We ran the same 50-question AI-visibility audit on one local retailer three times in a month. Correcting the facts AI kept getting wrong was fast and stuck — but staying the top recommendation turned out to be the harder, ongoing fight.
An AI engine was confidently feeding potential customers misleading information about a business we work with. Not once — in nearly half of the answers where the business came up.
The business is an independent scuba and dive-training retailer: a couple of physical storefronts plus a high-traffic nationwide e-commerce site. Real reviews, real expertise, real in-house repair work most competitors don't offer. And yet, when buyers asked AI engines about it, the answers were often wrong in a small but trust-eroding way.
So we did something most businesses never do: we measured it — three times.
We built a 50-question audit out of the things a real buyer would actually type — "best place to buy a regulator online," "scuba certification near me," "is this shop legit." Then we ran that exact same set of 50 questions three times over about a month, scoring each answer on whether the business showed up, where it ranked, and whether what the AI said was accurate.
Three snapshots on identical questions is the part that matters. AI answers wobble from run to run, so any single audit is a coin-flip's worth of noise. The trend only shows up when you ask the same things, the same way, more than once.
The first surprise was a good one: this wasn't a discovery problem. The business was named in the top spot in 80% of answers and mentioned in 88%. AI search already knew it existed.
The problem was accuracy. Of the answers that mentioned the business, only about half were correct. 22 of 50 answers contained misleading information about the business — repeated as if it were settled truth.
Being found isn't the same as being described correctly — and the second one quietly costs you trust on exactly the "is this place legit?" questions where it matters most.
The diagnosis was clear: not a visibility problem, an accuracy problem. And accuracy is fixable.
We corrected the ground-truth facts at the source — the places these engines actually pull from — and re-ran the same 50 questions.
The wrong answers dropped from 22 to 3 in about two weeks. By the third audit, they were down to 1. Accuracy went from roughly 48% → 93% → 98%.
Fix the facts at the source, and AI engines re-cite the corrected version quickly — and keep citing it.
This is the most reassuring lesson of the whole exercise: the thing that feels scariest — an AI confidently repeating something false about you — is often the thing you can fix fastest. The correction propagated, and it stuck.
Here's what we didn't expect, and won't paper over.
While accuracy climbed, top-spot placement slid the other way: 80% → 74% → 66%. The business was still mentioned in about 90% of answers — it wasn't disappearing — but it was increasingly the second or third name, not the first. By the third audit, roughly a dozen new competitor names had started showing up in the answers.
That's the part a single audit would have hidden. Across three identical runs, it stopped looking like noise and started looking like a trend: other businesses are entering these AI answers, and the top recommendation is being contested.
Accuracy is a cleanup you can finish. Being the one the AI recommends is a position you have to keep earning.
This one business, measured three times, surfaces a distinction most owners never get to see:
Most businesses aren't doing either yet. They've never seen how AI describes them, and they've certainly never tracked whether they're winning or losing the top spot over time. That's the opportunity: the businesses that start measuring now find out where they actually stand before their market catches on.
A footnote in the honest direction: AI-referred traffic is still a small slice of total visits for most businesses today. This isn't about a flood of clicks tomorrow — it's about holding position in the channel customers are increasingly using to decide who to trust.
You don't have to fix everything to start. You have to look — and then keep looking.
Ask the AI engines your customers use the questions a real buyer would type. Are you there? Is what they say true? Are you the first name, or has a weaker competitor slipped in ahead of you? Then check again next month, on the same questions, so you can tell a real change from a random one.
The facts you can fix fast. The top spot you have to defend. Both start with measuring what AI actually says about you — more than once.