How much should you read into one AI answer?
The test takes two minutes and the result stings. Before you act on it, it is worth knowing what that answer is evidence of and what it isn't.
There is a post that appears on small business forums every few weeks, and it has the same shape every time. An owner types their own trade and town into ChatGPT out of idle curiosity, three businesses come back, and one of them is the firm down the road. Theirs is not on the list. The replies are always the same too: check your search ranking, get on Bing, post more often. The owner already knows none of that is the problem, which is usually why they posted in the first place.
Before deciding what to do about it, it is worth working out what you have actually learned. One answer is weaker evidence than it feels, and treating it as a verdict leads people to spend money on the wrong thing.
Why it lands harder than a lost ranking
Dropping from second place to fifth is a number on a screen, and most owners can shrug at a number. This is different, because the app wrote a sentence and the sentence had somebody else's name in it. It reads like a judgement even though it is nothing of the kind. Sit with that for a minute and then put it down, because the emotional reading of the result is the least accurate one available. Nobody chose against you.
What a single answer is actually worth
Less than it appears. Ask the same question twice and you may well get two different lists, and asking from another account or with your town spelled slightly differently can change it again. These answers are not stable the way a search ranking is stable, so one run tells you about one run.
What it is not: a score, a ranking, or an opinion about the quality of your work. What it is: a fair sample of the sort of thing a stranger might be told about your trade in your town. If your name fails to appear across five or six attempts, spread over a couple of weeks, that is a real signal even though any single run is noise.
The second question is the one that pays
Most people stop after the first prompt, which is a shame, because the useful information is in the follow-up. Ask the app directly about your business by name. Then push on the things that decide a job: what it thinks your opening hours are, which areas it believes you cover, roughly what it thinks you charge.
This is where owners tend to find the real problem, and it is rarely silence. It is a confident description of a business they half recognise, built from hours that changed two years ago, a service they stopped offering in 2023, and a town they have never worked in. The recommendation failed earlier and more mundanely than anybody assumed.
What people actually change afterwards
The fixes that come up again and again are unglamorous and free. Write the service area out as real place names instead of "and surrounding areas". Make the opening hours agree between the website, the Google profile and every directory that still lists you. Add one plain sentence about the work you do not take on. Then delete the page that has been quietly wrong since before the last van.
None of that requires buying anything, and it is worth doing before you do. If you want the mechanism behind all of it rather than the anecdote, we set it out in how AI apps decide who to recommend.
A word about what happens next
Within a day of posting about this anywhere public, most owners get a message from somebody offering to fix their AI visibility. Some of those people are genuinely useful. Many are selling a technique for a mechanism that nobody outside the model companies can actually see, and the pitch works because the original result felt like a judgement.
Run the second question first. Find out whether your problem is being invisible or being described wrongly, because those are different problems with different fixes, and only one of them is worth paying anyone for.
Written by the team at hailera — we help small businesses answer customer questions without being there.