Best AI Scribe for Paediatrics: What to Check Before You Pick

A paediatric practice should check five things before picking an AI scribe: how it handles several voices in one visit, how consent is captured, whether templates fit well-child and acute visits, whether the draft stays accurate on ages and doses, and the real per-clinician cost. Test two tools on one scripted visit and read the drafts side by side.

Paediatrics is harder for scribes than adult consultations because the conversation is not a pair. A visit might include a parent, an anxious older sibling, a child who answers some questions and ignores others, and a clinician who talks to the child in one voice and the parent in another. Most reviews of scribes assume two speakers, so test on the three- and four-voice visits you actually have.

Follow me on Instagram@sagnikteaches

Test the speakers first: one scripted visit

Before comparing prices, write a two-page visit script and play it into each tool, or act it out with a colleague. Make it hard on purpose.

Connect on LinkedInSagnik Bhattacharya

• A parent who gives the history and interrupts the child.
• A four-year-old who says "it hurts" and points rather than describing.
• A sibling asking loudly for a snack.
• A clinician who asks the same question twice, once to the child and once to the parent.
• A moment where the parent says something the child should not hear, and the clinician asks them to step outside.

Read the draft for three failures: sentences attributed to the wrong person ("patient reports abdominal pain for three days" when the parent reported it), sibling chatter that reached the note, and the child's own words lost. An illustrative error: the parent says "she's had a temperature since Tuesday" and the draft writes "patient states she has had a temperature since Tuesday". The fix is a template that distinguishes "history from parent" and "child's account", and a tool that lets you correct attribution quickly.

Subscribe on YouTube@sagnikteaches

Ask each vendor directly whether the tool separates speakers, in how many languages, and whether it handles a clinician and an interpreter in the same visit. Of the pages I read on 4 November 2026, only Freed said anything on languages (more on that below), and none said how speakers are told apart, so treat each as a question for the trial, not an assumption.

Consent: parents, children and teenagers are different cases

Whose agreement you need to record or process a visit depends on the child's age and on local rules, so use your professional body's guidance and a qualified adviser for the wording. The practical shape is usually this:

• Young children. A parent or guardian agrees. Tell the child in plain words, if they are old enough to follow: "I use a computer helper that listens so I can write my notes."
• Older children. The parent agrees, and the child is asked too. A child who says no should be respected where you can.
• Teenagers. Confidential parts of a visit, such as a private conversation about health, mental wellbeing or relationships, may need to stay out of any recording, and out of any note the parent can see. Decide in advance how the tool will be paused, and test the pause button.

A sample consent line for the booking confirmation or waiting-room sign:

We use an AI note-taking tool to help us write your child's visit notes.
It listens during the appointment, and a clinician reviews every note.
You can say no at any time and we will write the notes by hand. Please
tell us at reception or at the start of the visit if you would prefer we
do not use it.

Add a line in the chart when a parent declines, so the next clinician does not switch it on. Our tutorial on handling children's data with AI covers the data side of this in more depth.

Audio handling and the agreement behind it

Children's visit audio is sensitive. Ask each vendor the same five questions: is audio stored and for how long; where is it processed; is any of it used to train models; will you sign a business associate agreement (BAA) for my plan; and what happens to the audio and transcript when I delete a note.

The answers on the pages I read differ in useful ways. Heidi says it is a business associate and makes a BAA available to every covered entity it works with, and its support centre says live sessions are transcribed in real time and the audio is not stored; how long transcripts are kept is a setting in your account. Freed says audio is saved temporarily until notes and quality checks are complete and then deleted, that it signs BAAs with healthcare organisations (without naming a plan, so ask for yours), and that its AI is trained only on de-identified notes. Freed's FAQ also suggests wording for asking patients' permission, but neither vendor describes a feature that records consent, so consent stays your job. For a longer treatment see AI scribe security questions on audio, retention and BAA.

Templates: well-child, acute and developmental visits

A paediatric practice has at least four visit types, and a scribe that produces one generic note will cost editing time on all of them.

Freed's page lists specialty templates on its Starter plan, names paediatrics among the specialties it writes for, and adds an instant template builder on Core, while Heidi lists standard templates on its free tier and limits advanced templates there. Ask each vendor whether a paediatric template exists, and if not, how easy it is to build and share your own with colleagues. Template-sharing across clinicians may sit on a higher plan: Heidi lists team template sharing only on its Teams plan, and Freed lists note sharing under its Groups plan.

Accuracy on ages, weights and doses

Children's dosing depends on weight and age, and a tool that misstates either puts a draft one signature away from harm. Speech-to-text tools confuse numbers ("fifteen" and "fifty"), drop units, and mishear drug names. Build a short number check into your routine:

1. Say the weight, age and dose aloud once, clearly, with units: "weight fourteen kilograms, amoxicillin two hundred and fifty milligrams in five millilitres, five millilitres twice a day".
2. After the draft appears, read only the numbers against your own record before signing.
3. Never let a tool calculate a dose for you. It may draft text, but the arithmetic is yours.

An illustrative failure: a clinician dictates "two point five millilitres" and the draft says "twenty-five millilitres". It reads smoothly and passes a quick skim. The fix is the habit of reading numbers separately, and a template that places dose lines in their own section.

Per-clinician cost: what the pages say

Price pages I read on 4 November 2026 gave this much, and no more:

Do the sum for your own volume. A clinician seeing 18 patients a day, four days a week, has about 288 visits a month. A 40-note cap is under 14 percent of that. A $39 plan is therefore a trial, not a working plan, for that clinician, and the cost comparison should use the unlimited tier. If Freed's Core tier, displayed at $79 a month, saves six minutes a visit, that is 1,728 minutes, or about 29 hours, saved for every clinician each month. Treat those figures as an illustration and replace them with your own timed numbers.

For a wider view of what small clinics pay, see AI scribe costs compared.

A visit, before and after: what to look for in the draft

Here is an illustrative acute visit, so you can see the kind of error to hunt for. A parent brings a five-year-old with two days of cough and a temperature. The clinician says aloud: "Five-year-old, two days of cough, fever up to thirty-eight point five at home, eating less but drinking well, no vomiting, no rash. Chest clear, ears normal, throat mildly red. Viral upper respiratory infection. Fluids, paracetamol if uncomfortable, return if breathing is fast or he is not drinking."

A draft that goes wrong might read: "Patient is a five-year-old with a two-day history of cough and fever to 38.5. Reduced appetite, vomiting present. Chest clear, ears normal, throat erythematous. Diagnosis: viral URTI. Advise fluids, paracetamol, return if breathing is fast or not drinking." Spot it: "vomiting present" contradicts "no vomiting". It is a one-word slip in an otherwise good note, and a quick read would miss it because the rest is fluent. This is the argument for reading every negative statement, and for a template where symptoms are listed as present or absent in a separate block.

A draft that goes right keeps the negatives, labels the history as from the parent, and leaves the safety-net advice as you said it. Check one more thing: the note should not contain a drug dose you did not state. If it does, that is a sign the tool is filling gaps from habit rather than from the visit.

Teenagers and private conversations

With adolescents, much of what matters is said when the parent is out of the room. Decide before you switch anything on:

• Whether the tool is paused, stopped or left running during the private part of the visit.
• Whether a parent can see the note through a portal, and which sections are shared.
• What the tool does with a private conversation that was captured by mistake: can the segment be deleted from the note and the transcript, and is the audio removed too.

A short script for the clinician: "I'm going to pause the note-taker while we talk on our own. It will go back on when your parent comes back in." Say it out loud so the young person hears it, and test the pause and resume buttons before you rely on them. A failed pause is the sort of error that undermines trust with a teenager for years.

Interpreters, accents and background noise

Paediatric rooms are noisy and often multilingual. When an interpreter is used, a scribe may hear three people in two languages. Freed's FAQ says it can capture a visit in more than a dozen languages and write the note in English, which helps with a family speaking one of those but is not the same as handling an interpreter and a parent talking over each other. Ask each vendor which languages it supports for speech and for the note, and test with a recorded visit that includes an interpreter. If the tool does not handle the second language, decide whether the interpreter's words should be captured at all, or whether you dictate in your own language after the visit. For noisy rooms, check how well it copes when a child cries, and whether the draft says "[inaudible]" or fills the gap with a guess. A visible gap is better than an invented sentence.

A scoring sheet filled in for two tools

Score each line from 0 to 2 after your scripted visit, weighting what matters most to your practice. This is an illustrative sheet for a three-clinician community practice.

Tool A: 3 + 6 + 6 + 4 + 4 + 2 + 1 = 26. Tool B: 6 + 6 + 3 + 0 + 4 + 4 + 2 = 25. Close, but the lines tell you more than the total: A is stronger on templates and numbers, B on speakers and agreements. Since speaker attribution is the paediatric weak point, and a template can be built in an afternoon, a practice might reasonably choose B and build its own templates. The sheet makes that trade-off visible and stops the decision resting on a nice demo.

Parent handouts and letters

Some plans draft patient instructions, referral letters and visit summaries as well as notes; Freed lists these on its Premier plan. They save time because the clinician already said the content during the visit. They also repeat the dose-and-age risk, in a document a parent will follow at home. Read every handout as though you were a worried parent at 2 am: is the dose written with units, is the age right, does it say when to come back, and is there anything you did not say. Print one example in the trial and have a colleague read it cold.

How to tell after a month that the tool is helping

Count three things: minutes per note before and after, number of errors found at review, and the number of parents who declined. Suppose before the pilot each note took nine minutes to write after the visit and now takes three to review. For 288 visits a month, that is 1,728 minutes saved by one clinician, minus time spent on corrections. If errors found at review are more than one in twenty, change the template or the dictation habits before adding clinicians. If more than a handful of parents decline, revisit how the consent line is introduced. Reception staff say it better than a sign does.

Five checks on the draft itself

1. Negatives. "No rash, no vomiting" must not turn into "rash, vomiting".
2. Red flags you did not say. The tool should not add "no meningism" because it seems likely.
3. Parent versus child. History attributed to the right person.
4. Timings. "Three days" stays three days.
5. Safety-net advice. The drafted advice matches what you said, including who to call and when.

Run these checks on twenty drafts in the first week and keep a tally of errors by type. If one type repeats, change the template or the way you speak, not just the draft.

Three practices, three starting points

A solo paediatrician. Start with a free tier to learn the workflow on routine visits, then move to a paid seat after a fortnight. Keep teenagers' private conversations out of the tool.

A three-clinician community practice. Pick one tool for all three so templates can be shared, and ask for a group quote. Agree one consent line and one correction routine.

A practice with a large developmental caseload. Long visits with several speakers and observation notes. Test the tool on a recorded role-play of a developmental review, because these are the visits where attribution matters most.

Roll-out in four steps

1. Pick two clinicians and one visit type, such as well-child checks.
2. Run two weeks with consent and a daily error tally.
3. Review the tally and parents' reactions, and change the consent wording if people asked questions.
4. Add acute visits, then the rest of the team.

If you would rather plan this with someone, that is what my AI implementation consultation covers. For a baseline on how scribes work in a consultation, start with what an AI scribe is and how it works, and for a parallel checklist in another profession see AI scribes for occupational therapists.

Paediatric scribe questions worth asking

Does an AI scribe handle a parent, a child and a sibling talking at once?

Not reliably, and vendors rarely publish how well. Test it with a scripted visit that includes a toddler, a talkative parent and background noise, then read the draft for misattributed sentences. Ask whether the tool separates speakers and whether you can correct the attribution.

Who consents when the patient is a child?

Usually the parent or guardian, with the child's own agreement sought where age allows. Teenagers raise confidentiality questions of their own. Rules differ by place and service, so follow your professional body's guidance and ask a qualified adviser for your policy wording.

Is a free tier enough for a paediatric clinician?

A free tier is a good trial but rarely enough for a full clinic. Caps on notes, advanced templates or data usually bite within weeks. Count your visits per month and compare against each plan's limits before relying on it.

Should the scribe write the parent handout too?

It can save time, provided a clinician reads it before it goes out. Some tools draft patient instructions and letters on higher plans. Check dosing, ages and follow-up timing line by line, because small errors matter more with children.

Further reads

• Is an AI Scribe Worth It for a Small Private Clinic? — Whether a scribe pays for itself in a small clinic.
• How to Roll Out an AI Scribe Without Losing Patient Trust — How to introduce a scribe without losing patients' trust.
• AI Scribe for GPs: Costs, Free Tiers and the NHS Pilot Limit — Free tiers and costs for family-medicine style practices.
• AI Dictation Software for Doctors: Dictation or Ambient Notes? — Dictation or ambient notes for doctors.
• AI Scribe for Dentists: Will It Write Back to Your Software? — Whether a scribe writes back into your software.
• AI Scribe for Medical Practices: Pick by Country and EHR — Pick a scribe by country and EHR.
• Best AI for Paediatricians: Vaccine Recalls and School Forms — Build a vaccine recall list, letter templates and a school and camp form queue with AI, keeping parent data safe and symptom advice out.
• AI Tools for Paediatricians' Offices: Seven Admin Jobs — Seven front-office jobs for AI in a paediatric practice, from phones to rotas, each with an example and a rule for using it safely.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Heidi pricing, compliance and data security pages, Freed pricing and security pages, Carepatron pricing page and published BAA (all read 4 November 2026); general scribe practice. Vendor features and prices change, so confirm each point on the live page and in writing.