Questions

Frequently asked questions.

Every finding in the report carries the sentence or record it came from, so you can check it rather than take our word for it.

Is the scan really free?

Yes, and there is no account or card. We pay for the model calls because the scan is how people find us.

Why do you not show the models my website?

Because whoever asks a model about your company does not hand it a URL either. We read your site ourselves and use it as the reference for what is true, not as a hint for the model.

Which AI models do you test?

Four, from two providers, asked the same question with no web access. Where they disagree with each other, that gap is one of the six checks.

My company came back as not recognised. Is that bad?

Less bad than a wrong answer, and easier to fix. There is no strong record to draw on, and the risk is that the absence gets filled with an invented profile the next time somebody asks.

A model described a completely different company. How does that happen?

The public record carrying your name describes them, and the model used it. Nothing failed from the model's side, so it keeps happening until there is a record of your own for it to find.

Do you store or publish my report?

A result is cached for 30 days, so running the same scan again does not repeat the work. We keep a copy on our side. Nothing is published.

Can you remove something bad an AI said about us?

No, and nobody can delete a sentence out of a model. What moves the answer is a correct public record under your name for it to draw on, and that takes time.

Is this the same as AI visibility or share of voice tracking?

No. Those tools track whether your brand comes up when somebody asks a model for recommendations in your category. This one asks whether the answer to your own company name describes the right company with the right facts.

Do you scan individuals?

No. It is built around organisations, where a website and a public entity record give us something firm to check against.

Run a scan

Live scan console

One scan, start to finish

Four language models are asked about the company by name. They have no web access and never see the website. Their answers are then compared against the site and the public entity record.

Recorded sequence, plays onceMotion reduced, finished scan shown

SubjectNordvale Instrumentsnordvale-instruments.example

Models queried

Model AWaitingAnswered 1.2s

“A precision sensor maker, somewhere in the Nordic region.”

Model BWaitingAnswered 2.4s

“A catering equipment supplier, incorporated 2009.”

Model CWaitingAnswered 3.6s

“Catering and kitchen equipment, based in Norwich I believe.”

Model DWaitingAnswered 4.8s

“Industrial gas detection equipment, United Kingdom.”

Answers come from recall alone. The site and the public entity record are read separately. Nordvale Instruments and every quotation here are invented for this example.

01RecognitionAll four models recalled the name and answered.Clear
02Identity accuracyTwo models describe a catering supplier. One of the four is right.Issue
03Cross-system consistencyModels disagree on sector and founding decade.Issue
04Entity anchoringA public record carries this name. It describes a catering supplier.Issue
05Machine-readable identityStructured data on the home page, but no Organization type inside it.Review
06Name contentionThree other companies answer to Nordvale.Issue
04 feeds 02

The record carries the company name and describes a different business. A model that leans on it describes that business, and nothing in the answer looks wrong.

Hollow, clearHalf, reviewSolid, issueColour names the check. The fill names the result.