AI Visibility Auditor™
What AI assistants say about you, measured rather than guessed.
When somebody asks an assistant what your organisation does, whether you serve their region, or who they should use — a model answers. It answers from what it has absorbed, which may be outdated, partial, or about a competitor. This measures that, and keeps each measurement as evidence you can show somebody else.
What it actually does
You give it the questions your buyers really ask, in their words. It puts those questions to AI systems, records what comes back — every answer, which model gave it, and when — and reads your own site so the answers can be checked against what you actually publish. Where a model says something your own pages contradict, that gap is the finding.
This is currently invisible to you. Analytics show who arrived. They cannot show the conversations where you were described wrongly, described as somebody else, or not mentioned at all.
It does not stop at measuring
Measuring is the part people expect. The rest is what makes it worth acting on. It crawls your site and renders your pages the way a machine reads them, so what a model said can be set against what you actually publish. It keeps sealed snapshots of your figures, so a change can be given a date. And where a finding has a fix, it drafts that change out of your own published words — marking anything it cannot support from them as unsupported rather than writing it anyway.
A finding someone else can check
Sealing is built and tested, and a customer's runs can be sealed on request. Each measurement is hashed as it is taken, and a batch of them is time-stamped by an outside authority, with every measurement in the batch provable against that batch's root. The time stamp comes from the authority and not from us: a time we could set ourselves would prove nothing to anybody. The records are then held on infrastructure in New Zealand or the EU, and no US-owned cloud holds them. Months later you can show a board, a regulator or a customer what was true on a given date — and they can check it without taking our word or yours.
This is not a diagram. The path has been run end to end against a real authority, Asseco Data Systems (Certum) in Poland, and the receipt is on file. A qualified EU authority, Aruba PEC (Italy), is contracted, built and tested, and is switched on per customer as an opt-in; each qualified stamp is priced. Every record shows which authority stamped it and its grade.
The word is tamper-evident, never immutable. Nothing anybody builds prevents a record being altered. What the seal does is make alteration detectable by someone who does not trust us. A vendor claiming immutability is claiming something no software can deliver about data it holds.
What it does not do
- It does not promise a result. The work is to make you findable for the terms you choose — the words and combinations your buyers actually use — and every effort goes into that. What nobody can promise is the outcome, because nobody controls what a model says. So the result is measured rather than asserted: what was true before, what was changed, what is true after, each sealed so none of it can be quietly restated.
- It does not prove that a change you made caused a change in an answer. Models change underneath any such comparison, and a before-and-after that ignored that would be a story rather than a measurement.
- It samples the providers' APIs, not the consumer chat products. What a logged-in person with their own history sees is not what we measure, and that is a real difference. One row is not like the others and is named rather than buried: Perplexity searches the web while it answers, so it measures retrieval and not recall alone.
- Not every measurement is timestamped, and the record says which are. A qualified timestamp costs real money, so it is spent on the batches you intend to show somebody. An unsealed measurement is still taken, still hashed and still kept — and is shown as unsealed rather than quietly presented as though it were not.
- The measurement path is not sovereign, and cannot be. Finding out what a US model says about you means asking a US model: your questions and its answers travel to them. What is held in a jurisdiction you chose is the record.
- The records are sealed under our key, not yours. We act as an independent custodian — which is the point. A seal you could generate yourself would prove nothing to a third party.
If you are searching for this by another name
This field goes by several names: AI search, AI search optimisation, answer engine optimisation (AEO), generative engine optimisation (GEO), and LLM or AI visibility. They describe the same work — making sure AI systems can read your site, and working on what they say about you, for the terms and phrases your buyers actually use. That is the objective here too. The one thing set out above still holds: the outcome is measured rather than promised, because nobody controls a model's answer.
The assistants sampled are ChatGPT and Claude through their documented APIs, Perplexity, and Mistral in France. You can also point it at a model running on your own hardware, so that those questions never leave your building. Which of those you sample is your choice, and it is the choice that decides whose jurisdiction your questions travel through.
How to get it
Available on request. There is no signup form, no trial and no free tier: an invitation goes to one named person at one address, and it is the start of a commercial relationship rather than a funnel. It does not require a Village — it is priced and sold on its own.
An invitation names you. The conditions of access are on the page in full rather than behind a link, and what is recorded is the version together with a digest of the exact words you were shown — so the document cannot be edited afterwards and have you counted as agreeing to the new one. Then you enrol a phone authenticator. There is no password on the account: a code from your own device is what signs you in, and we hold nothing you type that could be reused.