What does a free AI visibility score tell me?
That the assistants named you, or didn't, on the day the tool ran. It is a real measurement of one thing, taken once.
Several platforms now offer a free AI visibility score. You type in a brand name or a website, the tool puts some questions to ChatGPT and a few other assistants, and a score comes back within minutes, with a competitor comparison and some recommendations. The upgrade is a subscription.
The score is real. Whatever questions the tool asked, the assistants either named you or they didn't, and the number reflects that. If you have never checked, a low score is worth knowing about. So is a high one.
Here is what the number leaves out.
It is one run
Ask an assistant the same question twice and you will often get two different answers. When I put an identical set of questions to the same four assistants on two separate occasions, with nothing changed in between, the results differed by about two points. A single run cannot tell you whether the gap between you and a competitor is real, or whether the change you made last month moved anything. You need the same questions asked many times over, and you need to know how much the answers move on their own before you read anything into a movement.
It counts mentions
Tracy Park, a wedding and golf venue near Bath, ceased trading on 30 September 2024. In July 2026, four assistants were still recommending it, in 98 recorded answers. Counted as mentions, that is a strong score. Read as answers, it is a venue that has not existed for two years being put on couples' shortlists.
The finding came from reading the answers. The same reading, in a Zürich study the same month, found an assistant recommending a business that has never existed. A count of how often your name appears will not show you either of those things, and the second one may be taking the recommendation that should have been yours.
It does not say where the answer came from
An assistant builds its answer from two places: what it absorbed during training, and what it reads on the web if it searches. The two fail differently. A website the assistant cannot read is a search problem. A venue that closed in 2024 and is still recommended is a memory problem, because the training data never heard the news. A score that blends the two cannot tell you which one you have, and the fix for one does nothing for the other.
The split matters for a second reason. Many of the products built on these assistants have no web search at all, and the ones that do often answer advice questions from memory anyway. What an assistant already believes about your business sits underneath whatever it finds when it searches.
Ask any tool whether it separates the two. If it can't, the recommendations at the bottom of the report were written without knowing which problem you have.
Look at what the upgrade includes
Read the upgrade page before you read the recommendations. If the same subscription sells the measurement and the tools to move it, the measurement is marking its own homework. The point is structural rather than personal: a number that goes up when you buy more of the seller's product is a number you cannot use to judge the seller.
I price measurement and fixes separately for this reason. If your own agency does the work, my figures judge it the same way.
When a platform is the right tool
If you have a marketing team, several product lines and buyers in more than one country, and you want a number every week across ten assistants, buy a platform. It will be cheaper per answer than I am, and if you email me describing that situation I will say so.
What I do is slower and narrower. A set of realistic customer questions for your business, put to four assistants many times over, every answer recorded, every figure traceable to the recorded answer it came from, and a person who has read the answers writing down what they mean. The first look is free here too. It takes days rather than minutes, because someone reads every answer.
If you have already run a free score, send it to me along with the report it came with. I'll tell you what it caught and what it missed, at no charge.