Best AI for Paediatricians: Vaccine Recalls and School Forms

A paediatric practice can use AI for recalls by letting the record system produce the list of children due or overdue, then using AI only to write and test plain-English letter templates that are merged by the system. For school and camp forms, use AI to pre-fill factual fields and keep a clinician signing every form.

The split matters. The list comes from the record, so the data stays inside the system that already protects it. The AI works on wording, not on children's names. And nothing the AI drafts gives symptom advice or says whether a child should be seen. Those three rules keep the useful part of the tool and remove most of the risk.

Follow me on Instagram@sagnikteaches

Where the time goes in a paediatric admin week

Two jobs eat a surprising amount of reception and nurse time in a paediatric practice. Recalls are repetitive: someone exports a list, checks it, writes letters, chases replies. School and camp forms are irregular and urgent: they arrive in bursts before term, before a sports season and before a camp start date, each with a different layout asking for much the same facts.

Connect on LinkedInSagnik Bhattacharya

Say a four-clinician practice with 2,400 registered children. If 15% are due for a vaccine or check in a given quarter, that is 360 recall contacts. At four minutes each when done by hand (look up, write, send, log), that is 24 hours. A merged letter with a log entry written by the system takes under a minute of checking per child, so the work falls to roughly six hours plus the exceptions. Those figures are an illustration, so time your own first batch before you plan around them.

Subscribe on YouTube@sagnikteaches

Which AI does which part of the job

"Best AI" here is three tools doing three jobs, and the one that matters most is the one you already run. Splitting it this way also tells you where patient data may go.

If a vendor offers one product that claims to do all five, ask it to show each job on a test record, and ask where the data sits for each. The answer to "where does the recall list live?" should always be "in your record system".

The recall list comes from the record system, not from AI

Resist the temptation to export a spreadsheet of children and paste it into a chat tool. A vaccine history linked to a name and date of birth is health data about a child. Whether a given AI plan is acceptable for that depends on your agreement with the vendor, your data-protection obligations and your local rules, and a free consumer chat account will not meet them. Your practice's data-protection adviser can confirm what is allowed.

What to do instead:

1. Ask your record-system vendor how to build a recall query: children whose next dose or check-up date has passed, or falls in the next 30 days, excluding those already booked.
2. Export only the fields the letter needs: parent or guardian name, child's first name, the item due, the preferred contact route. Leave out the date of birth and diagnoses.
3. Run the merge inside the record system or in your approved messaging tool.
4. Log each contact in the child's record so the next query skips them.

If your system cannot do step one, that is the real problem to fix, and it is a conversation with the vendor, not a job for AI. A practice manager's wider prompt set is in ChatGPT prompts for healthcare practice managers.

Writing the letter templates with fictional children

AI is good at the wording job, and you can do it with no patient data at all. Give it a fictional child and ask for a template with merge fields. Here is a prompt you can copy.

Write a vaccine reminder letter template for a paediatric practice. Reading age 11. Under 140 words.
Merge fields to use exactly: [parent name], [child first name], [item due], [due by date], [booking phone], [booking link].
Rules: say what is due and by when; say how to book; do not describe symptoms, side effects or risks; do not say whether any child is unwell; do not give medical advice; include a line telling parents to call the practice if they think they have already had this done elsewhere.
Tone: warm, brief, no pressure or guilt.

A realistic output:

Dear [parent name],

[child first name] is due for [item due]. Please book an appointment by [due by date].

You can book online at [booking link] or call us on [booking phone]. Appointments for this are short, and you are welcome to bring a favourite toy or book.

If [child first name] has already had this at another clinic, please let us know so we can update our records.

Thank you,
The practice team

What you would fix: check "short" is true at your practice, remove the toy line if you do not want to make that promise, and confirm that "item due" will never merge as something sensitive in a letter that could be read by someone other than the parent. The last point is a real one: use a neutral phrase such as "a routine immunisation" if letters go to a shared household address.

Three letters, not one: first notice, reminder, final contact

A recall sequence works better when each message has a different job and, ideally, a different route.

The intervals are examples to adapt to your own policy and schedule. The point is that the third step is different in kind: it is where a person looks at the record, and it is where a worried parent may finally say why they have not booked. A practice that wants to see how another recall flow handles escalation can read how dental practices use AI for recalls and reminders, which has a different clinical setting but a similar structure.

Rules that keep clinical advice out of automated messages

Replies are where recalls go wrong. A parent answers a reminder with "She had a fever after the last one, should I still bring her?" and an eager auto-responder writes something reassuring. Write these rules down and apply them to every tool:

• Any reply that mentions a symptom, a reaction, a worry about safety or a question about whether to attend is routed to a human the same working day.
• Automated replies may confirm a booking, give opening hours and give a phone number. Nothing else.
• The template never includes a statement about risks or benefits of any vaccine. That conversation belongs to the clinician.
• If the parent says they do not wish to proceed, the system records it and stops the sequence; a clinician decides about any follow-up.

If parents call rather than reply, the same boundaries apply to a phone line. The trade-offs are in paediatric answering service: staff, AI or a nurse line.

Handling parent data without creating new risk

A child's record involves a second person, the parent or guardian, so check consent to contact and who may receive information. Four habits reduce risk:

1. Verify the contact before the first send. Separated parents, foster carers and changed numbers cause the worst errors. A message to the wrong adult about a child is a data breach in many places.
2. Use minimum content. The letter needs the child's first name and what is due. It does not need the date of birth, a diagnosis or a full vaccine history.
3. Keep AI out of the loop for real records. Test with fictional children. If a vendor's tool is to handle real data, put the agreement and data location in writing first.
4. Record opt-outs. A parent who asks to stop a route (text, say) should be switched to another with a logged note.

The questions to put to any vendor are listed in AI scribe security: audio, retention and agreement questions, and they apply to messaging tools as much as scribes.

The school and camp form queue

Forms come in a dozen layouts: school entry health forms, sports clearance, camp medical forms, nursery medication forms, travel letters. Most ask for the same few facts and one signature. A queue turns the pile into a process.

Give every request a row in a shared sheet or your task system, with these fields:

Four statuses are enough: received, ready for clinician, signed, sent. A form older than five working days in "received" gets a nudge to the owner. A request marked needed-by within 72 hours jumps the queue.

What AI does on a form, and what only a clinician does

AI earns its place on the boring parts. It can read a blank form a parent has scanned and turn it into a field checklist, draft a polite acknowledgement to the parent, and explain a confusing section in plain English for staff. It should not decide clinical content.

• Staff may let AI do: name, address and contact fields from the record (checked by a human), a list of what the form is asking, the parent acknowledgement, the cover note, the reminder to the parent to sign their part.
• Only a clinician does: allergies and medication statements, fitness for activity, conditions the camp should know about, any signature and date.

A form should leave the practice only after the clinician has read it, so make the status "signed" something they set themselves.

Five sample jobs, worked through

1. Turning an unfamiliar form into a checklist. Remove the child's details, paste the form text and ask: "List every question on this form that needs a clinician, every question staff can answer from the record, and every question the parent must answer." The output will be a three-column list you can attach to the task. What to fix: the AI sometimes misses a question hidden in a footnote, so scan the original too.

2. The parent acknowledgement. Template: "Thank you for sending the [form type] for [child first name]. We have it in our queue and aim to return it by [date]. If it is needed sooner, please call [phone]." It contains no clinical content and goes out automatically.

3. A sports clearance request. The practice decides whether it does these by appointment only. The auto-reply should say so and link to booking. A form for a child who has not been seen in two years cannot be completed from the record alone, and the reply should explain it without hinting at a clinical reason.

4. A mistake in practice. A pre-filled camp form carried the previous year's allergy list because the template pulled an old field. The clinician caught it because the process says every medical section is read against the live record. Without that read, the form would have gone out with out-of-date information. That is why the sign step is not optional.

5. A nursery medication form. It asks for dose and timing of a daily medicine. The AI should never fill this. Staff copy from the prescription record, and the clinician checks it. Mark these forms "medicine" in the queue so they always get a second check.

Rewriting a text message that reads like pressure

Text recalls have a character limit in practice, and it is easy to sound sharp. Here is a before and after, with fictional details.

Before: "OVERDUE: [child first name] has missed their immunisation. Book now to avoid problems."

After: "Hello [parent name], [child first name] is due a routine immunisation at the practice. Booking takes two minutes at [booking link], or call [booking phone]. Reply STOP to pause these reminders."

The first version blames, threatens a vague consequence and makes a clinical claim ("problems"). The second states the task and offers an exit. Ask the AI to produce five variants at different lengths, then choose by reading them aloud as the parent would hear them. Pick one, freeze it and review it twice a year.

Specifying the recall query so the list is right the first time

Most recall errors begin in the query, not the letter. Write the specification before you ask the vendor or your IT contact to build it. A filled-in example:

Recall query: routine immunisations and checks
Include: children registered and active; item due date on or before today + 30 days; no booking for that item.
Exclude: children with a recorded parent refusal for that item; children recorded as moved away or deregistered; children whose preferred contact is "phone only" (handled by a call list).
Output fields: parent name, child first name, item due, due date, contact route, contact address or number, record ID.
Sort: by due date, oldest first.
Check: first run limited to 25 records; two staff compare ten of them against the live record.

The record ID matters because it lets you log each contact without putting extra personal data into the merge file. The "phone only" exclusion matters because sending a text to someone who asked for calls is the sort of thing that generates complaints.

Exceptions: the families the template does not suit

Some groups need a person, not a template. Build a short exceptions list and hand it to a named member of staff.

• Children in care or with a social worker involved, where the contact is not the parent.
• Families recorded as needing an interpreter. Do not auto-translate clinical letters without a check by a qualified person; use the practice's interpreter route.
• Children with complex health needs, where a standard schedule may not apply and a clinician sets the plan.
• Families who previously declined, where a repeat letter could feel like pressure.
• Any contact flagged for safety, such as a protected address.

Say a batch of 100 produces nine of these. Nine tailored contacts take a nurse perhaps an hour, and they are the ones where a form letter would have done harm.

A weekly rhythm for both jobs

1. Monday morning: run the recall query, review the exceptions list (no valid contact, recently moved, opted out) and approve the batch.
2. Daily: the form owner clears "received", the nurse or clinician clears "ready for clinician" before the end of the day.
3. Friday: count recalls sent, bookings made from them, forms signed and any that missed their date. Four numbers on one page.

Count bookings made within 14 days of a recall as a simple response measure. If the first batch of 100 letters produces 20 bookings, that is a baseline; change one thing (the route, the wording or the timing) and compare the next batch of 100.

Charging for forms and setting expectations with parents

Many practices treat school, camp and sports forms as non-urgent paperwork and some charge a fee; others include them in care. Whichever you choose, say it in one sentence in the acknowledgement so parents are not surprised. AI helps by drafting the wording in three tones for your manager to choose from, and by producing a one-page "forms guide" for the website: what you complete, how long it takes, what to bring and when to ask. Keep the guide to facts you can stand behind, since a promised turnaround you cannot meet creates the very chasing calls you were trying to avoid.

A realistic example: a practice that promised "48 hours" on the website found that forms needing a clinician's review took five working days in term-start weeks. The fix was to publish "up to five working days, ten in the first two weeks of term" and to add an "urgent" route for forms due within three days, which needs a phone call. Complaints dropped because the published time matched what actually happened. If you change the number later, update the website, the acknowledgement and the reception script together.

Measuring whether the changes helped

Pick three numbers before you start and read them again after eight weeks: the share of recall contacts that led to a booking within 14 days, the median days from form request to sent, and the count of parent calls chasing a form. Write the starting values down this week. If bookings rise but chasing calls also rise, your reminders may be reaching people while your forms process is still slow, and that tells you where to look next.

Checking the letters before they go live

Before the first real batch, read the final template with a colleague who has not seen it. Look for the lines that could be misread: "overdue" may sound like blame; "required" may sound like a legal duty you cannot support; "free" may be wrong for a private appointment. Send five test messages to staff phones and email accounts to check the merge fields and the booking link. Then send the first real batch to a small group, say 25 families, and review replies by hand for two days.

For other admin jobs a paediatric office can pass to AI, AI tools for paediatricians' offices: seven admin jobs compares the options. If notes are your larger problem, the best AI scribe for paediatrics checklist is the place to start. And if you would like help mapping which letters your system can merge and which need a different tool, that is what my AI implementation consultation covers, through the contact page.

Questions from paediatric practice managers

Can I paste a recall list into ChatGPT?

Not with names, dates of birth or vaccine histories unless your plan and agreement with the vendor cover patient data. The safer pattern is to let your record system build the list and merge the letters, and to use AI only to write and test the template with fictional children.

Should the AI answer parents who ask about side effects?

No. Replies about symptoms, reactions or whether a child should be seen belong to a clinician or your triage process. Build a rule that any message mentioning a symptom skips the automated route and goes to a person the same day.

How do I stop school forms piling up?

Give them a queue with four statuses (received, ready for clinician, signed, sent), a target turnaround and a named owner. AI can pre-fill the factual parts from the record, but the clinician reads and signs every form.

How often should recall letters go out?

Follow your own recall policy and the schedule your practice uses. A common pattern is a first notice, one reminder and a final contact by a different route, with a stop point after which a clinician decides what happens next.

Further reads

• GP Practice DNA Policy: Wording, Reminders and Letters With AI — Missed-appointment letters use similar tone rules.
• How Nurseries Use AI to Write Daily Reports for Parents — Writing to parents about children, with the same care.
• AI for GP Reception: Signposting Scripts and Role-Play Training — Signposting scripts for reception staff who handle replies.
• Using AI for Childcare Observations and Learning Journals — Another place AI drafts text about children for adults.
• Friends and Family Test Comments: Summarise Them With AI — Summarise parent feedback on the new recall process.
• AI Scribe Consent Form: US Wording and State Recording Laws — Consent wording if you later add a scribe.
• AI Marketing Automation for Doctors: Recall, Reminder and Review Flows — Build recall, reminder and review-request flows in the practice software you already own, with consent, opt-outs and a worked month of numbers.
• AI for Nurse Practitioners: Prior-Auth Letters and Practice Policies — Prior-auth letters, patient messages and policy drafts for an NP-owned practice, with a clear line on what needs a signed BAA.
• Childhood Immunisation Reminders: AI-Drafted Letters and Texts — AI-drafted immunisation reminder letters and texts for a practice, plus an overdue-tracking sheet and opt-out handling that plain templates skip.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: no vendor prices or product claims are stated; the workflow follows general practice-administration steps. Check your own recall policy, record-system manual and data-protection advice before use.