What we review
We cover AI scribes and documentation tools that clinicians can buy or that health systems deploy. Each tool gets a fact table built from the vendor's own pricing, security, trust center, legal, and help pages. News coverage is used only for company background (founding, funding, customers).
Eligibility gates (separate from scores)
Safety and compliance basics are gates, not points. A tool is only described as suitable for US patient data when the vendor publicly states HIPAA compliance and a BAA is available. Where a BAA is not publicly documented, we say so and recommend getting it in writing. Gates are never traded off against price or features.
How comparisons are scored
We do not publish numeric star ratings. Comparisons weigh fit by practice type across: published pricing and free access; EHR integration depth; specialty and psychiatry templates; data retention and deletion controls; model training policy; certifications; platforms; and support and contract terms. Verdicts state who each tool fits and who it does not.
Evidence labels
- Vendor-reported A claim the vendor makes on its own site (for example a certification or a no-training statement) that we have not independently audited.
- Documentation reviewed We read the vendor's official documentation (pricing page, terms, BAA, retention policy) and summarize what it says.
- Third-party sourced From a non-vendor source such as a reseller listing or a news report. Used only when no official source exists.
- Not verified The vendor does not publicly disclose it. Shown as "Not publicly disclosed (ask vendor)".
- Not yet independently tested Anything requiring hands-on testing, such as note accuracy.
We never leave a cell blank.
Research score and planned note-quality testing
The published scorecard gives each vendor a research score from 0 to 100, labeled "Research-based score, not clinical testing". It has five categories worth 0 to 20 each: pricing transparency and value; EHR integration depth; privacy and security; specialty and psychiatry fit; and workflow features. Each category has four sub-items worth 0 to 5, scored only from public vendor documentation. Anything not publicly documented scores 0 and is shown as "Not disclosed". A publicly documented BAA is an eligibility gate: vendors without one are shown as not ranked and must not be used with PHI until a BAA is signed. Scores are stored in data/scores.json with an "as of" date.
Planned note-quality protocol (not yet run): standardized synthetic encounters (no real patients), psychiatry first, recorded identically for every tool. Two clinician reviewers will independently rate each note for omissions, unsupported additions (hallucinations), contradictions, and editing time to a sign-ready note, and we will report inter-rater reliability. Each result will publish the product version and test date. Once testing exists, note quality will weight about 50% of the overall score, with the research score making up the rest. Until then, every vendor shows "Independent testing coming" and note_quality is null in the data.
Hands-on testing
No hands-on clinical testing of any tool has been done yet. We do not report accuracy scores. Any accuracy figures quoted are vendor-reported and labeled as such.
Update cadence
Pricing and terms are rechecked weekly, and every page shows its last-checked date. Changes are logged on the corrections page.
Limitations
- Vendors change prices and terms without notice; always confirm before buying.
- Public documentation can lag behind the product.
- Enterprise contract terms are often negotiated and not public.
- Nothing here is legal, compliance, or medical advice.
Who curates this
ClinicalAIWeekly is curated by a practicing psychiatrist, which is why psychiatry and behavioral health fit get extra attention. The curator stays anonymous; see the editorial policy for independence rules.