Study finds AI chatbots directing clients to firms fined or shut down by SRA

What happened

A new study reported by Today's Conveyancer has found that AI chatbots are recommending UK law firms to consumers that have previously been fined, sanctioned or shut down by the Solicitors Regulation Authority (SRA). When asked for conveyancing or legal service recommendations, the chatbots surfaced firms with adverse regulatory histories without flagging any concerns.

The research points to a gap between how generative AI tools present themselves as helpful, authoritative assistants and the accuracy of the information they actually return when consumers ask for recommendations in a regulated sector such as legal services.

Why it matters

This is fundamentally an AI reliability story: it shows what happens when large language models are used for real-world referral decisions without verification against authoritative, up-to-date regulatory data. Chatbots are increasingly used as a first port of call for consumers seeking professional services, yet the underlying models appear to lack the live connection to regulatory registers that would let them flag a firm's compliance status.

For organisations building or deploying AI-powered recommendation or advisory tools, the finding is a reminder that plausible-sounding outputs are not the same as verified ones. In regulated industries — legal, financial, medical — a confidently wrong AI recommendation can cause real consumer harm and reputational exposure for both the AI provider and any business whose brand appears beside it.

The Renascence take

The deeper issue here is not that AI chatbots occasionally get facts wrong — it is that consumers tend to extend the same trust to a chatbot's answer that they would to a human expert's advice, without applying the scepticism they'd use on an unverified web search. That trust transfer is a behavioural blind spot, and it is exactly where harm concentrates in regulated sectors.

Most commentary on this story will focus on the AI's factual failure; the more useful lesson is about design incentives. Chatbots are built to be fluent and reassuring, not necessarily cautious, and users read fluency as credibility — a classic case of the ease-of-processing heuristic overriding due diligence. Any organisation offering AI-generated recommendations in a regulated space should treat "don't know" as a legitimate, even preferable, answer, and should hard-wire live checks against authoritative registers rather than relying on training data that can be stale or incomplete. For professional-services brands, this is also a wake-up call to monitor how AI tools represent them to prospective clients, since a chatbot's error can shape a customer's first impression before any human conversation even begins.