A client who is deciding and a client who has decided are two different meetings, and only one of them is a conversation. Research the FCA published on 27 August 2026 says that the second kind is becoming the normal kind, at least among younger investors, and that the thing doing the deciding is a chatbot. For an accountant that is not an abstract technology story. It is a change in what walks through the door: fewer questions, more conclusions, and a conclusion that the client believes came from something supervised.
The number worth stopping on is not the adoption figure. It is that 44% of those surveyed mistakenly believed AI-generated financial information is regulated. A client holding a plan they think is regulated behaves differently from one holding a tip they know is not — they argue for it, they have already told their spouse, and they are asking their accountant to help execute rather than to help think. That is the shift, and the profession that notices it first is the one that sees the bank statements.
What the FCA actually found
The regulator surveyed 18- to 40-year-olds who own investments or are considering them. Among the less experienced investors in that group, four in five had used AI for help with investing, and around two-thirds reported doing so occasionally or regularly. Trust levels put AI ahead of the established channels: 56% said they trust AI tools, against 47% for TV and radio, 46% for the press and 29% for social media influencers. Two-thirds expected to lean on AI more over the coming year.
Alongside that, the FCA recorded a set of misunderstandings about protection. Beyond the 44% who believed AI-generated financial information is regulated, more than one in three — 38% — believed it acceptable to make an investment decision based solely on the outputs of AI, and around a third, 32%, wrongly thought they would receive compensation from the Financial Services Compensation Scheme or the Financial Ombudsman Service if AI advice went wrong.
The picture is not uniformly bleak, and it would be misleading to present it that way. Almost three quarters, 73%, knew that AI can produce inaccurate information, and 86% understood the need to check the sources it references. Lucy Castledine, the FCA's director of consumer investments, framed the useful version plainly: "AI can help you research companies, understand jargon or explore options" before a decision is made. The regulator's caution was about protection and judgement, not about the tool.
The FCA also drew the regulatory line explicitly. General purpose AI chatbots are not regulated by it. A tool deployed specifically to provide financial advice would be likely to fall within its remit. Which side of that line a client's answer came from is knowable, and the client almost never knows it.
The misunderstanding is about redress, not accuracy
Most commentary about AI and money lands on accuracy — whether the output is right. That is the wrong worry to lead with in an accountancy practice, because the client already knows the answer might be wrong. Three quarters of them said so.
What they do not know is that being wrong carries no consequence for anyone but them. The Financial Ombudsman Service exists to handle complaints against firms it covers. The FSCS exists to pay out when an authorised firm cannot meet claims against it. Both are downstream of an authorised firm having been involved. Take the firm out of the chain and the redress goes too — not because the rules failed, but because the rules were never engaged in the first place. That is the practical content of the 32% figure, and it is a fact an accountant can state without straying anywhere near regulated advice.
The client is not usually wrong about what the answer said. They are wrong about what happens if the answer was wrong.
This is the same structural point as the one in the loan note warning, arriving through a different door. An unregulated product and an unregulated source of guidance fail in the same way: quietly, and outside every mechanism the client assumes is standing behind them.
Why it reaches the accountant first
Accountants see decisions before advisers do because they see money moving. A director's loan drawn down in March, a lump sum leaving a business account, a new brokerage name in the bank feed, a pension provider appearing in correspondence — these arrive in the practice as bookkeeping, and only later as planning. The client did not withhold the decision. They simply did not think it was a decision that needed anyone else.
There is a second reason, and it is more uncomfortable. The accountant is the professional the client already pays, already trusts and can already reach. When someone has produced a plan themselves and wants one professional to nod at it, the accountant is the cheapest available nod. That is a real risk, and it is not a risk about AI. It is the same risk described in the permission check: the moment when explaining the tax treatment of a step slides, in the client's hearing, into approving the step.
The mitigation is small and verbal. Answer the tax question. Then say which part of the decision the answer did not cover. One sentence, said deliberately, does the whole job — and it is the sentence that goes missing when the question arrives by email on a Friday afternoon.
A worked example
What follows is a composite built to show the shape of the problem. It is not a real client and not a comment on any firm or product.
A limited company client, 37, two years into contracting through her own company, emails her accountant in September. She has £64,000 sitting in the business account above what the company needs, has read that leaving it there is inefficient, and has a plan. She will take a £30,000 dividend this year, put £20,000 into a stocks and shares ISA and make a £20,000 employer pension contribution from the company. She wants the accountant to confirm the tax and process the dividend. She mentions, in passing, that she "checked it all properly first".
Several parts of that are the accountant's own work, and answering them is straightforward. The dividend has a tax cost that can be calculated exactly. The employer contribution has a corporation tax consequence and an annual allowance question attached to it. The £20,000 ISA subscription limit is a fact. None of that requires a regulated permission, and none of it should be withheld.
Three things in the same email are not the accountant's work at all: whether an ISA or a pension is the right home for the money given her circumstances, what the money should be invested in once inside either wrapper, and whether taking a dividend at all this year is preferable to leaving the funds in the company. Those are regulated planning decisions. The arithmetic does not settle them, which is precisely why the plan sounded so complete to her — the arithmetic is the part a chatbot does well.
The useful reply separates the two lists and adds the protection point once: the plan she checked was not produced by a regulated source, so nothing about it is covered by the Ombudsman or the FSCS. That reply takes ten minutes, contains no advice, and turns an execution request back into a decision she can still make properly.
What joined-up looks like from here
The version of this that works is not the accountant becoming sceptical about AI, or the client being told off for using it. The FCA's own tips point the other way — stay in the driving seat, verify the sources, understand that past performance does not predict returns. A client who has already read around a subject arrives better informed than the profession is used to, and a financial planner picking up that conversation starts several steps further along than usual.
What that requires is a handover that exists at all. In practice it looks like this: the accountant identifies which parts of the client's plan are regulated decisions, says so, and puts the client in front of someone qualified to make them; the planner documents suitability and takes responsibility for the recommendation; and the tax mechanics stay with the accountant, who is the only one holding the company accounts. Both professions keep their own work. Neither guesses at the other's. The mechanics of doing that cleanly are set out in the anatomy of an introduction, and the division of labour in accountant and adviser.
It is worth being honest about why this matters commercially as well as professionally. The lang cat's State of Advice Report 2025 found that just 9% of UK adults had paid for financial advice in the preceding two years, while 91% of those who did found it valuable — in a market of fewer than 5,000 IFA firms, with an average new client portfolio of £411,000 and an average client age of 59. A 37-year-old contractor with £64,000 of surplus company cash is not the client that market is built to attract, and she knows it. The space that leaves is the space the chatbot filled. That is the argument set out on the advice gap, and this research is it happening in real time.
Two things worth doing this week
Add one question to the year-end meeting. When a client presents a decision rather than a question, ask where it came from. Clients answer this readily and without embarrassment — nobody hides having used a chatbot. The answer changes what the practice says next, because a plan drawn from a general-purpose AI tool carries no redress behind it and a plan from a regulated adviser does. Recording the answer in the file takes one line.
Agree the practice's standing sentence. Not a policy document — a sentence, agreed between partners, that separates the tax answer from the decision it sits inside. Something in the shape of: this is how that would be taxed, and the question of whether to do it is a regulated one we do not answer. Firms that have settled this in advance say it calmly. Firms that have not end up improvising it at the worst possible moment, which is the moment a client has already moved the money.
What is not settled
Three things, and they are worth stating rather than papering over. First, the FCA's research covered 18- to 40-year-olds who own or are considering investments. It says nothing about how a 62-year-old business owner approaching a sale uses these tools, and the read-across to older clients is an assumption rather than a finding. Second, the FCA has not published rules aimed specifically at consumers using general-purpose AI for financial decisions; its position is the boundary statement described above, that general tools sit outside its perimeter while tools built to advise would likely sit inside it. Where a particular product falls is a question about that product.
Third, the interaction with targeted support is unresolved. A regulated regime designed to give people guidance short of full advice is arriving into a market where a large share of the intended audience has already found a free, instant, unregulated alternative and rates it above the press and television. Whether targeted support draws those people back inside the perimeter is a genuinely open question, and the evidence for it will take more than one year of take-up data to read.
What is settled is the part a practice can act on today. General-purpose AI output is not regulated. No Ombudsman complaint or FSCS claim follows from it. Clients are using it heavily, trust it more than most media, and a substantial share of them believe it carries protection it does not carry. Every one of those statements is published, dated and attributable — and together they change what an accountant should do with a client who arrives holding a plan.
Common questions
Is a general-purpose AI chatbot regulated by the FCA?
No. The FCA stated in its 27 August 2026 press release that general purpose AI tools are not regulated by it. Those tools respond to a wide variety of prompts and topics but are not set up to help consumers with financial advice, research or decision-making. The regulator drew a line alongside that statement: a tool deployed specifically to provide financial advice would be likely to fall within its remit. So the answer depends on what the tool was built to do rather than on how the person used it. Someone who obtains a detailed, confident, personalised-sounding financial plan from a general chatbot has not obtained anything the FCA supervises.
Does the Financial Ombudsman Service or the FSCS cover a decision someone made using AI?
Not where the output came from an unregulated general-purpose tool. The FCA’s research found that around a third of those surveyed, 32%, wrongly thought they would get compensation from the Financial Services Compensation Scheme or the Financial Ombudsman Service if AI advice went wrong. Both of those routes exist because a regulated firm sits behind the advice: the Ombudsman handles complaints against firms it covers, and the FSCS pays out when an authorised firm cannot meet claims against it. Remove the regulated firm and the redress route goes with it. This is a statement of fact about the compensation regime, not a view on any investment.
Can an accountant answer the tax part of a question the client has already researched with AI?
Yes, and that is the accountant’s own territory. The care needed is about sequence rather than subject. Answering the tax mechanics of a step the client has already resolved to take can read to the client as agreement with the step itself, particularly when the step is a regulated decision such as a pension transfer or a choice of investment wrapper. The practical fix is to answer the tax question and say plainly which part of the decision the answer does not cover. Confirming how a gain would be taxed is not confirming that realising the gain is sensible, and saying so out loud costs one sentence.
What should a practice say when a client arrives with a plan from a chatbot?
Three things, in order. First, ask where the plan came from, because the answer determines everything that follows and clients volunteer it readily. Second, state the protection position as fact: output from a general-purpose AI tool is not regulated, and no Ombudsman complaint or FSCS claim follows from it. Third, separate the parts. Some of what the client is holding is tax mechanics the accountant can address directly; some is a regulated decision that belongs with a qualified adviser. A practice that agrees this form of words in advance stops improvising it in the middle of a year-end meeting, which is where the trouble usually starts.
Does the FCA's research apply to older clients and business owners?
The published research does not say, and it should not be stretched. The FCA’s work focused on 18- to 40-year-olds who own or are considering investments, so the figures describe that group and no other. A practice can still act on it, because that age band covers a great many contractors, first-time company directors and younger shareholders. What a practice cannot do is assume the same proportions hold for a 62-year-old owner approaching a sale. The honest position is that the behaviour is documented in one age band and undocumented elsewhere, which is an argument for asking the question rather than for guessing at the answer.
Is a client using AI a reason to involve a financial planner sooner?
It changes the timing rather than the principle. The FCA’s own framing is that AI can help someone research and explore options before a decision, and used that way it can produce a better-informed client than the profession is used to. The difficulty is the client who has moved from exploring to concluding without anything regulated in between, because by the time an accountant sees it the decision often has momentum. Noticing that at the year-end meeting, rather than after the transfer, is the whole value of the observation. What follows is a conversation between two professions, not a verdict from either one alone.