AI medical scribes vs assessment platforms for ADHD

An AI medical scribe turns one recorded conversation into a well-drafted clinical note. An ADHD diagnostic report is a different artefact: it synthesises the interview with parent and teacher questionnaires, school reports and developmental history, and it must survive audit. Done well, that synthesis can still be fast: on Cerebric pilots, drafting the report fell from around two hours to under 30 minutes.

Both tools use the same underlying technology, speech recognition plus a language model, which is why they get confused with each other. But they solve different problems, and buying the wrong one for your bottleneck can be a costly mistake. This article sets out what scribes genuinely do well, what a diagnostic report actually requires, and where the line between the two sits.

What an AI scribe does well

An AI scribe (also called an ambient scribe, or AVT) listens to a consultation, transcribes it, and drafts the clinical note in your chosen template while you keep your attention on the patient. For the job they were built for (the short, high-volume consultation typical of GP and primary care), the leading tools do it well.

Heidi Health has real NHS traction, listing NHS Dudley, Hertfordshire and West Essex, Walsall Healthcare and WWL Teaching Hospitals among its users, alongside dictation, template switching and coding support. Accurx Scribe, from the company behind much of NHS primary care messaging, reports documentation time per appointment falling from around seven minutes to just over one, with one-click saving into EMIS and SystmOne.

Their sweet spot is the short encounter: a ten-minute appointment, a follow-up, a titration review, where the note is essentially a faithful account of one conversation. A long neurodevelopmental assessment is a different shape of encounter, and scribes are a weaker fit for it. Even where a scribe is used during an assessment, it only compresses the part of the work it actually touches, the transcription and first-draft prose, while the multi-source synthesis a diagnostic report needs still has to be done by hand.

So, put simply: if your bottleneck is turning conversations into notes (follow-ups, titration reviews, general psychiatry clinics), a scribe can be useful.

What an ADHD diagnostic report actually requires

An ADHD diagnostic report is a multi-source clinical document, not a long consultation note. NICE guideline NG87 is specific about what sits behind a diagnosis: a full clinical and psychosocial assessment, a complete developmental and psychiatric history, and observer reports alongside a mental state examination. It restricts diagnosis to specialist psychiatrists, paediatricians and other appropriately qualified professionals with training in ADHD, and, critically for anyone hoping to automate the whole thing from a transcript, states that a diagnosis "should not be made solely on the basis of rating scale or observational data".

NG87 also requires that symptoms cause at least moderate psychological, social, educational or occupational impairment, present in two or more settings. Two or more settings means two or more informants. In practice a defensible report weaves together:

• the clinical interview (a structured instrument such as DIVA for adults, or ADOS-2 and ADI-R for autism pathways)
• parent, teacher or partner questionnaires (NICE names Conners' rating scales and the Strengths and Difficulties Questionnaire as useful adjuncts)
• school reports, occupational evidence and prior clinical records
• a developmental history that usually predates the appointment by decades

What sits behind an ADHD diagnostic reportCaptured by an AI scribeGathered, scored and synthesised by handAI scribe hears thisClinical interviewDIVA · ADOS-2 · ADI-RInformant questionnairesConners, SDQ, each scored by handSchool reports & prior recordsOccupational evidence, past notesDevelopmental historyOften predates the appointment by decadesADHD diagnostic reportClinician-led synthesisper NICE NG87The three unshaded sources are gathered, scored and synthesised by the clinician, not the scribe. A scribe compresses one input; the clinician still owns the report.A scribe sees one source. The report synthesises four.

A scribe, by design, sees exactly one of those sources: the conversation in the room. Everything else still has to be chased, scored, cross-referenced and written in by hand. That is why clinics using a scribe for assessments still report substantial write-up time: the scribe compresses the transcription-and-prose step, not the synthesis. (For UK turnaround norms, see how long an ADHD report should take.)

The verification gap

AI makes drafting a report easy; the effort simply moves downstream, to verification, the step where a clinician confirms that every claim in a draft is true before signing it. For a neurodevelopmental report, whose length and range of sources are unlike a standard consultation note, thoroughly checking that accuracy is effectively a new workflow the clinician now owns. It is also where the scribe and the platform genuinely diverge.

With a consult note, verification is easy: you were in the conversation ten minutes ago, so you read the draft against memory. With a diagnostic report, that breaks down. The draft makes claims about a teacher's questionnaire from six weeks ago, a school report from 2014, and something said in minute 97 of a 150-minute interview. Checking those from memory is not verification. It is hoping. And an unverified AI-drafted diagnostic document is precisely the thing that will not survive an audit, a complaint, or a tribunal.

Where one report claim actually comes fromChildhoodDevelopmental history, decades earlier2014School report6 weeks agoTeacher questionnaireMinute 97of a 150-minute interviewTodayReport drafted and signedEvery claim in the draft must trace back to one of these sources, not be recalled from memory. Checking a diagnostic draft from memory is not verification; it is hoping. This is why per-claim citation and a side-by-side evidence view matter.A single claim can rest on a 2014 school report, a childhood developmental history, a six-week-old questionnaire and minute 97 of the interview: none verifiable from memory.

This is the problem assessment platforms are built around. Cerebric's approach is that every sentence in a drafted report is traceable to its source (a transcript timestamp, a questionnaire answer, or an uploaded document) and reviewed in a Split View that shows the claim and its evidence side by side. Sections can be regenerated individually rather than re-rolling the whole document, every change is tracked, and QA checklists gate the report at approval. Verification becomes a systematic pass rather than a re-read, which is how the whole draft-to-signature pass drops to under 30 minutes.

Clinicians notice the difference where the structure lives. As one consultant psychiatrist using Cerebric put it: "I had my report, recommendations and all, finished by 5pm… I love how quick it makes filling in all the DIVA boxes." The DIVA boxes are the point: a diagnostic report is mostly structure and evidence, and prose is the easy part. We cover the audit side in more depth in making AI clinical reports audit-defensible.

When to use each

A scribe compresses one step, turning a single conversation into a note. An assessment platform is the infrastructure around the whole pathway (screening and intake, patient and informant questionnaires and scoring, scheduling, recorded assessment, report drafting, then letters and outcome reporting), and unlike a scribe, a platform like Cerebric handles the long neurodevelopmental assessment and the short consultation equally well. So this is rarely about running both tools; it is about whether one step or the whole pathway is your bottleneck.

Where each tool covers the assessment pathwayAI scribeAssessment platformScreening & intakePatient & informant questionnaires + scoringSchedulingRecorded assessmentReport draftingLetters & outcome reportingAI scribe: compresses one stepfirst-draft prose onlyAssessment platform: covers the pathwaypathway operations end to end: documentation and scheduling, not diagnosisA scribe compresses one step. A platform is pathway infrastructure. The clinician still makes every clinical decision.A scribe compresses one step. A platform is pathway infrastructure.

The fair comparison looks like this:

If you read the left column and nodded, buy a scribe: Heidi and Accurx are strong options with real NHS footprints. If you read the right column and nodded, a scribe will help with one step and leave the rest untouched.

The takeaway

Match the tool to the bottleneck, not the demo. Time one full assessment end to end (referral to signed report) and note where the hours actually go. If most of them are in typing up conversations, a scribe pays for itself quickly. If most of them are in chasing informants, scoring questionnaires, assembling multi-source reports and verifying drafts before sign-off, a scribe compresses only the smallest slice, and an assessment platform is the category to evaluate. Either way, put the same three questions to every vendor: where does each sentence of the output come from, how do I verify it, and what happens when an auditor asks the same thing.