Article

Hanah found to be the best AI scribe in rigorous industry benchmarks

Key takeaways

  • The cited evaluation compared six AI scribes on the same 12 recordings, with note identities hidden during grading.
  • Hanah includes care plans through a clinic-branded patient app alongside an AI scribe.
  • In that evaluation, Hanah ranked first on 18 of 19 reported measures; the article also reports the measure it did not lead.
  • These results describe the cited test. Inspect the recordings, outputs and grading method, and review every clinical draft.

ScribeStandard tested Hanah against Heidi, Lyrebird, Freed, CliniScripts and PatientNotes on the same 12 recorded consultations and graded every note blind against the same list of clinical facts. Hanah ranked first on 18 of the 19 measures. Read the full write-up: Hanah is the best AI scribe.

The test

Every scribe heard the same audio. Each consultation has an inventory of every fact said about the patient's conditions, symptoms and signs, 358 facts across the 12 consultations. A note gets an error for each fact it missed or captured only partly, and for each claim the consultation does not support. Errors are counted the way Weiner et al. (2020) counted them in doctors' notes, and every scribe was run at least five times on every consultation.

The results

Scribe Errors per note Facts captured Hallucination flags per note Note ready to finalise
Hanah 4.1 93% 0.80 2m 23s
Freed 8.3 87% 2.47 4m 40s
Heidi 9.0 81% 1.00 3m 25s
CliniScripts 9.7 80% 1.40 4m 03s
Lyrebird 11.6 82% 3.42 4m 06s
PatientNotes 18.0 61% 2.58 5m 39s

Weiner et al. found 5.0 errors per note in doctors' notes checked against recordings of the visit. The ScribeStandard test counts more strictly than theirs, and Hanah was the only scribe below that figure.

Reviewing and fixing a Hanah note took 2 minutes 2 seconds. Writing the note yourself takes a median 8.1 minutes (Apathy et al. 2023), which makes a Hanah note 75% quicker to finish than writing it by hand. The next closest was Heidi at 63%.

Hanah also had the lowest word error rate at 12.8%, the fewest mis-heard words that change clinical meaning, a note a median 19 seconds after you press stop, and the least data per consultation at 9.1 MB.

The one measure Hanah did not lead

In a 50 minute stress test with every scribe recording at once on one machine, CliniScripts used less memory, 972 MB against Hanah's 1,103 MB.

Hanah also had the highest share of notes ready to finalise with no edits at all, 8%. Every note from every scribe still needs your review before it goes in the record, and Hanah's drafts wait for it.

Every grade is published

ScribeStandard publishes the recordings, the transcripts, every scribe's notes and the reason behind each grade. You can listen to a consultation, read what each scribe drafted and check the marking line by line. Read the full results and how the test works.

Then run your test. Try Hanah free for a month on your consults, no card needed.

Frequently asked questions

How many consultations were in this evaluation?

The evaluation cited in this article used 12 recorded consultations across six AI scribes. Other benchmark posts may describe different runs or sample sizes.

Did Hanah lead every reported measure?

No. The article reports first place on 18 of 19 measures and identifies memory use as the measure led by another product.

Can I inspect the evidence?

The article links to the results and methodology, including the recordings, outputs and reasons behind the grades. Read those alongside the headline results.

Sources cited in this article

Update note: Highlighted home exercise programs and the clinic-branded patient app in the key takeaways. Source-check and benchmark dates remain as stated in the article.

How we write, source and review articles

See Hanah in your own practice.

Book a demo