Which side of that line a tool lands on is decided by whoever built and deployed it, not by what the user asked. Consumers appear to price protection off the question; the rulebook prices it off the product [9]. Lucy Castledine, the FCA's director of consumer investments, frames the remedy as using your own judgement and checking sources [10], which addresses accuracy and leaves redress exactly where it was.
The survey's own answers show how little accuracy awareness buys. 86% said they understood the need to check the sources an AI cites [8], while 38% said it is acceptable to make an investment decision on AI output alone [5], which puts a floor of 24% on the share of respondents who agree with both propositions [13]. Knowing the model can be wrong is a different piece of knowledge from knowing nobody underwrites the error.
There is a second gap inside the misconception itself. 44% believe AI-generated financial information is regulated, but only 32% believe the FSCS or the Financial Ombudsman would pay out if AI advice went wrong [4][6], a spread of 12 points [17]. That middle group has half the picture: they think something supervises the output, and they do not expect money back. For a broker or app with an assistant on the roadmap, that is the commercial shape of the problem. Users who assume coverage will behave as though it exists, and a firm that tunes its assistant far enough toward advice buys the opposite outcome, which is regulated advice with regulated liability attached [9].
The evidence is one day of fieldwork: 666 UK adults aged 18 to 40, surveyed through Attest on 24 July 2026, all of whom own investments or would consider buying within twelve months [11]. At that sample size a reading near 50% carries roughly plus or minus 3.8 points at 95% confidence [16], so AI's placement above TV and radio at 47% and press at 46% is inside the noise [2]. The influencer comparison is not: 56% against 29% is close to two to one [14]. And 44% of 666 is about 293 people who told their regulator, on the record, that they think a chatbot's financial output is supervised [15]. Whether 73% of them also know the same output can be inaccurate [7] is the part the FCA's five tips cannot fix, because tips are aimed at the wrong belief.
