You can already see AI in everyday NHS work in 2026 — from automated scan reads in hospitals to chatbots used by GP reception teams. Faster scan reads and triage bots are showing up in some services, and they're changing waits and GP access; this piece looks at what's different, which tools staff actually mention, rough costs and the risks to watch.
Quick reference
- Waiting lists: roughly 6–7 million patients on elective lists in England (2025–26).
- Top AI tools in use: Ultromics (cardiac echo), Kheiron (breast screening), Babylon/GP digital triage, Microsoft Copilot in trusts, Google/Anthropic large models for admin.
- Prices (typical): ChatGPT Plus c.£16/month; Gemini Advanced via Google One c.£15–£20/month; Microsoft 365 Copilot for business c.£28–£35/user/month; private GP consultation c.£50–£100.
- Regulation and oversight: MHRA for AI medical devices, ICO for data, NHS AI Lab for evaluation and deployment guidance.
Overview: the current state in 2026
AI appears in many NHS areas — diagnostics, back-office admin, GP triage and a few frontline treatments — though uptake varies by trust. Hospitals and trusts use tailored algorithms to read images, prioritise patients and forecast demand. GP practices use symptom-checkers and auto-triage to manage demand. Central government has pushed funding and pilots, while regulators have started to set clear rules for clinical AI.
The effects aren't uniform: some trusts report faster radiology and pathology reporting, while others say AI adds extra steps and raises real concerns about bias, liability and procurement costs. And waiting lists remain stubborn: around 6–7 million people are still on elective pathways in England, so AI has improved throughput in places but hasn’t magically solved capacity shortages.
Top picks and analysis
Below are AI products clinicians and private GPs frequently mention in 2026, along with how people say they use them and rough cost notes from users.
- Ultromics — echocardiography analysis. Widely used in cardiology pathways to flag reduced ejection fraction and valve disease. Trusts pay per-scan or subscription; typical hospital agreements run into low tens of thousands of pounds annually depending on volume, though larger deployments negotiate integrated contracts.
- Kheiron Medical — breast screening AI. Deployed in several screening programmes to reduce false positives and speed reporting. Pricing is usually per-case or per-seat for radiographers; individual trusts report contracts in the tens of thousands a year.
- Babylon/GP digital triage. Used by some practices and private clinics for symptom-checking and remote consultations. Consumer subscriptions for direct-to-patient access sit around £10–£15/month in 2026, while private GP consultations cost roughly £50–£100 per appointment depending on the clinic.
- Microsoft 365 Copilot / Azure OpenAI. Adopted by trusts for document summarisation, clinic letters and coding support. Enterprise pricing sits around £28–£35 per user per month for the Copilot licence; cloud hosting and extra compliance layers add to cost for NHS uses.
- Large language models (LLMs) — ChatGPT, Gemini, Claude. Some clinicians use consumer LLMs to draft letters, prepare patient leaflets and help think through options — but they don't treat them as sole decision-makers. ChatGPT Plus is about £16/month for personal use; Gemini Advanced via Google One is in the same ballpark (£15–£20/month). But clinical deployments require on-prem or enterprise agreements with extra safeguards.
Comparison table
| Use | Product | Typical cost (UK) | Strength | Limitations |
|---|---|---|---|---|
| Cardiac imaging | Ultromics | £10k–£50k/year (trust size) | Fast echo reads; triage | Integration, validation per device |
| Breast screening | Kheiron | £10k–£40k/year | Reduces recall rates | Local audit needed |
| GP triage | Babylon / practice SDKs | £10–£15/month (consumer) | Reduces admin load | Patient trust, accuracy edge cases |
| Admin & letters | Microsoft Copilot | £28–£35/user/month | Time savings for staff | Costly at scale; integration work |
| Drafting & summarising | ChatGPT / Gemini | £15–£20/month (consumer) | Rapid drafts, translation | Not approved as sole clinical decision-maker |
Industry impacts — diagnosis, waiting lists and GPs
In diagnostics, AI tends to work best on narrow, high-volume tasks — routine reads where patterns repeat. Radiology and pathology have seen measurable speed-ups; automated reads flag urgent cases and reduce reporting backlogs. For example, AI-assisted mammography and echo analysis are shortening reporting times and lowering false-positive rates in pilot programmes.
On waiting lists, AI can flag who needs faster care and who might safely wait, which helps clinics focus the limited slots they have. That saves clinician time and reduces wasted clinic slots. But AI can’t add theatre capacity or staff.
So while some trusts have reduced time-to-treatment for specific pathways, national elective backlogs remain in the millions because the underlying capacity gap is still the constraint.
GP services: Primary care has seen two big changes. First, digital triage and symptom-checkers reduce routine telephone triage and filter urgent work. Second, remote consultations enabled by video and AI-generated summaries have sped admin. But there are trade-offs. Patients report convenience gains, yet watchdogs and some campaigners warn about a growing private option for faster access — contributing to a two-tier feel where those who can pay skip the queue.
Expert views and debate
Clinicians who back AI point to clear gains in turnaround times for imaging and reduced report fatigue. They say AI frees specialists to focus on complex cases. But critics — including patient groups and some GPs — warn about overdependence, algorithmic bias and the risk of displacing staff decisions with opaque models.
Regulatory authorities are central to the debate. The Medicines and Healthcare products Regulatory Agency (MHRA) now treats many clinical AI tools as medical devices, requiring conformity assessment. The Information Commissioner’s Office (ICO) enforces data-protection rules when patient data fuels models. And the NHS AI Lab provides central tool repositories and evaluation frameworks so trusts can compare performance under real-world conditions.
That said, healthwatch England has flagged the rise in private healthcare use as a result of waiting lists, noting a jump in people paying for private treatment when NHS waits are long. That trend colours the argument about AI: cheaper, quicker tech may improve throughput, but where capacity is limited, private spending can widen inequalities.
Practical tips for trusts, GPs and patients
- Trusts: insist on third-party validation and real-world performance data before buying. Budget for integration and staff training — licence fees are only part of total cost.
- GP practices: use digital triage to cut admin, but keep clear escalation routes to clinicians. Monitor patient satisfaction — automated triage can save time, but not at the cost of access.
- Patients: know the difference between consumer chat services and clinically‑approved AI. Paid subscriptions (ChatGPT Plus, Gemini Advanced) help with admin or general health queries, but they don’t replace a clinician’s judgement. Private GP appointments typically cost £50–£100; weigh speed against cost.
- All users: check data handling. Any system using NHS patient data should meet MHRA and ICO expectations and have clear audit trails.
Privacy and safety
Patient data is the most sensitive currency in any AI deployment. The ICO enforces data‑protection rules and expects lawful bases for processing. For clinical tools that influence decisions, the MHRA requires device classification, CE/UKCA marking where applicable and post-market surveillance. So trusts must not deploy powerful LLMs for clinical decision-making without enterprise agreements and auditability.
Still, many trusts use cloud-based AI for admin tasks under controlled contracts. Those arrangements typically include data residency clauses, encryption in transit and at rest, and breach notification protocols. Clinicians must document when they rely on AI, and trusts should run regular bias and safety audits.
Related Articles
AI in the NHS in 2026 is real, useful and messy. It speeds image reads, trims some admin and offers better triage. But it hasn’t removed the need for beds, theatres and GPs — those are human and budgetary limits. Regulators are tightening rules; patients and staff are demanding transparency; and the trick for the next five years will be to make AI practical, safe and fair, not just clever. The NHS may get smarter, but the questions left are social and political as much as technological.
This article was created with AI assistance.
