The Telegraph’s investigation into AI receptionists in GP surgeries painted a troubling picture. Patients hung up on, unable to get past basic identity checks. Elderly callers asked to use the NATO phonetic alphabet. A prescription error that could have cost a life. Feedback described by Healthwatch England as “wholly negative.”
These are real problems. But they are not inevitable ones. At Auxilis AI, we built Jackie to prove that AI receptionists can work for patients, for practices, and for the NHS.
Getting the basics right
If an AI receptionist can’t reliably catch a name or a postcode, nothing else matters. We selected a state-of-the-art speech recognition system optimised to avoid confidently guessing when it isn’t sure. Getting a name wrong isn’t just frustrating. It’s a safety risk: clinical data logged against the wrong patient record could lead to missed results or incorrect prescriptions. Jackie is designed to recognise uncertainty and escalate rather than guess.
Jackie is deployed across practices with diverse accents and patient populations, from Scouse in Liverpool to areas with large immigrant communities. And patients don’t need the phonetic alphabet. “S for Susan, M for Michael” works just as well as “Sierra, Mike.”
Fast, responsive, and relevant
Jackie is purpose-built for low latency and scales dynamically with demand, so response times stay consistent whether it’s one patient calling or a hundred at the 8am rush. Every workflow collects only the information necessary to safely process that specific request. Jackie adapts her questioning to the patient’s presentation: flu symptoms prompt different follow-ups than a urinary tract infection. No generic checklists, no unnecessary repetition.
No patient left behind
Jackie has successfully handled calls from elderly patients well into their nineties. For patients with learning disabilities, she recognises intent even with minimal input. For non-native English speakers, the feedback has been particularly encouraging: patients tell us Jackie is easier to understand than a human receptionist, and easier to explain their issue to. Where a fast-speaking receptionist with a local accent might be hard to follow, Jackie is clear, patient, and consistent.
We run extensive testing with varied patient populations before every deployment to make sure Jackie is as inclusive as possible in the real world, not just in the lab.
When AI should step aside
AI cannot replicate human empathy, and we don’t claim it can. What it can do is recognise when a human is needed.
Jackie detects urgent cases and clinical red flags, and when triggered, immediately transfers the patient to a human receptionist. Practices can also flag sensitive patients who should never interact with Jackie at all. Pharmacists, carers, or anyone else can request a transfer at any point with no questions asked. And Jackie never hangs up on a patient. If she cannot handle a request, she transfers the call.
Every call counts
Every call is recorded and stored, no exceptions. Jackie writes directly into clinical systems like EMIS Web and Accurx, and every call is also recorded in our own platform where our clinical team regularly reviews a sample for safety and accuracy.
Jackie verifies her understanding with the patient during every call, reading back what she’s captured before anything is logged. Nothing Jackie records leads to clinical action without being reviewed by practice staff first.
What patients actually say
We survey patients after every call with Jackie. The feedback isn’t universally perfect, and we wouldn’t claim it is. Some patients will always prefer speaking to a human, and that’s a legitimate preference we respect.
But the overall picture is encouraging. Patients consistently describe Jackie as polite, easy to talk to, and fast. Several have told us she reduced their anxiety around calling the surgery. For non-native English speakers in particular, the feedback has been that Jackie is clearer and easier to understand than a human receptionist.
We take every piece of negative feedback seriously. It feeds directly into how we improve Jackie. That loop between real patient experience and product development is how we make sure Jackie keeps getting better.
Built for the NHS, not bolted on
Jackie was purpose-built in the UK for NHS primary care, with DCB0129 and DCB0160 clinical safety certification, DTAC alignment, Cyber Essentials Plus, and NHS DSPT compliance.
The problems described in The Telegraph’s investigation are not inherent to AI in healthcare. They are the result of systems that were not designed with enough care. With Jackie, we’re proving AI receptionists don’t have to be like this, and we’ll keep working until that’s the standard, not the exception.