ChatGPT for Doctors: What It Is Useful For, and the One Thing You Must Not Do With It

The short answer

ChatGPT is a general-purpose writing and reasoning tool made by OpenAI. It is genuinely useful to a doctor for tasks that involve language and no patient identity: drafting a patient information leaflet, rewriting discharge advice in simpler Hindi, summarising a guideline you have pasted in, preparing a talk, or drafting a reply to an insurance query.

It is not clinical software. It cannot see your patient's history, does not know what is in your pharmacy, has no record of the allergy you documented eight months ago, and cannot write anything into your records.

And there is one thing you should not do with it: paste identifiable patient information into a consumer chat product. That is the part of this article that matters most.

MedKit Care is not affiliated with OpenAI and does not use ChatGPT in its product. This is an independent assessment.

What it is genuinely good at

Rewriting for patients

Turning your clinical language into something a patient with limited literacy can follow is a real task, and language models are good at it. "Take one tablet twice daily after food for five days, and return if the fever does not settle by day three" in simple Hindi is a legitimate use.

Drafting non-clinical documents

Consent form wording, clinic policy notes, staff instructions, a letter to a landlord, a reply to a corporate panel. All ordinary writing tasks.

Explaining something you already know to someone who does not

Preparing patient education material, a talk for a local association, content for your clinic's page.

Working through a guideline you supply

If you paste in a protocol document and ask it to summarise or restructure it, that is a comprehension task on text you provided, which is squarely within what these tools do well.

What it is not reliable for

Clinical decisions

It produces fluent, confident text whether or not it is correct, and it has no way to signal the difference. A wrong dose stated confidently reads exactly like a right one. There is no clinical validation behind a general chatbot, and none is claimed.

Drug dosing and interactions

It will answer. The answer will sometimes be wrong, and it will not tell you which time. For prescribing, use a source with a maintained pharmacological database behind it.

Anything requiring your actual data

It does not know your patient, your stock, your tariffs or your schedule. It cannot check whether the drug you are about to prescribe is on your shelf, because it has never seen your shelf.

Current information

Its knowledge has a cutoff and it does not know what changed after it. For anything time-sensitive, verify.

The thing you must not do

Do not paste identifiable patient information into a consumer chat product.

Name, phone number, address, hospital number, or a combination of details that identifies someone — that is personal data, and often sensitive personal data, being sent to a third-party service under consumer terms that were not written for healthcare. India's Digital Personal Data Protection framework treats you as responsible for the personal data you hold about your patients. Handing it to a general consumer service, without a purpose your patient consented to, is not a defensible position.

The practical rule is simple and easy to follow: if you would not read the sentence aloud in the waiting room, do not paste it into a chatbot. Describe the clinical situation without the identity — "a 54-year-old man with type 2 diabetes on metformin" is fine; the same sentence with his name and number is not.

Where clinical software is different

This is the actual distinction, and it is not about which model is smarter.

A clinical system is built to hold patient data lawfully, restricts who can see what by role, records every access in an audit trail, and — crucially — can act on your own data. When MedKit Care's AI drafts a prescription, it is working with the patient's recorded allergies attached from the record on the server, and its medicine suggestions are matched against your own pharmacy stock. It writes the confirmed result back into the patient's file.

A chatbot in a browser tab can do none of that, however capable the model behind it.

For completeness: MedKit Care's own AI features run on Google's Gemini models over the API with DeepSeek as a fallback. Not ChatGPT. That is a vendor choice, not a partnership, and MedKit Care is not affiliated with or endorsed by any of those companies.

A workable policy for your clinic

• Use general chatbots freely for writing that contains no patient identity.
• Never paste names, numbers, addresses or hospital IDs into one.
• Never take a dose or an interaction from one without verifying it in a proper source.
• Keep anything that must enter the patient record inside your clinical system.
• Tell your staff the same three rules. The risk in most clinics is a receptionist pasting a patient list to "tidy it up", not the doctor.

Related reading

• Gemini for doctors — the same assessment for Google's assistant
• AI software for doctors — what AI inside a clinical system does differently
• AI and patient data privacy in India — the rules in more detail
• AI healthcare software — what the category actually means