Public backs NHS tech and AI, but safety checks come first

Attitudes to technology and AI in NHS care: what 10,000+ people told us in 2025

Published 4 March 2026

Key points

• Public sentiment on health tech stays positive: 55% say technology improves care quality; 13% say it makes it worse (up from 8% in 2024).
• AI support is rising but still more cautious: 38% of the public think AI will improve quality (33% in 2024); 19% think it will make quality worse.
• NHS staff are more supportive than the public: 80% back AI for patient care vs 54% of the public; 86% vs 66% for admin uses.
• People want human oversight. When asked to choose, the public prioritises checks and evidence over speed and economic benefits by margins of around 70/30.
• Support is uneven. Women, 16-24-year-olds, and those in socioeconomic groups D and E are consistently less positive.

The survey at a glance

From 30 July to 1 October 2025, 8,000 UK adults and 2,027 NHS staff completed an online survey, with a booster of 240 adults at risk of digital exclusion via telephone interviews. The public sample was representative by age, gender, ethnicity, region and socioeconomic group. Differences reported are statistically significant at the 95% level.

Sentiment on technology: positive overall, with a small lift in negative views

The public still sees value in health tech: 55% say it improves care quality, while 13% say it makes care worse, up from 8% over the past two years. Among staff, 60% say tech makes care better, but those saying it makes care worse rose to 19% (from 6%).

The signal: the benefits remain clear, but day-to-day friction with current tools appears to be creeping in.

AI: improving support, but still behind general tech

On AI, public optimism grew: 38% think it will improve quality (33% in 2024); 19% think it will worsen it (18% in 2024). Staff confidence is higher: 57% say AI will improve quality (49% in 2024); 10% say it will make it worse.

Worth noting: questions on "technology" asked about the present; AI questions asked about the future. People can be frustrated with today's tools and still hopeful about AI's potential.

NHS App: strong appetite for basics, caution on AI advice

In England, people want the NHS App to do more of the basics. Around three-quarters would use it to book hospital appointments (76%), choose a hospital for treatment (73%), and access procedure information (73%).

But AI-generated advice for non-urgent care is a line some won't cross yet: 49% would use it; 32% would not. In socioeconomic group E, slightly more would refuse (36%) than accept (35%).

What people back AI for: admin first, then clinical

Public support is stronger for admin uses (66% support; 24% oppose) than for patient care uses such as diagnosis and treatment (54% support; 33% oppose). The uptick from 2024 to 2025 among the public came mainly from admin use cases (61% to 66%).

Staff back both: 80% support AI in patient care and 86% support admin uses.

Safety first: human checks and clear rules

• Support drops when AI decisions are not checked by staff. The decline is steepest for clinical scenarios like diagnosis and treatment selection.
• Asked to pick between competing priorities, the public chooses:
  • Evidence over speed (72% vs 28%).
  • Strict rules over economic development (71% vs 29%).
  • Human checks over faster results (70% vs 30%).
  • Accuracy vs explainability is evenly split (50% vs 50%).
  • Faster advice vs always-from-a-human leans to speed (55% vs 45%).

Bottom line: human-in-the-loop and high evidence bars buy trust.

Knowledge is rising; preferences are steady

Public knowledge about how the NHS uses health data rose from 34% (some/great deal) in 2023 to 39% in 2025. Among staff, it jumped from 67% to 80%. Knowledge of tech use also climbed (public: 55% in 2025 vs 50% in 2023).

But preferences barely moved. Comfort with robot-assisted surgery sits a little over four in ten across all three years; views on self-monitoring at home and software-supported triage remain stable. Familiarity changes faster than values.

Who is more cautious

• Women: 51% say tech makes care better vs 59% of men.
• Young adults (16-24): 48% say tech makes care better vs 55% overall.
• Socioeconomic group E: 40% say tech makes care better vs 68% in group A. Only 22% think AI will improve care; 28% think it will worsen it.

What this means for NHS leaders and teams

• Make implementation your first-class citizen. The rise in "makes care worse" is a warning. Fund go-live support, workflow redesign, training, and local superusers. Budget time for optimisation after launch, not just delivery.
• Co-design NHS App features with real users. Prioritise high-demand basics. For AI advice, be transparent about limitations, show who checks what, and make escalation to a clinician effortless. Provide non-digital routes and assisted access.
• Adopt human-in-the-loop by default for clinical AI. Define what staff must check, how often, and how it's documented. Stand up audit trails, bias monitoring, and incident reporting. Align with evolving guidance such as the MHRA's work on software and AI as medical devices: SAIA MD change programme.
• Start with low-risk admin wins, measure the gains. Letters, coding support, rotas, and queue management can free capacity fast. Track time saved, error rates, and staff satisfaction-and watch for unintended impacts on equity.
• Close the trust gap where it's widest. Engage women, 16-24-year-olds, and groups D/E through targeted outreach, community partners, and plain-language materials. Offer digital inclusion support, device access, and telephone fallbacks.
• Set clear evidence thresholds. Predefine performance, safety, and fairness criteria for each use case. Run real-world evaluations (A/B or stepped-wedge), publish results, and retire tools that don't meet the bar.
• Support staff ownership. Involve clinicians and operational leads early, compensate time for testing, and route procurement through teams who live with the consequences. Feedback loops should be short and visible.

Where to build capability next

• For clinical and operational teams: AI for Healthcare
• For policy and governance teams: AI for Policy Makers

The takeaway

People like useful tech, expect proof for AI, and want a human in the loop-especially for diagnosis and treatment. Staff are ready to move faster than the public, but they need better tools and better rollouts. Do the basics well, earn trust with checks and evidence, and bring those on the fence into the design room.