AI-generated client scrutiny is becoming a familiar headache for professional advisers across the UK, and independent financial planner Dan Wiltshire has written candidly about both sides of that experience. Writing in Investment Guide, Wiltshire describes how he first encountered the technology’s limitations as a consumer, before finding himself on the receiving end of its outputs as a practitioner.
When AI advice meets the real world
During a recent home renovation, Wiltshire used AI to interrogate builders’ quotes and interpret building specifications. For someone who describes his communication with tradespeople as “distinctly Partridge-esque,” the tool offered a welcome sense of control. Yet the exercise carried a sting. On AI’s recommendation, he insisted on installing an external temperature sensor that turned out to be incompatible with his boiler, shutting the system down whenever the temperature rose above 20 degrees. That single mistake cost him the best part of a thousand pounds.
His experience is far from unusual. According to Realtor.com, 1 in 4 homeowners say they have used AI to verify or challenge a professional’s recommendation. The same research found that a tenth of homeowners reported AI gave them conflicting information compared to a technician, a gap that can prove costly when the conflicting advice is acted upon without proper verification.
Wiltshire draws a parallel with the medical profession. Doctors he knows describe managing patients who arrive having self-diagnosed using AI, armed with confident but flawed conclusions. The pattern is consistent: a tool that provides fluency without genuine understanding creates friction for the expert who must then unpick the damage.
AI-generated client scrutiny and the asymmetry of accountability
Having learned from the building-site experience, Wiltshire might have expected to feel more comfortable when AI-generated challenges arrived in his own professional inbox. He does not. He now regularly receives lengthy lists of queries distinguished by clean grammar and neatly organised subheadings, bearing all the hallmarks of a prompt-and-paste exercise. The questions, he says, often miss the point of what was originally communicated.
The dynamic is further complicated by what Wiltshire identifies as a fundamental asymmetry. A client can paste a 500-word excerpt into an email in seconds. The adviser, by contrast, must choose every word in response with care, knowing that response may itself be reviewed. Satisfying the large language model, rather than educating the client, increasingly shapes what a written reply must look like. That is a significant shift in the nature of client communication.
There is also a cost dimension that deserves attention from a portfolio-management perspective. AI is frequently presented as a tool that compresses costs and improves efficiency. Wiltshire’s point, well-made, is that those gains in one area can generate hidden overheads in another. Professional time spent working through AI-produced queries is not billed separately; it is absorbed. Over time, that changes the economics of advice delivery in ways that are not yet fully priced in by either firms or their clients.
Redefining expertise in a world of frictionless information
The deeper question Wiltshire raises is about what expertise actually constitutes in a world where any well-phrased prompt can produce a superficially authoritative answer. His conclusion is one that experienced investors might recognise from their own decisions: real expertise is not the recitation of facts but the exercise of judgement, the capacity to understand context, cut through noise and, above all, ask the right questions in the first place.
For clients engaging professional advisers regulated by the Financial Conduct Authority, that distinction matters. AI can surface information; it cannot yet replicate the contextual judgement that separates a generic answer from one calibrated to a specific financial situation. The risk of conflating the two, as the Realtor.com research on home improvement suggests, is measurable and real.
