Let me start with something some doctors won't: I know you ask AI about your child's symptoms. Everyone does now - at midnight, in the bathroom, while the child sleeps in the next room. I am not going to tell you to stop. I am going to tell you how to do it so it helps instead of terrifying you, because the difference between those two outcomes is almost entirely in how you ask.

Why the same question gets wild answers

Try it: type "child has fever and leg pain" into an AI tool. The answer will list everything from a viral illness to leukemia, in the same calm tone, with no way for you to tell which one is likely for your child. That is not the AI being bad. That is the question being empty.

Symptoms are like words - their meaning depends completely on the sentence around them. Fever plus leg pain in a 2-year-old for six hours after a vaccination is one story. The same five words in a 7-year-old for three weeks, with weight loss, is a completely different story. A doctor carries that context automatically from the questions we ask you. An AI only has the context you type. No context in, haphazard answer out - and a haphazard list that includes a terrifying possibility is how a manageable evening turns into a 2 a.m. spiral.

The five details to include every single time

This is the actual checklist. Before you hit send, make sure your question contains:

1. Age - exactly

"Baby" and "child" are useless to medicine. Three weeks, eight months, six years - the same symptom means different things at each. A fever at three weeks is an emergency; the same fever at six years is usually a virus. Age is the single biggest context switch.

2. The symptom, with its timeline

Not "has a cough" but "cough for four days, worse at night, started after two days of cold, today a little breathless while playing." When it started, how it has moved, what makes it better or worse. Trajectory is information.

3. Numbers where you have them

The temperature as measured, and how - "39.1°C with a digital thermometer in the armpit at 9 pm." Breathing rate if it is a breathing worry. How many wet nappies or toilet trips today. Vague words like "high" and "a lot" get vague answers back.

4. What your child is doing, not just what they have

Drinking or refusing? Playing a little or flat on the bed? Passing urine normally? Sleeping more than usual but waking normally? Doctors call this the child's "activity" and it outweighs almost any single symptom. An AI that knows your child drank two glasses of water and asked for cartoons will rank its answer completely differently.

5. The background only you know

Medical history (premature birth, asthma, a heart condition), medicines already given today with doses and times, recent vaccinations, sick contacts at school, travel, and what you have already tried. This is the part that stops the AI from suggesting the one thing that is actually wrong for your child.

One question, asked two ways

Empty: "child vomiting what to do"

With context: "4-year-old, vomiting since this afternoon, five episodes, no blood or green colour. Mild tummy ache, no fever, drinking small sips of ORS and keeping some down, passed urine twice today, sleepy between episodes but wakes fully and answers. Ate at a birthday party yesterday, two other children there are sick today. No medical problems, no medicines given yet. What should I watch for and when should I take her in?"

Same child, same evening. The second question will get you an answer you can actually use - calm, specific, and pointed at the right red flags.

Then do the two things that keep you safe

First, sanity-check the answer before it enters your bloodstream. A confident paragraph is not a correct paragraph - AI tools state wrong things in the same smooth voice as right ones. If an answer would change what you do - start a medicine, skip the doctor, stop worrying - verify it against a source like the AAP's healthychildren.org or your paediatrician before acting on it. (I wrote a separate piece on checking AI health answers.)

Second, respect the bright lines. There are three things AI should never decide for you: medicine doses, actual diagnoses, and emergencies. I explain why in Three things to never ask AI - the short version is that all three need the one thing no chat window has: your actual child, examined by an actual person.

What AI is genuinely good for

Used this way, it is a real tool. It is excellent at explaining a diagnosis you already have in plain language, listing the questions worth asking at tomorrow's appointment, turning a discharge summary into human sentences, and telling you which red flags belong to which symptom - so you know when to move fast and, just as importantly, when you can safely put the phone down and go back to sleep.

That last part is the point of this whole section of the site. Context in, calm out.