AI in Perio

A patient arrives having already consulted ChatGPT about their bleeding gums. They have a working diagnosis, a list of treatment options, and a pointed question about why you haven’t mentioned laser therapy. Meanwhile, your radiograph analysis software has flagged bone loss the naked eye might have missed, your note-writing tool has drafted the clinical letter before you’ve taken your gloves off, and the complaint that landed in your inbox this morning is four pages long, structurally impeccable, and references the 2017 classification system correctly! Welcome to the dental surgery in 2026. The algorithm didn’t replace you. It just moved in.

The Radiograph

The most clinically mature application of AI in periodontics is also the most quietly significant. Automated bone level measurement on periapical radiographs - identifying alveolar bone loss, furcation involvement, CEJ position - has been studied extensively, and the findings are uncomfortable for those who assumed radiograph reading was an exclusively human skill. In controlled studies, AI platforms have performed comparably to experienced periodontists in detecting bone loss, with some demonstrating superior consistency. The advantage is not necessarily accuracy on a single image. It is the removal of inter-examiner variability - the same radiograph, read with the same criteria, every time, regardless of how tired the clinician is or how many patients have been seen that afternoon.

Platforms such as Pearl AI are already being used in practices to flag findings, generate structured reports and provide a second-opinion layer on routine radiographic assessment. The critical caveat, and it matters, is that AI reads the image it is given. Positioning errors, angulation inconsistencies and image quality remain entirely human problems. An AI that reads a poorly angulated bitewing with the same confidence it applies to a perfect periapical is a liability, not an asset. The clinical implication is straightforward: AI as a second pair of eyes, not a replacement for clinical judgement. It should make you more rigorous, not less attentive.

The Letter

If radiographic AI is the most clinically significant application, AI-assisted documentation is the most immediately transformative for day-to-day practice. The referral letter that previously took fifteen minutes (or more) to compose - recalling pocket depths, summarising staging and grading, articulating the clinical rationale - now takes three minutes to review and send. Ambient voice tools transcribe clinical encounters in real time, generating structured notes that capture the appointment as it happens. Periodontal charting summaries translate raw numbers into readable clinical narratives. The administrative burden that quietly erodes clinical time is, for many clinicians, the most tangible early benefit of AI in practice.

The risk is proportional to the convenience. AI-generated clinical records can be fluent, plausible and wrong. The phenomenon of hallucination - where a language model confidently generates information that was never in the source material - is not theoretical in clinical documentation. It is a documented problem. A referral letter that states findings the clinician never recorded, or a clinical note that inverts the treatment sequence, is not merely inaccurate. In a medicolegal context, it is dangerous. The principle must be non-negotiable: AI as a first draft, clinician as the author. The record carries your name. It must reflect your judgement.

The Treatment Plan

AI risk stratification tools are beginning to enter periodontal practice - using staging, grading, systemic risk factors and patient history to generate treatment recommendations and predict disease trajectory. The appeal is obvious: pattern recognition across large datasets, identification of Grade C risk profiles, flagging of cases where the clinical picture suggests a response that non-surgical treatment alone may not achieve. Under genuine time pressure, in a busy general practice where periodontal assessment competes with everything else, a tool that prompts “this patient’s risk profile warrants specialist consideration” before the appointment ends has real clinical value.

The honest limitation is equally real. AI is trained on datasets - and datasets reflect populations, settings and clinical contexts that may not map onto the patient in front of you. The algorithm that performs well across ten thousand cases from a university hospital in one country may not account for the complexity of your patient’s specific biology, their treatment history, or the conversation you had three minutes ago that changed your entire understanding of their compliance. Where AI helps most is in prompting consideration of factors the clinician might deprioritise under pressure. Where it falls short, consistently, is in the irreducible particularity of the individual patient. The patient in front of you is not a dataset.

The Patient Who Did Their Research

Nothing has changed the texture of the clinical consultation quite like the patient who arrives having consulted an AI before you. They have a differential diagnosis. They have read about osseous surgery, about antimicrobial photodynamic therapy, about the microbiome. They have a specific question about a treatment you didn’t mention and a quiet suspicion about why. This is not a niche phenomenon. It is the new baseline for a significant and growing proportion of patients - and it requires a recalibration of how we position ourselves in the room.

The instinct to be defensive is understandable and almost always counterproductive. The AI-informed patient is not a threat to clinical authority. They are an opportunity - for deeper consent, more engaged treatment partnership, better long-term compliance. The framing that works is not “the internet got it wrong” but “you’ve done good research, let me show you how it applies specifically to what I’m seeing in your mouth.” Position yourself as the interpreter of the information, not its adversary. The clinician who can navigate this conversation fluently is not losing ground to AI. They are demonstrating exactly what AI cannot do.

The Complaint

This is the section nobody is talking about loudly enough. AI has fundamentally changed the complaint landscape - not by generating complaints, but by enabling patients to write them in ways they never could before. The frustrated patient who previously sent a handwritten note, or made an angry phone call, or said nothing at all, can now produce a structured, articulate, four-page formal complaint that references clinical guidelines, cites the relevant regulatory framework and identifies specific gaps in the documented care. The floor of complaint quality has risen dramatically. Practices that have not updated their documentation standards accordingly are exposed in a way they were not two years ago.

The implications run in both directions. AI tools are also available for complaint response - drafting, tone-checking, ensuring the clinical narrative is coherent and complete. But the more important implication is preventive: if your clinical notes cannot withstand AI-assisted scrutiny, they are insufficient for the standard now expected. The periodontal chart that records numbers without clinical interpretation, the letter that omits the discussion of prognosis, the consent record that doesn’t document what the patient was told about maintenance - these were always inadequate. Now, there is a tool in patients’ hands that can identify the gap and articulate it precisely.

What AI Cannot Do

It cannot examine the patient. It cannot perceive the anxiety in the room, the inconsistency between what the patient says and what the tissues show, the clinical instinct that something is being underreported. It cannot exercise judgement in the face of genuine complexity - the case where the evidence points one way and experience points another, where the right decision requires weighing factors that no dataset has ever captured. And it cannot take responsibility. Every AI output in clinical practice is, ultimately, endorsed or rejected by a clinician. The liability does not transfer. The professional judgement does not outsource.

The periodontist who understands AI’s limits is not being left behind by the technology. They are the most valuable person in the room - because they know what the algorithm is good for, what it misses, and how to use it without being used by it.

AI Is Here to Remove Inefficiency. Not You.

The most useful reframe is this: AI in perio is not here to replace clinical expertise. It is here to eliminate the friction that surrounds it. The fifteen minutes spent writing a letter that could be reviewed in three minutes. The bone loss on a radiograph that a tired eye might underweight at the end of a long day. The complaint response drafted and redrafted under stress when a structured first draft would have taken the pressure off entirely. These are not clinical tasks - they are administrative and cognitive burdens that sit around the clinical task, dulling it. Strip those away and what remains is more time, more clarity and more capacity for the work that only a clinician can do. The consultation. The judgement. The relationship. AI doesn’t replace the periodontist. It clears the path for one!

The pattern repeats with every significant technological shift: the anxiety that the tool will replace the person, followed by the realisation that what it actually does is raise the baseline. Documentation standards go up. Detection sensitivity goes up. Patient expectations go up. Those who adapt find the technology amplifies their clinical judgement. Those who don’t find themselves held to a standard they didn’t know had changed. AI in periodontal practice is not the end of clinical expertise. It is a new demand on it. The question isn’t whether to engage. It’s how quickly you choose to.