AI scribes for doctors in India: what the evidence says, and what to ask

Ambient AI scribes are the first AI tool most doctors meet. The promise is simple: the scribe listens to the consultation, drafts the note, and the doctor stops typing. For a doctor seeing sixty patients in an out-patient session, even a small saving per patient adds up.

The evidence is now good enough to say what scribes actually do, and what they do not. This article summarises it, explains what is different about Indian consultations, and lists the questions to ask before choosing one.

How an ambient scribe works

1. The patient is told the consult will be recorded, and agrees.
2. The conversation is recorded and transcribed into text.
3. A language model drafts the clinical note from the transcript: complaints, history, findings, assessment and plan.
4. The doctor reads the draft, corrects it, and signs it. Only then does it become part of the record.

The last step is the one that matters. A scribe produces a draft, not a record.

What the trials found

Time saved: real, but modest. The strongest evidence is a randomised trial at UCLA, published in NEJM AI in 2025. It compared two commercial scribes with usual care across 238 physicians in 14 specialties. One scribe cut time spent on notes by 9.5% compared with usual care; the other made no significant difference.

A larger observational study, published in JAMA in 2026, followed 8,581 clinicians at five academic centres, of whom 1,809 took up a scribe. Users spent about 16 fewer minutes on documentation and 13 fewer minutes in the EHR per eight scheduled patient hours, and saw about half a visit more each week. The authors described the effects as modest. Time spent on records after hours did not change significantly.

Burnout: probably better. In the UCLA trial, measures of burnout, task load and work exhaustion suggested improvement with both scribes. A separate stepped-wedge trial, also published in NEJM AI in 2025, found less work exhaustion without a rise in professional fulfilment, and shorter documentation time without a loss of note quality.

Errors: occasional, and they matter. Clinicians in the UCLA trial reported clinically significant inaccuracies “occasionally”, and the trial recorded one mild adverse event. Studies of AI-written clinical summaries give a sense of scale: one, published in npj Digital Medicine in 2025, found that 1.47% of sentences contained something that had not been said and 3.45% left something out.

The honest summary: a scribe saves some time, may ease burnout, and needs a doctor who reads what it wrote.

What is different about Indian consultations

Every one of those studies was done in the United States, and Indian practice differs in ways that could change the result in either direction.

• Language. Many Indian consults move between English and Hindi, Tamil, Telugu or another language, sometimes in the same sentence. Speech recognition for this kind of code-mixed clinical speech is much less studied than for English, and early research suggests it is where systems fail most.
• Volume. A busy OPD sees far more patients in far shorter consultations than a US clinic. A scribe saves seconds per patient here, but across sixty patients that may matter more.
• Who writes the note. In many hospitals a junior doctor or assistant writes the notes. A scribe may change their work more than the consultant’s.
• Drug names. Indian prescriptions are full of brand names and combinations. A scribe has to get these exactly right, because “Dolo 650” and a misheard alternative are not interchangeable.

None of this has been measured in a published trial in India yet, so treat any claimed time saving as a hypothesis to test in your own clinic.

Questions to ask before choosing a scribe

1. Does it draft, or does it write to the record? Nothing should reach the patient’s record until the doctor signs it.
2. How does it handle the languages you and your patients speak? Test it on real consults, not a demo script.
3. How does it handle drug names, doses and abbreviations? Read the drafts for exactly these.
4. What does the patient see and agree to? Consent should be asked and recorded every time.
5. Where does the audio go, and how long is it kept? Does the data stay in India? Is it used to train the vendor’s models?
6. Can you edit the draft easily, and regenerate it? The review step has to be fast or it will be skipped.
7. What happens to the transcript? Can you read it beside the note, to check what was actually said?
8. Can you measure it? Agree a baseline (minutes per note, notes finished the same day) before the trial starts, and measure again after a month.

Where 92health fits

92health includes an ambient scribe in its consultation workspace. The consult is recorded with the patient’s consent and transcribed, and a draft note is produced for the doctor to edit and sign. The transcript can be read alongside the draft, to check what was actually said. On eye consults the draft is a clinician’s note the doctor adds to the record by choice, never automatically. Doctors can also dictate into a note, or say a medicine and get one prescription line back.

Like every AI feature in 92health, the scribe drafts and a person decides. You can see the full list on our AI page.

Sources

• Lukac PJ et al. Ambient AI scribes in clinical practice: a randomized trial. NEJM AI, 2025;2(12). doi:10.1056/AIoa2501000
• Afshar M et al. NEJM AI, 2025;2(12). doi:10.1056/aioa2500945
• Rotenstein LS et al. JAMA, 2026;335(16):1408-1417.
• Asgari E et al. A framework to assess clinical safety and hallucination rates of LLMs for medical text summarisation. npj Digital Medicine, 2025;8:274. doi:10.1038/s41746-025-01670-7