CQC dental well-led: 10% failure rate. What AI adds in 2026

Your two sites run differently. Site one has a written policy for almost everything, because your practice manager there has been with you eight years. Site two, opened eighteen months ago, runs on verbal instructions and whatever the principal there decided made sense at the time. Now you are looking at a third site, and you have just added an AI scribe to speed up your treatment notes. It works well at both existing sites. Neither site has a hazard log, a named clinical safety officer, or a written record of what patients have been told about it.

CQC's own dental inspection data explain why this gap matters more than it might feel like it does. Of 1,336 dental practices inspected in 2017/18, 10% received requirement or enforcement action under the well-led key question, against just 2% under safe (CQC, Dental Mythbuster 27, last updated December 2018). Well-led is, by CQC's own figures, the area where dental practices are most likely to fail. Adding an ungoverned AI tool across inconsistent sites does not create a new risk category. It sits directly inside the one that already catches the most practices out.

Why well-led is the sharpest edge for AI governance

Well-led is one of five key questions CQC asks at every dental inspection, alongside safe, effective, caring and responsive. It is built around Regulation 17 (Good governance), which requires providers to have effective systems and processes to assess, monitor and improve the quality of care, and to maintain an accurate, complete record for each service user (CQC, Single Assessment Framework, well-led guidance, last updated February 2024). In plain terms: can you show that leadership, management and governance actually assure the care being delivered, not just that the care itself is fine.

An AI tool does not change what well-led is assessing. It adds a new thing to govern. A hazard log, a named clinical safety officer, a training record and a patient disclosure process are all governance artefacts in the same sense as your existing safeguarding policy or infection control audit. The difference is that most dental practices using AI tools today do not have them, because the tools arrived faster than the paperwork did.

So what for you: if well-led is already your weakest key question on paper, an AI tool used without governance is the fastest way to turn a soft finding into a documented one.

What actually changes in 2026

CQC is rebuilding its Single Assessment Framework from scratch. Two independent reviews found the existing framework opaque and poorly suited to individual service types. In its place, CQC has drafted four sector-specific frameworks, one of which, Primary Care and Community Services, covers dental practices alongside GP practices, pharmacies, optometry and community nursing (Agilio Software, dental compliance analysis of the CQC draft framework, 4 June 2026).

The structural change that matters most: the 34 "quality statements" used under the old framework are being replaced with 26 reintroduced Key Lines of Enquiry (KLOEs), mapped under the same five key questions dental practices already recognise. Under the old system, dental inspections were only ever assessed against 7 of the 34 quality statements, a gap that was never made explicit in published guidance. Whether the new KLOEs will be phased in for dental from day one, and how the framework's rating characteristics apply to a sector that receives no star rating, are both still open questions raised directly with CQC during the consultation, which closed on 12 June 2026 (Agilio Software, 4 June 2026). Piloting runs June to October 2026, with a final evaluation expected after that.

Dental practices remain non-rated throughout this. You will not receive an Outstanding, Good, Requires Improvement or Inadequate judgement. You will receive "regulations met" or "not all regulations met." That distinction matters for how you read everything that follows: CQC is not scoring your AI tool. It is checking whether the regulations, well-led chief among them, are being met around it.

So what for you: the framework is still being finalised, but the pilot window is live now. If your group is due an inspection between June and October 2026, these are not hypothetical future questions.

Where CQC's AI guidance actually comes from

CQC has not published dental-specific AI guidance. What exists is GP Mythbuster 109 (published July 2025), written for general practice, and a broader position paper, "Artificial intelligence in health and social care: CQC's role, expectations and plans" (CQC, 21 May 2026), which sets out principles for all registered providers: human oversight and continual monitoring of AI outputs, transparency so people can make informed choices, effective governance, staff sufficiently trained and confident in the tool, and Data Protection Impact Assessments to manage privacy risk. CQC is explicit that it is not building a separate AI-specific assessment process. It is clarifying how the existing fundamental standards, the same ones behind well-led, already apply.

Because the underlying regulations are shared across all registered provider types, the checks GP Mythbuster 109 describes translate directly into a dental well-led assessment. In practice, inspectors visiting a dental group using an AI tool are likely to ask about the same things: procurement evidence, a completed risk assessment and hazard log, a named clinical safety officer, human oversight of outputs, a route for learning from errors, data protection documentation, patient disclosure, staff training, equity of access for patients who decline the tool, and vendor assurance on bias.

The British Dental Association added its own advice page on AI use and governance in May 2026, including a downloadable AI Use and Governance Policy template for Expert-tier members (BDA, "Using artificial intelligence in dental practice," last updated May 2026). Its core message matches CQC's: AI is a useful adjunct, but accountability for anything that goes wrong sits with the dental professional, not the AI vendor.

Why this lands differently for a multi-site group

Single-site practices can often run governance informally and still pass, because one principal knows what is happening everywhere. That option disappears once you add a second site, and it disappears completely at a third. Well-led explicitly assesses whether leadership and governance work consistently across the whole organisation, not whether any individual site happens to be doing the right thing.

AI scribe adoption in UK dentistry is no longer a one-off trial. PortmanDentex, one of the largest dental groups in the UK and Ireland, signed a two-year partnership with Heidi Health to roll out AI clinical scribing across its practices (Digital Health News, April 2026). That is a vendor-announced partnership, so treat the scale claims with the usual caution, but the direction is clear: AI documentation tools are moving from single-site experiments to group-wide rollouts. A three-site expansion is exactly the point at which informal, site-by-site AI adoption becomes a well-led liability rather than a convenience.

So what for you: if you introduce the same AI scribe at your third site the way it was introduced at your second, without a shared governance framework across all three, you are replicating the same well-led gap CQC's own figures say is already your biggest exposure.

What this means for your practice

Four actions close most of the gap, and none of them require an external consultant to start:

First: nominate one named clinical safety officer for the whole group, not one per site. A senior clinician with current General Dental Council registration can hold this role across all locations, provided they complete digital clinical safety training and the hazard log is kept current for every site.

Second: write one risk assessment and hazard log per AI tool, applied consistently across every site that uses it. A tool introduced at site two under an informal trial still needs the same documented risk assessment as one introduced at site one under a formal process.

Third: confirm a signed Data Processing Agreement and a completed DPIA exist for each AI vendor, covering every site the tool touches. Patient dental records are Special Category data under UK GDPR regardless of which site generated them.

Fourth: standardise patient disclosure. A waiting room notice or a line in your patient information leaflet at one site and nothing at another is precisely the inconsistency well-led is built to catch. Make the wording identical across sites and brief every member of staff, not just the ones who were there when the tool was introduced.

If you want a structured view of where your group's governance actually sits before your next inspection, the AI Opportunity and Growth Assessment includes a well-led governance review as part of the CARE Framework, built for practices operating across more than one site. You can also book a 20-minute call to talk through where the gaps are likely to be before committing to anything further.

The single most important thing to take from this

CQC's own historic data say well-led is where dental practices already fail most often. The 2026 framework rewrite does not remove that risk. It restructures how it is assessed and folds AI-specific expectations directly into it. For a group adding a third site, the fix is not clinical and it is not expensive: name one clinical safety officer, write one hazard log per tool, confirm your data protection paperwork, and say the same thing to every patient at every site. Build that once, consistently, and the well-led question stops being the one you are least prepared for.

For the checks that apply to any registered provider using AI, not dental-specific ones, see the 10 AI governance checks CQC will look for in 2026. For the admin economics behind why dental groups are adopting AI recall and scheduling tools in the first place, see dental practices lose 18% of revenue to no-shows: what AI recall data shows.