Does your clinical AI
earn the right to be trusted?

How faithfully does the AI capture what was actually said? We measure hallucination rate — content generated that was never spoken — and omission rate — clinically relevant content that was spoken but not captured. Tested across 100 standardised clinical scenarios validated by practising clinicians.

In clinical AI, an error is not a bug report. It is a patient safety event. We score drug name accuracy against a standardised medication corpus, clinical risk flag detection, and whether the platform has governance controls that catch errors before they enter the permanent record.

Where does patient data go? For how long? Under whose jurisdiction? We score data hosting location, security certification level, government compliance frameworks achieved, and whether the vendor can demonstrate that patient data never leaves the country of care. This dimension is a pass or fail for government health systems.

Adoption rate tells you whether clinicians actually use the tool. Edit rate tells you how much they trust what it produces. Satisfaction score tells you whether it helps or adds a new burden. We measure all three because a clinical AI platform that clinicians quietly stop using is not a solution. It is an expensive pilot.

Patients have a right to know they are being recorded. We score whether the platform requires explicit documented patient consent before recording begins, whether patients are notified clearly and in plain language, and whether consent records are retained and auditable. In 2026, lawsuits have already been filed against ambient AI companies for recording without adequate consent. This dimension is not optional.

The edit rate is the most honest measure of real world AI accuracy that exists. It tells you what clinicians actually do with AI output after the demo is over. A low edit rate means clinicians trust what the AI produces. A high edit rate means they are essentially rewriting notes — and the AI is adding work, not removing it. We score vendors on their independently verified edit rate from live clinical deployments.

When a clinical consultation ends, what happens to the recording? What happens to the transcript? What happens to the draft note? Most vendors cannot answer these questions precisely. We score the anonymisation standard applied to stored data, the verified retention period, the data destruction protocol and proof of destruction, the right to erasure process under applicable privacy law, and whether a complete audit trail of every deletion event exists and is independently verifiable. In a world where patient conversations contain some of the most sensitive data that exists, the question of what happens after the note is signed matters as much as the note itself.