Healthcare is not immune to this AI age, with more cases of ‘patients’ consulting AI with their physical and mental health issues, occasionally with concerning outcomes. In dentistry too, AI scribes and diagnosticians (most notably in radiography) are becoming more prevalent – with the scribe being the vanguard of the AI wave about to wash over the profession.

Yet most of us who have dabbled with AI – playing around with ChatGPT usually being the gateway – know that it is fallible. We see it ‘hallucinate’ and soon realise it doesn’t think like us – it a skill to even understand how to ask the right questions. We’re not quite there yet – the list of unrealised predictions for autonomous Tesla vehicles made by Elon Musk even has its own Wikipedia page. 

Why dentists remain responsible for AI‑generated notes 

As is always the case with new technology, regulation and the law tend to plod along rapid advances. It is no different in dentistry. Akin to the driving analogy (where drivers are still primarily responsible for control of the car, regardless of the sophistication of the self-driving aids), in the current legal and regulatory environment, practitioners remain responsible for the AI generated outcomes in patient care. 

With respect to AI scribes, regardless of how a clinical note is generated, the treating practitioner who signs it off is responsible for its contents. If an AI scribe misinterprets, omits, or embellishes information and the clinician fails to review and correct the record before it enters the patient's notes, that inaccuracy becomes the clinician’s inaccuracy in the eyes of a regulator and the legal system. 

Case study 

Mr V, a regular patient of the practice in his late fifties attends Dr C for a routine examination. He is a known heavy smoker (20 cigarettes per day). This is well documented in the notes, and Mr V has mentioned in passing at previous appointments that he drinks most evenings, though this has never been formally recorded in units.

Mr V is the kind of patient the whole practice knows well - gregarious, self-deprecating, always quick with a joke, and flippant about anything that sounds like health advice.

During the soft tissue examination, Dr C notices a suspicious mixed red and white lesion on the right lateral border of the tongue. There is no obvious cause of frictional irritation and Dr C is sufficiently concerned that this has the appearance of erythroleukoplakia – a high-risk presentation, at a high-risk site, in a patient with two of the most significant risk factors for oral malignancy. Dr C is taking all of this into consideration in her differential diagnosis and asks the patient directly how long it has been there and whether it has been causing any discomfort. 

Mr V takes the hand mirror, glances down, makes a show of examining his own tongue in the dental light, and says with a grin: "no idea, probably been there forever, I'm not worried about it – I'm invincible." 

Dr C, who knows this patient well and understands exactly the sentiment in which this comment is being made, is not reassured by it. She is forming a clinical judgement that this lesion needs urgent attention. She continues dictating her examination findings to her AI scribe and then plans to do a routine clean before discussing the need for referral with the patient. 

With reference to the soft tissue finding, the AI records: "mixed red/white lesion noted lateral border tongue right side, patient reports lesion longstanding, patient states he is not concerned, no pain reported, patient confident no cause for worry." 

Dr C carries out the scale and polish and sits Mr V up to discuss the interdental bleeding and deliver some brief oral hygiene instruction, resulting in the usual dismissive quips. As this is occurring, Dr C’s assistant starts to change over the instruments, de-activates the scribe (as is their protocol when an appointment is finished) and disappears to the steri room.

By this time Mr V is standing and Dr C invites him to sit again to discuss her strong concerns about the lesion, which in combination with the known risk factors, gives her cause to recommend an urgent referral to an Oral Medicine Specialist. The word “cancer” is even mentioned as a possibility, to emphasise the gravity of the situation. Mr V does appear to straighten up and listen more than usual and then reverts to type. “Ah, we’re all going to die one day. I might as well have a good time on the way,” he says.

A referral card is issued with clear instruction for Mr V to book an appointment as a matter of urgency. 

By this time Dr C is running late, and at the end of the day she instructs the scribe to dictate a referral letter to the recommended Specialist based on the soft tissue findings on the tongue - to assess and manage the suspect lesion. 

The referral letter is generated from the information recorded by the scribe and sent. Regrettably, the referral is marked as routine rather than urgent and the patient is placed on the applicable waiting list at the Specialist.

Two months pass and close to his allocated time, Mr V reschedules his appointment, then cancels the next one given to him two weeks later without re-booking. By the time the Specialist passes care back, and Dr C is live to the situation, it is almost three months down the track. 

It is only at this point that Dr C reviews the referral letter and realises with horror the lack of urgency conveyed in the correspondence. This was not at all what was intended and she can see now that the scribe must not have been active during her conversation at the end of the appointment.

Dr C urgently tries to contact Mr V. When she finally speaks to him, he says he no longer wants to follow it up. He is more aware of it now and it might be slightly bigger, but he assures her that he still is, by his own account, “invincible”. Dr C tells him it’s time to get serious. She will make an urgent appointment for him, and in the spirit of their long-standing relationship she half-jokingly threatens to drive him there herself if he won’t go. 

When he is eventually seen by the Specialist, the lesion has progressed significantly. Mr V is referred immediately to the local tertiary hospital for assessment and work up at the Head and Neck Cancer Unit. 

During the oncology team's review of the file, including the dental notes, in preparation for the case discussion, a clinician reads the record and notes that at the time of referral no particular concern is expressed and the lesion is described as longstanding and asymptomatic. This is factored, along with the routine rather than urgent grading of the referral, into a narrative that was built up about a lesion that was stable and of minimal concern to all parties, rather than one that required immediate review. 

The entry "patient confident no cause for worry" has mutated from a misinterpretation of false bravado, into a formal clinical document, into a referral letter — each step moving it further from its original context, lending it a weight and authority it was never meant to carry, and ultimately narrowing the window of opportunity for urgent and timely investigation of the lesion. 

It started as a punchline. It ended as part of a multidisciplinary case discussion by a Head and Neck Cancer team. 

Learning points

Currently, clinician responsibility for record accuracy remains absolute. 

"The AI wrote it" is not currently recognised as a defence in clinical negligence or regulatory proceedings. The prudent position is to treat AI-generated notes as a draft requiring careful clinician review and sign-off rather than a finished record. 

Errors that read as competent clinical reasoning are the most dangerous.  

As the example in this article illustrates, the most harmful AI scribe errors may not be obvious transcription mistakes, but convincing clinical entries that have the potential to silently misdirect future care. A note that attributes a suspicious lesion to a pizza burn or converts a sarcastic remark into a reassuring patient history, may never be questioned precisely because it plausibly reads as the recording of a careful clinician.

Compounding errors across multiple clinicians have potential to create complex liability.  

When an AI-generated inaccuracy propagates through a patient's record and shapes the clinical decisions of several subsequent clinicians across months or years, establishing where responsibility lies becomes extremely difficult. Each clinician acted in good faith on the record as they found it, yet the cumulative harm may be substantial. This diffusion of responsibility does not necessarily protect individual practitioners – it may expose all of them.