Confidence Is the Scarce Input in Britain’s AI Roll-out
Ipsos’s latest Global Trends preview should give every business deploying customer-facing AI pause. It finds that 62% of British adults agree that technological progress is destroying our lives: the highest reading in the study’s UK trend since 1997. Only 35% say AI is having a positive impact on the world, while 49% disagree.
That does not mean customers are rejecting technology wholesale. People routinely use digital services because they are useful, convenient or unavoidable. But it does mean that businesses should stop treating adoption as a simple question of functionality. Public confidence is becoming a practical constraint on AI-led growth.
For brands, the commercial risk is not merely a consumer saying they dislike AI in a survey. It is a customer hesitating before accepting a recommendation, doubting a support answer, withholding data, escalating to a human or simply choosing a familiar alternative. The experience may work technically and still make the relationship weaker.
The decisive question is no longer whether AI can automate an interaction, but whether customers feel safe letting it do so.
Scepticism is broader than a chatbot problem
The significance of the Ipsos findings is their breadth. Technology anxiety is not confined to a small group of people who avoid digital services, nor is it just a reaction to one flawed product category. Ipsos reports that concern is widespread across age groups, declining only slightly among people with higher incomes and higher educational qualifications.
That matters because many business cases are built on an implicit assumption: once a new service is easy enough, resistance will recede. Sometimes it does. Contactless payments and online banking became normal through repeated use and visible utility. But AI presents a tougher proposition because it often combines unfamiliarity with opacity. A customer cannot always tell what information has shaped an answer, whether a recommendation is commercially influenced, or who takes responsibility when a decision is wrong.
Earlier Ipsos research points to the same underlying problem. A majority of Britons said they would trust a generative AI tool less if its answers were influenced by advertisers. Most also said they would not use AI output without checking it. This is not irrational caution. It is an understandable response to systems that can sound assured while making errors, and which frequently conceal their workings behind a conversational interface.
For customer-facing organisations, that distinction is critical. Convenience does not cancel the need for credibility. In some moments, particularly where money, eligibility, health, safety or a significant purchase is involved, credibility is the service.
Treat trust as part of the product specification
There is a temptation to address this mood with a broad reassurance campaign: explain that a business is innovative, responsible and using AI to improve service. Such messaging may be necessary, but it is insufficient. Trust is more likely to be formed in the detail of the interaction.
Can a customer easily see when AI is involved? Can they obtain a clear explanation without deciphering technical language? Can they correct a bad assumption, decline personalisation or reach a competent human being before the issue becomes expensive or distressing? Are the incentives behind a recommendation visible?
These are product and service-design choices, not a communications afterthought. They should sit alongside accuracy, latency and cost-to-serve in an AI programme’s success measures.
Ofcom’s work on online advertising offers a useful parallel. Its research has found that people do not always recognise why content or placements appear before them, even where the commercial mechanism is familiar. AI-assisted discovery and advice could compound that ambiguity if brands blur the lines between assistance, persuasion and paid influence.
This does not require businesses to make every interface feel defensive. It requires them to be proportionate. An AI tool that helps a customer find a replacement part needs an obvious route to verify compatibility. A virtual financial assistant needs clearer boundaries and escalation than a tool suggesting a playlist. The greater the consequence of a wrong answer, the stronger the proof and human recourse should be.
Research must move closer to the moment of choice
The new mood also raises the standard for insight teams. Measuring general sentiment towards AI remains useful, but it cannot tell a business whether a particular application earns confidence in a particular customer journey.
The better questions are behavioural and contextual. At what point do customers abandon an AI-led flow? Which explanations reduce uncertainty, and which sound like evasion? Does an AI label reassure customers by signalling honesty, or alarm them because it appears only when something goes wrong? Do customers accept automation for routine queries but resist it when they feel misunderstood?
Those questions demand mixed evidence. Qualitative research can identify the language people use when describing control, fairness and confidence. Journey analysis can expose where assistance becomes friction. Experimentation can test whether transparent choices improve completion, satisfaction and repeat use. Complaints, calls to human agents and repeat-contact rates should be treated as trust signals rather than mere operational leakage.
There is also a governance implication. Teams building AI experiences should be able to explain not only how a model performs in testing, but why customers are likely to accept its role. A technically valid answer is not always a commercially acceptable one.
Britain’s AI anxiety is not a reason to postpone every deployment. It is a warning against assuming that speed of implementation equals readiness for market. Businesses that make customer agency, evidence and accountability tangible will have a better chance of turning AI from a source of unease into a service people choose to use.
Comments