narrow query is where someone has already decided what they want and is looking for who does it. Cited issue 17 follows a trainer who owns exactly that one.
Named on one engine only
Your footprint is concentrated in whatever that engine leans on. Usually it means Google Business Profile is strong and third-party corroboration is thin, or the reverse.
Named but described wrong
Worse than absent in some ways, because it is being repeated. Almost always an entity consistency problem, and the fastest thing on this list to fix.
Competitors named that you have never heard of
They are being cited by something you are not on. Click their citations and find out what.
Nobody named at all
The engine hedged and gave general advice instead of names. That is an open category, and it is the best possible position to move into.
Before you run the set.
What is a prompt set?
A prompt set is a fixed list of the questions a real customer would ask an AI assistant before choosing a business like yours, run the same way each time so results can be compared across engines and over time. It covers the whole decision rather than the single question most people test, which is who is the best in town.
How many prompts do I need?
Fifteen to twenty is enough to see a pattern for one category in one market. Fewer than ten and a single odd answer skews the whole picture. More than thirty and the exercise becomes something you will not repeat, which defeats the purpose.
How often should I re-run it?
Quarterly is the right rhythm for most businesses. AI systems need time to crawl, retrieve, and trust changes, so anything shorter than 60 days is measuring noise. Running it monthly is only worth it when active work is underway and you want an early read.
Will I get the same answer twice?
No, and that is expected. Answer engines are not deterministic the way a search ranking is, so the same question can return different sources between requests. This is why a set matters more than a single prompt. You are looking for patterns across fifteen answers, not a verdict from one.
Should I be signed in when I run these?
Start signed out. A signed-in session personalizes results based on your history, and your own history is the least representative there is. Most engines will let you run several prompts in a private window before asking you to sign in. When that happens, sign in with an account that has no history with your business, or run the rest of the set on a different engine and come back later. What matters is that you never run them from the account you use to search for your own name.
What if I am not named anywhere?
That is the common result and it is not a dead end. It means the engines have not been given enough to work with, which is a supply problem rather than a penalty. The work is entity clarity, structured data, and content specific enough to quote, in that order.
Is this the same as the tracked baseline in the audit?
It is the same idea done by hand. The audit stands up a monitored prompt set in AI visibility tracking software, runs it across the engines through the engagement, and reports visibility score, share of voice, and which domains are cited instead of you. Running the set yourself gives you the shape of the problem. The tracked version gives you the measurement.
Running the set tells you where the gaps are. It does not tell you what to fix first.
See every gap, scored.
The AI Visibility Audit tracks a prompt set across ChatGPT, Google AI Overviews, and Perplexity for seven days, scores your AI visibility across six areas and 30 criteria, and sets what to fix first and who should do it.