01 · Who you areCorrect
GPT-6 Sol OpenAI · ChatGPT
Acme Coffee is a Spanish online specialty coffee roaster based in Valencia, founded in 2019. It sells single-origin beans by subscription, roasts to order and ships within 48 hours across Spain.
Free AI brand benchmark · no account
What do AIs know
about your brand?
Your buyers ask ChatGPT, Claude, Gemini or Perplexity before they reach you. We ask nineteen models six questions about your company and send you every answer, word for word, with the errors marked. The questions are the same for every brand, so the report is a benchmark: you can set yourself against your competitors, and against your own next edition.
Who answers
- GPT-6 SolOpenAI · ChatGPT
- GPT-6 AstraOpenAI · ChatGPT
- Claude Opus 5.5Anthropic · Claude
- Claude Sonnet 5Anthropic · Claude
- Gemini 3.8 FlashGoogle · Gemini
- SonarPerplexity · searches the web
- + 13 open modelsServed from EU infrastructure
Called through the API with web search off, so you read what each model remembers. Sonar is the exception: Perplexity searches before it answers, and the report lists its sources.
Request your reportSix questions · nineteen models · about five minutes
Want a quick look first? Get your AI visibility score from 0 to 100, live on the page and without leaving an email.
What you get back
A sample report,
before you ask for yours.
An invented coffee brand run through the same six questions. The layout, the four states and the closing correction are the real report's; the brand, its competitors and the answers are made up, so no company is on display without having asked. The full sample report is one click away.
Sample · fictional brandAcme Coffee · specialty coffee eCommerce
9 / 19described the brand correctly or with one error
3 / 17named it when asked for a brand like it, without being prompted
6 / 19confused it with another company
Question 01 · Who you are · one square per model
- Correct
- Partial
- Confuses it
- No usable answer
01 · Who you arePartial
GPT-6 Astra OpenAI · ChatGPT
Acme Coffee is a Spanish specialty coffee roaster that sells single-origin beans online by subscription. Besides its web shop it runs two cafés in Madrid.
One error: the brand sells online only and has no cafés.
01 · Who you areConfuses it
Open model · EU-served One of thirteen
Acme is best known as the fictional Acme Corporation from the Looney Tunes cartoons. I am not aware of a coffee company by that name.
Takes the name for a cartoon company. Three open models do.
Correction worth making at the source
Say on the About page, in one sentence a model can quote, that Acme Coffee sells online only and has no physical cafés — and use the name in full, never «Acme» alone, so the pages about the roaster stop competing with the cartoon.
Built by Sealmetrics
The report is free. It comes from the team behind Sealmetrics, the cookieless analytics these companies use to measure what their marketing brings in.
We ran it on ourselves first, with fifteen models
Five of fifteen models
knew who we were.
Sealmetrics is a company with customers, a decade of history and a name that ten of those models attached to somebody else — industrial seals, water metering, call tracking. Not one of the eleven that answered a purchase question recommended us. We publish the numbers because the report is only worth something if it is allowed to say this.
described the company correctly or with one error
recommended it when asked for a tool like ours, without naming anyone
factual errors serious enough to correct at the source
What the six questions ask
The answers a buyer gets
before they reach you.
Each question maps to a moment in a real conversation. The report keeps every answer whole, so you read what the model said rather than a score standing in for it.
Who you are
What the model says your company does, in its own words.
What it praises and criticises
The advantages and objections it repeats without being prompted.
Who it recommends
Ask for a tool like yours without naming anyone. See who it names.
How it reads your sector
The trends and shifts it believes are happening in your category.
Your alternatives
Who it puts in your place when someone asks for another option.
You against a competitor
How it frames the difference when the two names are side by side.
What it does not measure
Stated plainly,
so the number means something.
- Not what ChatGPT, Claude or Gemini answer with search on. That is a live search result, not model memory, and it moves for different reasons. The one search answer in the report is Perplexity's, and it is labelled as one.
- Not a visibility score. Every figure in the report carries its denominator — five of fifteen, zero of eleven — because a score with no denominator is a number nobody can check.
- Not a ranking against competitors. It counts how often each name appears in the answers you get, in one run, and says so.
- Not a person. The report covers companies, products and brands.
What you do with it
The errors are the part
you can actually fix.
A model that confuses your company with another one is reading the sources that exist. The report ends with the corrections worth making at the source, in the order that changes the next edition. How Sealmetrics measures what search sends you
Questions before you ask for one
Frequently asked questions
Is this the same as tracking what ChatGPT says with search enabled?
No, and the difference matters. Eighteen of the nineteen models answer from training, with web search switched off. It is the layer underneath: when a model has no source at hand, or when a tool calls the API without search, this is the answer it falls back on. What a user sees in ChatGPT with browsing is a different measurement. The one exception is Perplexity Sonar, which always searches before it answers: it is in the report because that is what people who ask Perplexity see, and its answer arrives with the sources it cited.
Does the report include ChatGPT, Claude, Gemini and Perplexity?
Yes. GPT-6 Sol and GPT-6 Astra from OpenAI, the company behind ChatGPT, Claude Opus 5.5 and Claude Sonnet 5 from Anthropic, and Gemini 3.8 Flash from Google are called through the API with web search switched off, so the report shows what each one holds in memory, not what its app would find by searching that day. Sonar, from Perplexity, is the exception: it searches the web before it answers, so its answer is what Perplexity finds that day, with its sources listed underneath.
Which models answer the questions?
Nineteen: thirteen open models served from European infrastructure, plus GPT-6 Sol, GPT-6 Astra, Claude Sonnet 5, Claude Opus 5.5, Gemini 3.8 Flash and Perplexity Sonar. Each one gets the same six questions, one run each, at the same output budget. The report names every model beside its answer.
Who decides whether a model got my company right?
A model does, and the report says so on the page. It reads the six answers for each model and assigns one of four states: correct, partial, confuses it with someone else, or no usable answer. Nothing in the report is a human verdict, and none of it is presented as one.
What does the report cost?
Nothing. It runs on Enroutia, the model routing platform behind it, and the inference is paid for by Sealmetrics. There is no card, no trial and no account to open.
What happens to the email address I give you?
It is used to send you the report, and it is kept as a record of the request for up to 24 months. It is added to a mailing list only if you tick the box for occasional emails, and you can unsubscribe from any of them in one click. Neither Enroutia, which generates the report, nor the AI models ever see it: they receive the brand, the sector and the competitors, nothing else. To access or erase what we hold, write to privacy@sealmetrics.com.
Can I ask about a company that is not mine?
Yes for a company, a product or a competitor. No for a person: the report is about organisations, and asking nineteen models what they believe about a named individual is a different problem with a different answer.
How often does the answer change?
Every time a model ships a new version, and slowly in between. A brand that publishes a citable company page and gets its facts into the sources models train on can look different in the next edition. That is the point of measuring it.
It takes five minutes
and one email.
You will get the report as a page you can open, keep and forward. If it says nothing is wrong, that is also an answer worth having in writing.