Somewhere in the next few years, a doctor in India may glance at a phone or a laptop at the start of your appointment and say something like, “I’m using a tool to take notes.” Or they may not say anything at all. The tool is an AI scribe. It listens to the conversation between you and the doctor, then writes a draft of the medical note.
The technology is real, it is spreading, and the honest answer to “is it good or bad?” is that it depends on a few things patients rarely get told. Here is what the evidence shows about whether it helps, how accurate it is, what happens to the recording, and what you can ask.
What an AI scribe actually does
An ambient AI scribe records the visit, converts speech to text, and uses a language model to turn that text into a structured note: your complaint, the examination, the plan. The doctor is supposed to read the draft, correct it, and sign it. The pitch is simple. Doctors spend a lot of time typing notes, and a tool that drafts them frees that time for the patient in front of them.
Does it help doctors?
A little, according to the best trial so far. A randomised trial in NEJM AI in 2025 (Lukac and colleagues) assigned 238 outpatient physicians across 14 specialties to one of two commercial AI scribes or to usual care, for about two months. The scribes were used in roughly a third of visits (33.5% for one product, 29.5% for the other).
The main result was modest and uneven. Time spent writing notes fell by 9.5% with one product (Nabla). With the other (DAX Copilot) there was no significant change. Both products showed potential improvements in burnout, task load and work exhaustion, which the authors say larger trials need to confirm. Clinicians also reported occasional inaccuracies that needed ongoing oversight.
So this is not a magic fix. It is a small saving with a caveat attached.
How accurate are the notes?
A 2026 quality study in JMIR Medical Informatics (Taylor and colleagues) looked at 31 physicians who used an ambient scribe to help create 7,545 clinic notes. The physicians evaluated 356 of them. The result was mostly reassuring: 94.7% were free from significant errors.
The errors that did turn up fall into a few types. Accidental omissions, where something said in the room is missing from the note, were the most common at 18% of the reviewed notes. Hallucinations, meaning something in the note that was never said, appeared in 11.5%. Accidental inclusions were 9.3%, and bias was rare at 1.1%. Most errors were mild to moderate, but 19 of the 356 notes (5.3%) contained an error rated as a serious or imminent risk if nobody caught it.
What catches the errors is the doctor reading the note. In this study, physicians changed a median of 9% of the AI’s words, and 14.9% of notes (143 of 960 in the editing data) were left entirely unedited. Some of those may simply have been fine. But it shows how much safety depends on a busy person actually checking.
India is a harder case
A September 2026 preprint (Jaiswal and colleagues, submitted for review and not yet peer reviewed) argues that these tools are mostly built and validated on Global North speech, languages and consultation styles, and that Indian consultations are different. The authors describe them as brief, three-party conversations that are multilingual, mixed between languages, and held in resource-constrained, noisy settings, which they say raises the chance of transcription and note errors “manyfold.”
The same paper reports that it found no large, real-world public benchmark for testing these tools in India, that existing datasets are overwhelmingly synthetic, and that the deploying organisations they interviewed each use their own private evaluation methods, so buyers have no independent way to compare. That is a preprint’s claim and should be read that way. But it points at a real gap: a scribe that works well in one hospital’s English-language clinic has not necessarily been tested in yours.
Consent and where the recording goes
This is the part patients notice least. A scribe has to send the audio somewhere to be turned into text. In an August 2026 opinion piece in New Atlas about Australian clinics, the author described a doctor who said “nothing was recorded” and would not carry on the appointment without the scribe. The piece quotes Tom Sulston of Digital Rights Watch: the audio does have to be transmitted to the AI service, and most services delete it after transcription.
Deletion sounds good for privacy, but Sulston points out the other side. If the recording is destroyed, nobody can later check the transcript against what was actually said, and AI systems can mishear words, leave out details, and invent information. That is an opinion piece about Australia, not an Indian finding, but the mechanism is the same everywhere.
In India, the main framework is the Digital Personal Data Protection Act, 2023 and the DPDP Rules notified in November 2025, which are being phased in over roughly 18 months. According to legal summaries of the Rules, consent is the default basis for processing personal data, health data included. A notice has to be clear, in plain language, and say what data is collected and why, and giving a notice is a separate step from obtaining consent. The Rules also require reasonable security safeguards, such as encryption and access controls. Medimadad did not find a rule written specifically for AI scribes, so the general framework is what applies.
What you can ask
You do not need to argue with anyone. Four plain questions cover most of it:
- Is this visit being recorded, and is it being transcribed by an AI tool?
- Where does the audio or text go, and is it kept? For how long?
- Can I say no, and will that change how I am treated?
- Will you read the note before it goes in my file?
If you get a visit summary or printed note afterwards, read it. Check the things where a small error matters most: medicine names and doses, allergies, which side of the body, and how long you said the symptom had been going on. If something is wrong, say so and ask for it to be corrected. A clinic that cannot tell you where the audio goes is also giving you an answer.
This article is educational and informational. It is not legal or medical advice, and it does not recommend for or against any particular tool or clinic. If you have a concern about how a hospital handles your data, you can raise it with the hospital first.
Further Reading on Medimadad
If you are curious about the other ways AI is entering health care, our pieces on using AI chatbots for health questions and on AI reading medical scans cover two different corners of the same shift.
Sources
- Lukac PJ, et al. Ambient AI Scribes in Clinical Practice: A Randomized Trial. NEJM AI, 2025. DOI 10.1056/aioa2501000.
- Taylor SL, et al. Quality of Clinical Notes Created by Ambient Listening Generative AI: Pragmatic Prospective Pilot Study. JMIR Medical Informatics, 2026. DOI 10.2196/86474.
- Jaiswal SD, et al. Evaluating Ambient Clinical Scribes in India: The Need for Multilingual Real-World Clinical Conversation Data. Preprint, submitted September 2026 (not peer reviewed). arXiv 2609.17355.
- Armitage H. AI medical scribe consent raises patient rights concerns. New Atlas, August 4, 2026 (opinion). Link.
- Digital Personal Data Protection Act, 2023, and Digital Personal Data Protection Rules, 2025, as summarised by legal and compliance commentators (for example Seclore).
To get informed about health issues, visit medimadad.com.
