Most Accurate AI Scribe for Medical Specialists – Complex Cases & Medical Terminology

핵심 요약

• Hanah is the most accurate AI scribe for specialist clinicians available on the market, with the most complete notes and least errors recorded
• On a heart failure review, Hanah's notes captured 94.3% of the required clinical facts. Heidi's captured 74.3% and i-scribe's 67.9%.
• Hanah recorded the corrected medication, bisoprolol, in all five heart-failure notes. Heidi recorded metoprolol in four and omitted the drug in one.
• On a chest pain consultation, Hanah's notes captured every required fact in all five runs. Heidi averaged 76.7% and i-scribe 80.3%.
• Hanah captured more clinical detail in the cardiology comparison while preserving the clinician's stated reasoning. In the separate added-interpretation test, none of Hanah's tested notes introduced unstated reasoning.

Hanah is an AI medical scribe that turns a recorded consultation or post-consultation dictation into specialist notes, referrer letters and patient plans.

We tested Hanah, Heidi and i-scribe on cardiology notes and on ICU and trauma transcription. The complete test records include the notes, reference transcripts, fact inventories and AI grades. The testing method covers note accuracy and transcription scoring.

Specialist notes, dictation and letters with Hanah

See public pricing and trial terms and the Cliniko API and optional browser-extension guide. The API connection and browser extension can be used separately or together. For another worked documentation format, see our SOAP note template and AI drafting guide.

Cardiology note accuracy

The cardiology comparison measured how much of each consultation's clinical detail appeared in the finished notes, using a reference inventory of 33 facts for chest pain and 53 for heart failure.

Hanah captured more clinical facts across both cardiology consultations and delivered more consistent coverage across repeated runs.

Only Hanah correctly captured bisoprolol in all five runs. Heidi incorrectly recorded metoprolol in four runs and omitted the medication in one, while i-scribe also recorded metoprolol.

Download the cardiology notes, transcripts, checklists and grades, or read the scoring instructions.

Worked example: a medication correction and an uncertain diagnosis

In the heart-failure benchmark recording, the patient names bisoprolol. The clinician asks whether it was metoprolol, and the patient repeats bisoprolol. Later, the clinician discusses a possible flare-up of heart failure.

Hanah's first heart-failure note included these entries (full note):

Medications: furosemide, aspirin, bisoprolol and simvastatin. Furosemide strength unknown.

Suspected flare-up of heart failure. Explained that the heart may not be pumping as well as it should.

Review in 1 week.

The same clinical details in referrer-letter format:

A flare-up of heart failure is suspected. The current medication list includes furosemide, aspirin, bisoprolol and simvastatin; the furosemide strength remains unconfirmed. Review is planned in one week.

Added interpretations

Added interpretation occurs when a scribe links findings to an assessment, cause or plan without the specialist making that connection. In the test notes:

• The cardiologist said "the symptoms that you've described to me indicate the possibility of a heart attack". Several scribes wrote that the patient's smoking, blood pressure and family history supported the diagnosis. The cardiologist never said that.
• A pain specialist listed lower back pain, a disc herniation and foraminal stenosis side by side. Two scribes wrote the back pain was "secondary to" the herniation, a link the specialist never made.

The added-interpretation comparison covered eight vendors across four consultations.

The test evidence includes the notes and grader explanations; the exact grading instructions define what this pass checked.

Transcript accuracy

We measured transcript accuracy and final-note accuracy separately. Word error rate measures differences from a verbatim reference transcript; lower scores mean more accurate transcription.

Hanah led transcript accuracy across most recordings, with particularly strong performance on ICU dictation and trauma audio. Its ICU transcription remained stable when background noise was added. Inspect the ICU reference, transcript and error counts, the noisy ICU evidence, and the trauma evidence. The transcript source download collects available Hanah and Heidi WER records and the i-scribe capture transcripts for these six scenarios, including the exploratory ICU captures. The cardiology download above includes the captured transcripts supplied for note grading. See the transcription scoring method for the WER definition and normalisation rules.

How we tested

We replayed the recordings into each scribe through a virtual microphone for consistent audio input. Transcript scores compare each captured transcript against the same reference for that recording.

Read or download the source material:

• Cardiology notes, reference and captured transcripts, inventories, verdicts and interpretation-test notes.
• Scoring formula, configuration details and exact AI grading instructions.
• ICU and trauma evidence browser
• Full testing methodology breakdown.

Testing a scribe in your practice

Use a consultation with a spoken medication correction and an uncertain diagnosis to generate your usual note and referrer letter. Try post-consultation dictation in your own template and run the documents through your practice's usual filing workflow.

자주 묻는 질문

Which AI scribe is most accurate for medical specialists?

Hanah led this cardiology comparison, capturing more required clinical facts across both consultations and showing more consistent coverage across repeated runs. It also preserved the spoken medication correction in every heart-failure note. The full notes and scoring are available in the linked test records.

Can an AI scribe record the wrong medication?

Yes. In our heart failure recording the patient takes bisoprolol, and the doctor briefly asks about metoprolol. Heidi recorded metoprolol in four of five notes and i-scribe recorded it in its note. Hanah recorded bisoprolol in all five.

What is added interpretation in an AI scribe note?

Added interpretation is reasoning introduced by the scribe, such as linking risk factors to a diagnosis when the specialist made no such link. Our test checked whether notes preserved the clinician's reasoning. Hanah, Heidi and i-scribe had no flagged added interpretations; other tested vendors had between two and eight flagged notes.

Does word error rate matter for a specialist's notes?

Transcript accuracy supports faithful capture of medication names, doses and diagnoses. Hanah led across most tested recordings and maintained its ICU transcription performance with added background noise.

Can I dictate a specialist note after the consultation?

Yes. Hanah accepts a post-consultation dictated summary as well as a recorded consultation. Choose your note template, dictate the history, findings, assessment and plan, then review the draft and generate any correspondence you need.

Can Hanah draft letters to the referring GP?

Yes. Hanah drafts referrer letters from a consultation or selected sessions, using your letter format. You can also produce a patient plan from the same source material.

How much does Hanah cost, and do I need a practice management system?

Practice costs A$100 per practitioner/month, or A$80/month billed annually, with unlimited sessions, supported integrations and a one-month trial without a card. Hanah works standalone. Cliniko offers API appointments and note filing; the optional browser extension supports recording and drafting on supported practice-system pages. See the public pricing and integration guides linked above.

Additional output observations

i-scribe's heart failure note recorded that the patient denied dizziness, when dizziness only came up in the doctor's safety netting advice. It called the heart failure flare-up confirmed when the doctor had only wondered whether it was one. In one of Heidi's chest pain notes, the doctor's name was recorded as "Dr Jane Doe", a name that appears nowhere in the consultation.

i-scribe drafts a Medical Note, a referrer letter and a patient letter when you stop recording. On the ICU dictation, the Medical Note came back as "no transcript was provided" in every run, while the two letters used the transcript correctly. In the interface tested on 23 and 24 October 2026, we used the change-output workflow to request a different note format, and a SOAP note took about 27 seconds from stopping the recording.

On one run, i-scribe returned to its start screen partway through a recording and the transcript was lost. The same recording worked on a second attempt.

Reviewing a draft before filing

Do I still need to review notes from an AI scribe?

Yes. Make review the final step before filing: check the medication list, confirm that spoken corrections and the clinician's level of diagnostic certainty are preserved, and verify the agreed follow-up in both the note and any correspondence.