Structured Templates or Ambient Notes? Match the Tool to the Appointment

Most practices that have abandoned ambient AI scribing did not abandon it because the transcription was bad. They abandoned it because they switched it on for every appointment in the book, and by week three the nurse was rewriting half the entries by hand at 5:40pm. The technology worked. The deployment did not.

The debate framed as “dental note templates vs ambient AI” is the wrong debate. Both work. Both fail. What determines which one you get is the appointment type you point it at, and almost nobody makes that decision deliberately at the point of purchase.

What ambient capture is actually good at

Ambient tools listen to the room and write the note. Dental-specific ones on the UK market include Kiroku (which has moved from template-assist toward conversational capture), Dentr’s note assistant, and general-purpose medical scribes such as Heidi Health and Nabla that practices have retrofitted with dental prompts. Carestream, Dentally and Software of Excellence have all shipped or announced some form of dictation-to-note layer.

The thing ambient does that templates cannot is capture unstructured human speech about uncertainty. That is the core content of a new patient exam and almost the entire content of a treatment planning consultation.

Consider a 30-minute new patient exam. The clinically meaningful material includes: the patient’s own account of the sensitivity on the upper right (“only when I drink something cold, and only for a second or two, it’s been about four months”), the smoking history negotiation, the fact that they are terrified of extractions because of something that happened when they were eleven, and the shared decision that you will monitor the UR6 crack rather than crown it now, with the patient explicitly accepting the risk.

A template captures none of that well. You would need free-text boxes for all of it, at which point you are typing, not templating. Ambient capture gets it verbatim and structures it afterward.

Here is what a Heidi output looks like for that visit, roughly, after a dental-specific prompt:

SUBJECTIVE
Pt reports cold sensitivity UR quadrant, ~4/12 duration.
Sharp, non-lingering (<5 sec), no spontaneous pain, no
disturbed sleep. No analgesic use. Smoker, 10/day, 18 yrs;
discussed cessation, pt declines referral at this time.
Dental anxiety re: extraction, historic (age 11).

OBJECTIVE
[clinician-entered charting]

ASSESSMENT / PLAN
UR6 MOD amalgam with mesio-buccal cusp crack, asymptomatic
to percussion and bite test. Options discussed: cuspal
coverage onlay vs monitor. Risks of catastrophic fracture
and possible RCT/XLA explained. Pt elects to monitor.
Review 6/12 with periapical.

That consent conversation, written up in the patient’s terms, is the difference between a defensible note and a note your indemnifier winces at. It took you nothing to produce. The tool heard it because it was said out loud.

Ambient shines wherever the note is a record of a conversation. New patient exams, recall exams, treatment planning, complaint resolution meetings, consent discussions for implants or ortho, paediatric behaviour management appointments where what you did was mostly talk.

Where ambient capture falls over

Now run the same tool through a molar endo.

The clinically meaningful material is: working lengths for each canal, apex locator readings versus radiographic confirmation, file sequence and final master apical file size, irrigant concentrations and volumes, activation protocol, intracanal medicament, obturation technique and sealer, number of rubber dam clamps and whether isolation was maintained throughout.

Almost none of that is spoken aloud. You are not narrating “irrigating with three millilitres of three percent sodium hypochlorite” to a nurse who already knows. You say “hypo” and hold your hand out. Ambient AI gets a transcript full of suction noise, the radio, and a conversation about the patient’s holiday.

We have seen practices where ambient notes for endo appointments arrived with roughly 40% of the required data fields empty, versus something like 3% for structured template entry, and the clinician then spent longer filling gaps than they would have spent on a template from the start. The numbers vary by practice, but the direction never does.

Restorative and surgical visits have a second problem: the data is numeric and repeatable. Twenty-two canals a month, each needing the same six fields. That is a form, not a story. Forcing a narrative tool to produce structured numeric output is asking it to do the one thing it is worst at, and it will hallucinate the gaps rather than leave them blank, which is considerably worse than leaving them blank.

The split, appointment by appointment

Roughly speaking, in a mixed NHS practice that split lands at 55 to 60% ambient by appointment count, dominated by exams and recalls, and 40 to 45% template, dominated by restorative. In a predominantly private restorative or implant practice the ratio inverts to somewhere near 30/70.

Why mixing them badly stalls adoption

Three failure patterns account for most of the abandoned rollouts.

The first is all-ambient. A principal buys seats for the whole practice, everyone switches it on Monday, and the associates doing four-surface composites all day discover it produces notes that need heavy editing. They stop using it. Two of the associates keep using it for exams, quietly, and the practice manager cannot work out why utilisation is 38% and concludes the purchase was a mistake. It wasn’t. It was a scoping mistake.

Second: all-template. This is the older failure and it is still common. Practices with 60 templates in Dentally or SOE, each one a fossil of some past MDU circular, and the exam template has a 400-word free-text box that everyone fills with three words. The templates are excellent for the restorative work and actively harmful for the exams, because a template teaches clinicians to record the fields that exist rather than the things that happened.

Third, and this is the subtle one: template-shaped ambient. The practice buys an ambient tool and then configures it to output into their existing 22-field restorative template. The tool now has to invent values for fields nobody spoke about. You get “isolation: rubber dam” on an appointment where you used cotton wool rolls, because the model inferred it from context. This is the configuration that produces genuinely dangerous notes, and it is the most common thing people do when they are trying to be sensible.

Your baseline for deciding what “good” looks like should be your own charting standards rather than the vendor’s demo. If you have not written those down, the wider clinical notes and charting material is the place to start, because you cannot evaluate a tool against a standard you have not articulated.

Running the split in practice

The mechanics are less painful than they sound. Most ambient tools take a per-appointment toggle, and appointment types already exist in your diary software. Map them once.

Set the default to off, and let the appointment book turn it on. In Dentally, that means tagging your exam and consult appointment reasons; in SOE you can drive it off the appointment book codes. The clinician should never have to decide at the chairside, because at the chairside they will always pick whichever is currently on.

Budget properly for the template half. Ambient licences run somewhere around £60 to £120 per clinician per month in the UK market as of this year. Templates are free and they cost you two afternoons of a senior clinician’s time to write well, which is a real cost that nobody puts on the business case. Write eight or ten good ones rather than sixty mediocre ones. An endo template, a direct restoration template, an indirect prep template, an extraction template, an implant template, a perio therapy template. That is most of your restorative volume covered.

Audit at six weeks, not six months. Pull twenty notes, ten from each modality, and ask a clinician who was not in the room to tell you what happened and whether they could defend it. That test finds the problem faster than any utilisation dashboard.

One more thing worth saying to your associates before you switch anything on: ambient tools are recording the room, which means the nurse’s aside, the trainee’s question and the patient’s off-hand comment about their ex-partner are all going into a third party’s processing pipeline. Check the DPIA, check whether the audio is retained or discarded post-transcription (Heidi discards by default, several others do not), and tell the patient. A verbal notice at the start of the appointment, logged once in the notes, is what most indemnifiers are currently asking for.

The practices getting real value out of this have stopped thinking of it as buying a note-writing product. They have decided, appointment type by appointment type, whether the record they need is a story or a form, and bought accordingly. That decision takes an hour with the diary open, and it is the hour that determines whether the £1,400 a month you are about to spend produces notes anyone wants to read.