Yorkshire Patients Say Their GP's AI Receptionist Cannot Understand Their Accent

Key points

• Healthwatch Rotherham, the local health and social care watchdog, has flagged complaints from patients at multiple GP practices in the area.
• An AI receptionist called Emma has been introduced at a number of those practices to handle incoming patient calls.
• Patients say the system fails to understand their Yorkshire dialect, making it difficult or impossible to get through.
• No national regulator has opened a formal investigation; this remains a complaint at the local watchdog level.

For some patients in Rotherham, South Yorkshire, calling the doctor has become something stranger: being told, in effect, that the computer on the other end of the line can't understand the way they speak.

Healthwatch Rotherham says several GP practices have replaced human receptionists with an AI voice-recognition system called Emma. It struggles, the watchdog reports, with the town's broad Yorkshire dialect.

What actually goes wrong on the call?

Emma mishears patients or fails to respond correctly, leaving callers stuck. Voice-recognition AI works by matching spoken words against patterns learned during training. When local pronunciation or vocabulary sits outside those patterns, the system loses the thread.

That's not a minor glitch. For someone trying to report symptoms or book an urgent appointment, a failed call means real delay. Healthwatch Rotherham, first reported by The Guardian AI, described the situation as leaving patients frustrated.

Should patients be worried?

Not alarmed, but informed. If Emma can't understand you, you're not imagining it and you're not alone. Most GP practices retain a fallback: ask to speak to a human, or check your practice's website for that option.

The deeper concern is fairness. AI voice systems are trained on data, and when that data skews toward certain accents, speakers of other dialects pay the price. Older patients, people with speech differences and communities whose voices were never well-represented in training sets can all find themselves on the wrong side of a system that wasn't designed with them in mind. Our earlier story on how AI health tools reach patients noted that the regulatory frameworks meant to catch exactly these gaps are still catching up with the technology.

No central regulator has opened a formal investigation. This is a local watchdog raising a local alarm, but it points to a broader question about how GP practices assess AI tools before rolling them out.

What happens next?

Healthwatch Rotherham gathers patient experience and passes it to the NHS bodies responsible. Its report puts pressure on practices to review how Emma is configured and whether it serves the population it was sold to serve. Practices could reconfigure the system with regional speech samples, though that decision sits with the software supplier.

Patients who find Emma unresponsive should log their experience with Healthwatch Rotherham directly. That feedback is what gives a watchdog the evidence to push for change.

Common questions

Can AI voice systems be fixed to understand regional accents?

Yes, but it takes deliberate effort. Developers need to train the system on speech samples that include the target dialect. Without that, it defaults to the accent patterns that dominated its original training data.

Is this happening at GP surgeries elsewhere in the UK?

No public data covers the full picture yet. Rotherham is the first place a watchdog has formally flagged the problem, but AI receptionists are spreading across NHS primary care, and similar complaints may surface elsewhere.

What to watch: the software supplier's response matters most here. A watchdog report is useful pressure, but if the company that builds Emma doesn't commit to dialect-inclusive retraining, every new surgery that signs up inherits the same flaw.