Definition
What is AI evidence assurance?
AI evidence assurance is a defined check of in-scope claims in AI answers, against the evidence competent to settle each one. This is AlphaCitation’s definition, and it is the work AlphaCitation does.
The definition
Evidence assurance is a defined check. For a claim in an AI answer, it identifies the record competent to settle it, in that market on that date, tests the claim against that record, and records the result and its limits. It covers the claims in scope, not every claim in an answer and not the answer's reasoning. It is not a certification, an audit opinion or a guarantee that any answer is correct, and it does not replace your own review, including medical, legal and regulatory review.
Why it is needed
People now put questions about brands and medicines to AI systems. An answer can be confident and still be wrong, out of date, or true only of another market. Counting how often a brand appears does not show whether what is said about it holds. Read one case of a wrong answer.
Three questions it asks of a claim
Which record can settle this claim?
Not every source can. The check starts from the authority competent to decide the claim.
Does that record apply in this market?
An approval in one jurisdiction does not settle a claim about another. At most it is a declared benchmark.
Did it apply on that date?
A record that was right last year may not be right today, and a newer one does not rewrite what was true before it.
Assurance and intelligence are two halves of one record
Evidence assurance
Is the claim supported? By which authority? In which market? As of which date?
Intelligence
What is AI saying? What is changing? How does the brand compare? What warrants action?
How it differs from what you may already have
AI brand monitoring
Reports how answers mention a brand. Some tools also check facts. AI evidence assurance tests a claim in an answer against the record competent to settle it, in that market on that date.
AI model evaluation
Measures a model against a test set, for its developer or an independent evaluator. AI evidence assurance reads what answers say about a given brand or medicine, in a given market.
Medical, legal and regulatory review
Approves your own materials. AI evidence assurance looks at what others' AI systems say, and does not replace that review.
Certification
Is a third party's formal attestation. AI evidence assurance is a defined check, not a certificate.
Questions
- Is AI evidence assurance a certification?
- No. It is a defined check with stated limits. It is not a certification, an audit opinion or a guarantee that any answer is correct.
- Does it check every claim in an answer?
- No. It covers the claims in scope. In Label Ledger those are claims about a medicine's approved uses and populations, decided in the US and in the EU and EEA; elsewhere the answers are shown beside both labels as benchmarks, with no verdict. A claim outside the scope stays unassessed, and is given no verdict.
- Does it assess the answer's reasoning?
- No. It tests whether a claim holds against the record. An answer can reach a correct claim by a poor route, and the check says nothing about the route.
- What happens where no authoritative record exists, or none can be read?
- It states what was read and what was not found. Absence of confirmation is not evidence of absence. Nothing is settled, and the claim is not marked right or wrong.
- Does it depend on AI being wrong?
- No. If AI answers improve, there is still a need to establish what they can substantiate, and to see what changes.
- Can it change what AI says?
- No. It measures what answers to a fixed set of questions, put to the engines through their APIs, say. An individual consumer session can differ. It does not control answers.
- Is it only for medicines?
- No. It applies where an authoritative public record can settle a claim: for medicines, the regulator's label; in other sectors, registers such as NHTSA safety ratings and USDA FoodData Central. The published standard covers medicines.
The standard and the limits are published
For medicines, the check rests on the Evidence-Authority Standard, which sets out which evidence has the authority to settle a claim about a medicine, for which market, as of which date (doi:10.5281/zenodo.23594909). The standard decides authority, not truth: reading the record and judging the claim against it is the check’s own work. The methodology says how each finding is read, and where the measurement stops.