Interview an AI about your own business. It is uncomfortable and it is free.
Somewhere out there, an AI is describing your business to people who have never heard of you.
It is doing this now, several times a day if you are lucky, and you have almost certainly never checked what it says. Nobody has. It is a strange gap, because if a human employee were introducing your business to strangers you would want to know roughly what they were telling them.
So here is a structured way to find out. Set aside half an hour. You will need a pen.
How to run it
Use at least two of these: ChatGPT, Google’s AI Mode, Perplexity, Gemini, Claude.
That matters more than it sounds. Some of them go and look things up live. Others lean more on what they already learned during training, which may be a year or two out of date. Different answers from different tools tell you different things: a live tool is showing you what is currently findable, a training-heavy one is showing you what has already stuck.
Two other rules before you start.
Run each stage twice. These systems are not deterministic and you will get different answers. You are looking for patterns, not single verdicts.
Do not sign in, or use a private window. Otherwise you are seeing a version influenced by everything else you have ever asked it, which is not what a stranger gets.
Write the answers down. You are looking for gaps, not grades.
Stage one: the cold open
Start with no context at all. This is your baseline.
“What do you know about [business name] in [town]?”
Then, whatever comes back, the single most useful follow-up in this entire exercise:
“What are you unsure about, or could not find out?”
That second question is where the value is. A confident-sounding answer hides its gaps. Asking directly makes them visible.
What to look for:
Is it actually your business, or has it merged you with somebody of a similar name? This happens constantly and it is worth knowing.
Is anything simply wrong? Old address, wrong services, a phone number you gave up in 2019?
Is it vague where it should be specific? “They offer a range of services” means it could not find out what you actually do.
Score it honestly: would this description make somebody want to ring you?
Stage two: the accuracy audit
Now go through what it told you, line by line.
“You said [X] about my business. Where did that come from?”
This one is worth the whole exercise. You find out which sources are shaping your reputation, and the answer is frequently not your website.
I have seen this turn up a directory listing nobody had touched in six years, a Facebook page with the wrong opening hours, and in one case a review site profile the owner did not know existed.
Follow up with:
“Is there anything you found about this business that contradicts something else you found?”
Inconsistency is the quiet killer here. If your website says one thing, your Google profile says another, and an old directory says a third, a machine has to pick, and it may not pick the one you would have chosen.
Action: every wrong source is a job. Claim it, correct it, or get it taken down.
Stage three: the buying scenario
This is the one that matters commercially, because it simulates what actually happens.
Ask it the way a customer would, without naming yourself:
“I need [your service] in [your town]. Who would you recommend and why?”
Then the follow-ups:
“Why did you choose those, and what would make you more confident recommending one over the others?”
“What information would you need before recommending someone for this?”
What to look for:
Did you appear at all? If not, that is the headline finding, and it is a content and visibility problem rather than a website design problem.
If you did appear, why? What did it cite as the reason? That tells you which of your material is actually doing work.
And if somebody else came top, go and look at what they have published that you have not. Usually it is not clever. Usually they have simply written down something you have not got round to.
Stage four: the hard questions
Now make it uncomfortable, because this is where the real gaps show.
“What are the risks or downsides of using [business name]? What would you want to know before committing?”
“What questions does their website fail to answer?”
“If I were being cautious, what would put me off?”
These are unpleasant to read and they are the most useful outputs of the whole session.
An AI answering these is effectively listing the objections in a cautious buyer’s head, and it is doing it based on the absence of information rather than the presence of anything bad. Nine times out of ten the answer is not “they seem terrible,” it is “there is no pricing, no reviews visible on the site, and no way to tell how long they have been trading.”
Every one of those is a page you have not written.
Stage five: the comparison
“Compare [your business] with [competitor] and [competitor]. Who would you recommend for someone who wants [the thing your best customers want]?”
Then:
“What does each of them make clear that the others do not?”
That last question is the gold. It shows you exactly what your competitors have communicated well, in a machine’s own words, which is a considerably more honest read than looking at their website yourself and deciding it is a bit rubbish.
Warning: do not use anything you find here to write comparison content naming those competitors. Publishing factual claims about identifiable local businesses is a genuinely risky thing to do, and machine-generated claims about them are exactly the sort you would struggle to defend. Use it for your own diagnosis and nothing else.
Stage six: the retest
Do all of that again in three months.
That is the actual point of this exercise. A single snapshot tells you where you stand. Two snapshots tell you whether anything you did made a difference, which is the only way to know if any of this is working.
How to read your results
Three things worth keeping in mind so you do not over-react.
A bad answer is usually an absence, not an insult. These systems describe what they can find. If the description is thin, that is because the evidence is thin, not because something has decided against you. That is fixable, and it is mostly fixable by writing things down.
One bad run means nothing. Ask again. Ask on a different platform. Patterns matter, single answers do not.
And being invisible is normal. Most small businesses currently are. If you come out of this session with a list of things a machine could not find out about you, you have not failed the test, you have just been handed a content plan.
The uncomfortable bit
The reason I like this exercise is that it removes the arguing.
I can tell somebody their website does not explain what they charge, and they will say their customers know, or that it depends, or that they would rather have the conversation. All fair.
But when an AI says, in plain English, “I could not determine their pricing, so I recommended the other company who publish theirs,” it lands differently. It is not my opinion. It is a demonstration.
Half an hour, no cost, and you will finish it knowing exactly what to write next.
Want somebody to run this properly and tell you what it means? Get your free audit, or email hello@inkon.co.uk. I read every message myself.