Britons Back AI Tools to Help GPs Spot Skin Cancer, Nationwide Survey Finds

Artificial intelligence is steadily moving from the research laboratory into the GP surgery, and one of its most promising applications lies in the early detection of skin cancer. A new UK-wide study has now asked the public directly how they feel about that prospect, using a rigorous survey technique known as a discrete choice experiment to map out the conditions under which people would accept AI support in primary care. The findings, published in the British Journal of Cancer, offer some of the most detailed evidence yet on patient preferences for algorithmic assistance in one of medicine’s most visible frontlines.

Skin cancer remains one of the most common cancers in the United Kingdom, and melanoma incidence has been rising for decades. Most patients first present to their general practitioner, whose task is to decide which suspicious lesions need urgent referral to a specialist and which can be safely monitored. That judgement is difficult: benign moles vastly outnumber malignant ones, and distinguishing between them by eye is a skill that varies considerably between clinicians. AI tools trained on tens of thousands of dermoscopic images have shown impressive accuracy in research settings, and several systems are now being trialled in NHS practices as decision-support aids that sit alongside, rather than replace, clinical judgement.

Yet technical performance is only part of the story. If patients distrust the technology, or if certain design choices put them off, even the most accurate algorithm will struggle to achieve its potential. This is the gap the new study set out to address. Rather than simply asking people whether they like the idea of AI, the researchers used a discrete choice experiment, an economic method that presents respondents with a series of hypothetical scenarios in which the attributes of a diagnostic service are varied systematically. By analysing the choices people make across many paired options, the method can quantify how much each attribute matters and how people trade one feature off against another.

The survey was designed to capture the dimensions most likely to shape real-world acceptance. Respondents considered factors such as where the assessment takes place, whether the AI acts as a first reader before a clinician reviews the image or as a second opinion after the doctor has made a judgement, the accuracy of the technology, the speed of the referral decision, and the degree of human oversight involved. Participants were recruited from across the UK to reflect a broad population sample, and their responses were modelled statistically to estimate the relative weight each attribute carried in shaping preferences.

The headline result is one that developers and health planners will welcome: on balance, people are receptive to AI being used to help detect skin cancer in general practice. The technology is not viewed as an unwelcome intrusion but as a potentially useful partner in a diagnostic process that many recognise as imperfect. Importantly, the strength and direction of that acceptance depended on how the system was deployed. Preferences were not uniform across all configurations of an AI-enabled service, which suggests that implementation choices, not just algorithm accuracy, will determine public trust.

One of the clearest signals in the data concerned the role of the clinician. Respondents placed substantial value on scenarios in which a human doctor remains central to the diagnostic pathway, with the AI serving as an aid rather than an autonomous decision-maker. This aligns with a recurring theme in the wider literature on medical AI: people generally support algorithms that support clinicians, but are far more hesitant about systems that appear to bypass human expertise altogether. For skin cancer detection, where a missed melanoma can be fatal, that desire for oversight is perhaps unsurprising.

Accuracy itself also mattered. Respondents preferred services in which the AI, or the combined human-AI process, was described as more reliable in distinguishing harmless lesions from those needing urgent attention. Trade-off analysis allowed the researchers to express these preferences quantitatively, revealing how much additional waiting time, for example, people would accept in exchange for a measurable gain in diagnostic accuracy. Such estimates are valuable because they convert abstract attitudes into concrete quantities that health service designers can weigh against cost and capacity constraints.

The study also found that preferences varied across different groups of respondents. Familiarity with AI, previous experience of skin checks, and demographic factors all influenced how people weighed the attributes of an AI-supported service. This heterogeneity matters for policy. A one-size-fits-all rollout of AI tools could encounter pockets of resistance, whereas a deployment strategy that communicates clearly what the technology does, who oversees it, and how errors are handled is more likely to build broad acceptance. The authors suggest that transparency about the assistive role of these systems should be a central feature of any NHS implementation.

The timing of this research is significant. Health systems around the world are under pressure from rising demand and workforce shortages, and diagnostic backlogs have become a persistent concern. AI triage tools for skin lesions promise to speed up the pathway, potentially allowing low-risk cases to be managed in primary care while ensuring that high-risk lesions reach specialists quickly. Early trials of such tools in the UK have reported encouraging results, but large-scale adoption will depend as much on public confidence as on clinical evidence. By quantifying what patients actually want from these services, this study provides an evidence base for designing AI deployments that people are willing to use.

The broader lesson extends well beyond dermatology. As algorithms enter ever more areas of medicine, from radiology to pathology to general practice triage, understanding the public’s terms of engagement becomes an essential part of responsible innovation. This UK-wide experiment demonstrates that acceptance is conditional, nuanced and measurable. People appear ready to embrace AI that makes skin cancer detection faster and more accurate, provided that clinicians stay firmly in the loop and that the promises made on behalf of the technology are honest ones. In that sense, the study offers not just a snapshot of current attitudes but a practical roadmap for introducing AI into primary care in a way that earns, rather than assumes, public trust.

The discrete choice experiment approach used in this study has a long pedigree in health services research, having been employed to elicit preferences for everything from screening programmes to vaccination schedules. Its strength lies in forcing respondents to make realistic trade-offs rather than simply endorsing or rejecting a technology in the abstract. When people are asked directly whether they support medical AI, many express generic enthusiasm or generic unease; when asked to choose between two concrete service designs, their underlying priorities become visible. This is particularly valuable for technologies at an early stage of deployment, where public attitudes are still forming and where policy decisions made now could lock in patterns of trust or distrust for years to come.

The context for this work is a diagnostic pathway under genuine strain. Melanoma, while accounting for a minority of skin cancer cases, is responsible for the large majority of skin cancer deaths, and its incidence in the UK has increased substantially over recent decades, partly attributed to historical trends in overseas sun exposure and an ageing population. At the same time, urgent suspected cancer referrals have grown faster than dermatology capacity in many parts of the country, creating waiting times that clinicians and patient groups have repeatedly flagged as concerning. Tools that can safely reduce the number of benign lesions progressing along the urgent pathway could free specialist time for those who need it most, which is precisely the promise that has attracted NHS innovation funding to this area.

It is worth noting how the regulatory landscape is evolving in parallel. Software intended to inform clinical decisions about suspected cancer falls within the scope of medical device regulation, and the United Kingdom has been developing its own post-Brexit framework for approving and monitoring such tools. Real-world evaluation is a central expectation of that framework, and studies of patient acceptability complement the technical validation studies that dominate the field. A system may pass accuracy benchmarks in retrospective image datasets yet still fail in practice if the public declines to engage with the service in which it is embedded. Evidence on preferences therefore feeds directly into implementation guidance and commissioning decisions.

The finding that prior familiarity with AI shapes preferences echoes a consistent pattern in the behavioural literature: experience tends to moderate both utopian and dystopian expectations. People who have encountered algorithmic tools in everyday life, whether in navigation apps or online services, often report more calibrated views of what such systems can and cannot do. In the clinical setting, this suggests that early, well-communicated deployments could themselves build the familiarity that supports later acceptance, whereas a poorly explained first encounter could colour attitudes across an entire community.

There are, of course, limits to what any stated-preference study can establish. Hypothetical scenarios do not carry the emotional weight of a real diagnosis, and respondents may behave differently when a genuine lesion of their own is at stake. Discrete choice experiments also require researchers to select which attributes to vary, and unmeasured concerns, such as data privacy or the fear of being deprioritised by an algorithm, may not be fully captured. Longitudinal follow-up of actual deployments, capturing both uptake and outcomes, will be needed to confirm that the preferences measured here translate into behaviour. Nevertheless, by quantifying the conditions of acceptance before widespread rollout, this study offers health planners a rare opportunity to design AI-enabled services around public expectations rather than retrofitting trust after the fact.

Subject of Research: Public preferences for AI-assisted skin cancer detection in UK primary care, measured using a nationwide discrete choice experiment.

Article Title: Preferences for the use of Artificial Intelligence (AI) technologies to help detect skin cancer in primary care settings: a UK-wide discrete choice experiment (DCE)

Article References: Jones, O. T., Walter, F. M., Matin, R. N., Calanzani, N., Emery, J., van der Schaar, M., & Morris, S. (2026). Preferences for the use of Artificial Intelligence (AI) technologies to help detect skin cancer in primary care settings: a UK-wide discrete choice experiment (DCE). British Journal of Cancer. https://doi.org/10.1038/s41416-026-03611-x

Image Credits: AI Generated

DOI: 10.1038/s41416-026-03611-x

Keywords: artificial intelligence, skin cancer, melanoma, primary care, general practice, discrete choice experiment, patient preferences, clinical decision support, dermatology, NHS, diagnostic accuracy, public attitudes