AI Chat for "Pet Mein Dard" Messages: Handling Vague Symptoms Safely

At 11.20 pm a WhatsApp arrives: "Hello, pet mein bahut dard hai, kya karun?" No name, no age, no detail. At 11.25 another: "Doctor sahab kal appointment milega? Gas bahut ho rahi hai." At 11.40 a third: "Papa ko khoon ki ulti hui hai."

Three messages. Three completely different situations. If an AI assistant answers all three the same way, with "Thank you for contacting us, please share your preferred time slot", you have a serious problem. Gastroenterology is one of the specialities where a vague message can hide an emergency, so the AI has to be built safety first.

What should the AI do with a vague gut symptom message?

Its job is narrow, and that is a good thing. I put it simply in Rich Doctors Walk Away: the AI books the appointment, the doctor gives the advice. In a gastro clinic, the AI does four things:

1. Checks for warning words first, before anything else, on every single message.
2. Collects a few details if there are none: name, age, how long the problem has been there, and whether they have seen a doctor or have reports.
3. Offers real slots from your calendar and books the visit, with the location and what to bring.
4. Hands over to a human whenever the conversation needs one, with everything so far attached.

It never says what the problem might be. It never suggests a medicine, a home remedy or a diet. It always introduces itself as the clinic's assistant.

Which warning words must go straight to a human or emergency?

Your doctors must write this list. Not your agency, not your software vendor, and not me. Below is the kind of thing a list covers, to show you the shape. Your doctors decide the final words and the actions.

The rule is simple. When in doubt, the AI hands over. A false alarm costs your team two minutes. A missed warning costs far more.

Rule: The AI checks for warning words before it does anything else, on every message, including the tenth message in a long chat.

How do you write the rules so the AI follows them?

Write them in plain "if this, then that" sentences. The clearer the rule, the more reliably the assistant follows it.

1. If a message contains any word on the warning list, in any language or spelling, stop booking, send the emergency message your doctors approved, and alert the on-duty staff member by call and WhatsApp.
2. If the patient asks what is wrong with them, say the doctor will assess that at the visit, and offer the earliest slot.
3. If the patient asks about a medicine or a remedy, say you cannot advise on medicines and offer a slot or a call from the team.
4. If the patient sends a report, say the team will look at it and the doctor will discuss it at the visit. Do not read it out or interpret it.
5. If the patient is upset, confused or asks for a person, hand over immediately.
6. If outside working hours, tell them when a person will call, and give the emergency instruction your doctors approved for urgent symptoms.
7. Always introduce yourself as the clinic's assistant, not as the doctor.

Spellings matter. Patients write "khoon", "khun", "blood", "bld". Test every version.

How do you test it before patients see it?

This is where most clinics cut corners. Do not.

• Have every team member send 20 messages as different patients: the vague one, the scared one, the one who writes half in English, the one who mentions blood in the fifth message of an otherwise routine chat.
• Have one doctor send the trickiest ones. Doctors know how patients describe symptoms in ways nobody expects.
• Write down every wrong, vague or unsafe answer and fix the rules.
• Retest after every change to the knowledge or the rules.

Then go live in phases: the instant reply and booking first, everything else later.

What should happen after go-live?

The AI is not "set and forget". Every week:

1. A team member reads a sample of conversations, including every one that triggered a warning word.
2. Any message that should have triggered a handover but did not goes on the list for a rule fix the same day.
3. The doctors review the warning list every quarter and add any new phrases patients used.
4. Only patients who have agreed to hear from you get follow-up messages, and patient data stays private.

Example: Say the clinic receives 300 WhatsApp enquiries a month, a third of them after hours. If the AI books even a quarter of the after-hours ones that used to wait till morning, that is 25 more visits a month at an example ₹800 consultation, ₹20,000, while every urgent message still reaches a human within minutes.

The step-by-step build, from mapping the journey on paper to going live in phases, is in the tools chapter of Rich Doctors Walk Away. When we build assistants for clinics through our CRM, WhatsApp and AI work, the warning-word list is always the first thing we write with the doctors. Our gastroenterology page shows how enquiries flow through a GI practice end to end.

The short version

• In gastroenterology, a vague message can hide an emergency, so the AI must be built safety first.
• The AI checks for warning words on every message, collects details, books visits and hands over to humans.
• Your doctors write and own the warning-word list, in every language and spelling patients use.
• Write rules as plain "if this, then that" sentences, test them hard, and go live in phases.
• Review conversations weekly and the warning list quarterly. When in doubt, the AI hands over.