Handling patient queries on WhatsApp without missing any
Let the bot handle timings, fees, directions and appointment booking, and route anything clinical to a person immediately. The critical design decision is deciding what the bot must refuse to answer, and a clinic's queue needs tighter reply deadlines than most other businesses.
- A bot must never interpret symptoms, results or medication.
- Clinical questions need a named owner and a short deadline.
- Patient data in chat is sensitive; set a retention period deliberately.
- Most clinic messages are logistics, not medicine.
What clinics actually receive
Read a week of a clinic's WhatsApp and the split is consistent. Around three-quarters is logistics: what time is the doctor in, how much does a consultation cost, where do I park, can I move my appointment, is the report ready.
The remaining quarter is clinical or near-clinical: photographs of a rash, a question about whether to continue a medicine, a report the patient wants explained, a symptom that has got worse.
Those two groups need completely different handling, and the failure in most clinics is that they sit in the same undifferentiated thread on a phone at the front desk. The logistics questions consume attention, and the clinical ones wait behind them.
Automating the first group is straightforward and safe. Handling the second group properly is a triage problem, not an automation one.
The split also varies by time of day in a way worth noticing. Logistics questions cluster before the clinic opens and clinical ones arrive during and after consultations, which means the queue is busiest with the wrong kind of message exactly when the front desk is checking patients in.
What the bot should answer, and what it must refuse
Draw the line explicitly in the system prompt rather than hoping.
| Bot answers | Always a person |
|---|---|
| Doctor timings and days | Any symptom description |
| Consultation fees | Photographs of a condition |
| Location and parking | Questions about medication |
| Booking and rescheduling | Interpreting a test report |
| Which documents to bring | Anything urgent or worsening |
Write the refusal wording yourself. "I can't advise on that, but I'm passing it to the clinic now and someone will reply shortly" is right. A vague deflection leaves the patient unsure whether anybody will respond, and some will simply wait.
Triage and deadlines
A clinic needs a shorter clock than most businesses, and it needs the urgent cases separated automatically rather than by whoever happens to read first.
With the Support Desk in Convo360, a message that reaches a person becomes a ticket with a priority, an owner and a reply deadline counted in clinic hours. Anything the bot flags as clinical should open at high priority, with escalation to the practice manager if nobody has picked it up.
Fifteen to thirty minutes is a reasonable first-response target for clinical questions during clinic hours. Logistics can sit at two hours without anybody minding.
The point is not speed for its own sake. It is that a worsening symptom described at 11 am should not be discovered at 4 pm because the front desk was busy.
Name the owner explicitly for clinical tickets rather than leaving them in a shared pool. In a clinic, the person who can answer is usually one specific doctor or nurse, and a pool everyone can see is a pool nobody picks from.
Patient data and what you are obliged to do
WhatsApp threads at a clinic accumulate exactly the data you should be most careful with: names, conditions, photographs, reports, sometimes identity documents.
Under India's DPDP Act this is personal data you hold as a fiduciary, and health information deserves stricter handling than most. Three practical obligations follow.
- Limit who can read it. Front desk staff need appointment threads; they do not all need clinical ones.
- Set a retention period. Transcripts should not accumulate indefinitely by default.
- Say what you do in your privacy notice, including that messages are processed by a platform and, if you use AI, by a model provider.
None of this is onerous, and all of it is much easier to decide now than after a patient asks.
Reminders reduce the queue more than automation does
A large share of clinic messages are caused by uncertainty the clinic could have removed.
An appointment reminder the day before, with the time, the doctor's name and what to bring, prevents a whole category of questions. A message when a report is ready prevents another. A clear fee message at booking prevents a third.
These go out as utility templates and cost very little. Most clinics find inbound volume falls noticeably within a fortnight, and the messages that remain are more likely to be the ones that needed a person.
That is the right order of work: remove the avoidable questions first, automate the routine ones second, and triage what is left properly.
Preparation instructions deserve their own reminder. A patient told the evening before to fast for twelve hours, or to bring previous reports, is a patient whose appointment is not wasted, and each wasted appointment generates its own small pile of messages.
A safe starting configuration
For a single-doctor clinic starting from a shared phone, this is enough for the first month.
Put the timings, fees, location, preparation instructions and booking process into the knowledge base. Write a system prompt that names the clinic, sets a calm tone and lists explicitly what must be handed over. Set clinic hours honestly, including the evening session. Set a fifteen-minute first-response target for high priority and two hours for everything else.
Then read every handover for a week. You are checking one thing: did the bot correctly refuse everything clinical. If it answered something it should not have, tighten the prompt that day.
Once that is stable, add reminders and a report-ready notification. Those two do more for the patient experience than anything else on this list.
Example conversation
Industry guide: AI agents for Clinics & Hospitals — Appointment booking, reminders & 24/7 FAQs.
Triage patient messages safely
Start free in minutes — no code, no credit card. Bring your own AI keys and pay no per-message markup.
Frequently asked questions
Can the bot answer questions about medicines?
No, and the prompt should refuse explicitly. Even a correct answer about a common medicine sets an expectation that the bot handles clinical questions, and the next one may not be common.
What about patients sending photographs of a condition?
Route them to a person immediately at high priority, and store them with the same care as any medical record. The bot should acknowledge receipt and say a person will respond, without commenting on the image.
Is WhatsApp appropriate for health information at all?
It is where patients already are, and messages are encrypted in transit. What matters is what you do with the data afterwards: who can read it, how long you keep it, and whether your privacy notice says so.
How do we handle emergencies?
State clearly, in the bot's first message and in your profile, that WhatsApp is not for emergencies and give a number to call. Then make sure anything urgent still escalates immediately, because people will use it anyway.
Does this work for a multi-doctor practice?
Yes, and routing matters more there. Assign clinical tickets to the right doctor's queue rather than a shared pool, so a question about one patient's treatment reaches the person who knows the history.
Related reading
Available across India — see WhatsApp & AI calling automation by state and city.