Your buyers stopped searching. They started asking.
ChatGPT, Perplexity, Gemini and Claude now answer the questions your customers used to type into Google - and most brands have no idea what those answers say about them. Arbitr measures it, then fixes it.
If the model cannot cite you, it cannot recommend you.
Your name never surfaces
Alternatives are recommended and the question closes without you in the room. Nothing about it reaches you.
Present here, absent there
The same question in another market returns a different answer, because that is where the least approved evidence about you exists.
Mentioned, but not chosen
You appear in the list and nothing commits - no first recommendation, and no clear reason to pick you over the others beside you.
Mentioned, and wrong
An expired accreditation, an old price, a discontinued feature - stated with confidence, carried by a source still online, and costing you trust.
Measure it. Fix it at the source. Prove it moved.
Where you surface, assistant by assistant, on a benchmark that holds still.
Contradictions surfaced with both sources named, then corrected from the approved claim.
Measured against a baseline and a holdout set before the change shipped.
One score. Seven weighted signals.
See exactly what the models see.
Nothing changes but the answer
Set the assistants, the markets and the prompt set once. Re-runs use the same set and the same weights, so a movement in the score is attributable rather than noise.
Every observation, laid bare
The exact response to each benchmark prompt - where you placed, in which language, and whether the claim made about you matches what you have approved. Inspect it, correct it, or challenge it.
Your share of the answer
Not market share - share of recommendation inside the benchmark. Who wins the citation when a buying question comes up, and which domains are feeding the answer on their behalf.
Parity, market by market
Coverage, prominence and accuracy for each market you operate in, measured against your reference language. The gap is usually wider than expected, and it is the one most tools do not look at.
Your approved facts, and what they rest on
Claims, evidence and lineage in one layer - each carrying its state, its language and the source supporting it. Only claims that are approved and public are eligible to ground generated content.
Prove the fix worked, honestly
An intervention is tied to a baseline and a holdout before it ships, and results are labelled observed, associated, or statistically distinguishable. "Caused" is reserved for designs that can support the claim.
Nothing publishes without knowing where the line is.
Your approved messaging, product and pricing detail, approved claims and the evidence behind them, and the legal, privacy and regulatory constraints that apply - held in one layer inside Arbitr.
It is also where you set the guardrails: what Arbitr must never claim, what needs marketing, legal or executive sign-off before it goes near a live channel, and who is brought in when something is ambiguous. Generated content is checked against it before it moves.
And when the evidence is not there, the draft says so. It returns insufficient approved evidence rather than writing a plausible sentence to fill the gap.
Twenty-five years experience in getting this right.
