Every AI search visibility checker on the market does the same thing underneath: it runs a list of questions past several assistants on a schedule and records how often your business gets named. None of them reports a ranking, because assistants do not have one. Understanding that is what separates a sensible purchase from an expensive dashboard, and it also means you can build a usable version yourself in about twenty minutes.
What these products actually measure
A sample, and nothing more than that. The software holds a prompt list, asks each assistant those questions repeatedly, and counts the answers that mention you. The better ones also log which competitors appeared and which web pages Google or the assistant cited as the source.
Two things follow from that. The number is a frequency of appearance across sampled prompts, so anybody presenting it as a position is misrepresenting how the systems work. And the whole result depends on whether the sampled prompts resemble what your customers really ask, which for a regional service business is often where the value falls apart. Our piece on AI search visibility covers the measurement problem underneath all of this.
Four questions before you pay for an AI search visibility checker
Which prompts, and can you supply your own? A product tracking only its own generic list describes a market you do not sell into.
Which assistants, and how often are they sampled? Answers vary between sessions, so one weekly run per assistant produces noise dressed as a trend.
Does it show the cited sources, or only the mention count? Sources are the actionable part, because they tell you which page or directory to go and fix.
And what happens to the history when you cancel? Trend data you cannot export is a reason to stay subscribed, not a reason to subscribe.
The free version is a sheet with five columns
Prompt, assistant, named or not, competitors named, and anything stated about you that is wrong. That is the whole structure, and it answers the same questions the paid tools answer for one business in one market.
Write twelve prompts the way a customer would actually type them: who does this work near Ashland; what does this cost in southern Oregon; how do I choose between two companies doing this; is it a job I could do myself. Then keep the wording fixed for ever, because a changed prompt makes this quarter incomparable with the last one. Our guide to getting cited by ChatGPT covers what the assistants are drawing on when they answer those.
Scoring it so the number means something
Three points when you are the first or only business recommended, one point for appearing somewhere in a list, and nothing at all when you are absent. Then subtract a point for every factual error stated about you, because a confidently wrong phone number does more damage than an absence does.
Twelve prompts across three assistants gives you thirty six runs and a maximum of 108, which is a denominator you can compare against next quarter. The absolute figure means little on its own. What you are reading is the direction of travel over three or four quarters, which is why the wording has to stay frozen. Our piece on brand mentions in AI search covers why those mentions carry the weight they do.
What to do with each of the three outcomes
Nobody local is named at all. In a market the size of the Rogue Valley this is the best result on the list, because the category is open and whoever gets consistent first compounds that advantage while everybody else argues about whether any of it matters.
Competitors are named and you are not. They are described more consistently across the web than you are, which is a corroboration problem rather than a content one, and writing four more blog posts will not touch it. Our guide to fixing NAP issues and cleaning up local citations is the first place to look, and our piece on improving brand visibility in AI search engines covers the rest.
You are named, but something said about you is wrong, and this is the outcome to treat as urgent. Trace the error to its source, which is nearly always an outdated listing or an old page still sitting on your own site, and correct it there, since there is no way to correct the assistant directly.
When a paid checker starts to make sense
Three situations justify the subscription. More than one location, where running the sheet by hand multiplies painfully. More than a handful of service lines, where twelve prompts stop covering the ground. And agencies reporting to clients, who need defensible sampling rather than one person’s quarterly afternoon.
Outside those three, a single location business gets the same decisions out of the spreadsheet for nothing. Revisit it once the fundamentals are done, because a tool telling you that you are invisible is only worth money once you have fixed the reasons. Our piece on AI citations and SEO covers making a page worth quoting in the first place.
If you would rather have the audit run properly and turned into a list of actions, that is part of what our services cover. Our own hours, directions and reviews are on AmpLocal’s Google Business Profile.
Frequently asked questions
Is there a free AI search visibility checker worth using?
Several products offer a free scan of one or two prompts, which is a sales tool rather than a measurement. The spreadsheet above costs nothing and tells you more, because you choose the prompts.
How many prompts do I actually need?
Ten to fifteen covers most single location businesses, provided they spread across buying stages rather than being fifteen versions of one question. Add one for each distinct service you want to be found for.
Should I be checking this weekly?
Quarterly is enough for almost every single location business. Answers vary between sessions, so weekly checks produce movement that looks like a trend and is not, and the underlying fixes take months to show anyway.
Does Search Console report any of this?
Search Console reports impressions and clicks from Google Search, which is a different measurement. Conversations that end in a phone call rather than a visit leave no record for it to report.
What counts as a good score?
There is no benchmark worth quoting, since it depends entirely on your market and your prompts. Compare yourself against your own figure from last quarter and against the competitors appearing in the same answers.
