Most AI advice aimed at advisers is either breathless hype or vague caution. Here is something you can actually act on: a practical way to keep client data and AI-drafted documents inside Consumer Duty, SM&CR and UK GDPR.
Ask a working broker what they have read about artificial intelligence and you tend to get one of two answers. Either it will transform the business, ten times the output, automate the admin away. Or it is risky, be careful, have a word with compliance first. Neither tells you what to actually do on a Monday morning with a real client in front of you. This is an attempt to close that gap.
Start with the good news. The FCA has confirmed there is no separate AI rulebook for firms like yours. The frameworks you already work under do the job: Consumer Duty, the Senior Managers and Certification Regime, and UK GDPR. That matters more than it sounds. It means the question is not “am I allowed to use this”. The question is “can I evidence the same oversight, records and traceability I would for anything else that touches a client”. Answer that, and you are most of the way home.
From there it comes down to three lines. Cross any one of them and you have a problem. Stay inside all three and AI becomes one of the most useful things in the business.
Line one: mind what you put in, and which account you use
Client information is personal data under UK GDPR. A name. A date of birth. An income figure. A property address. All of it. The mistake brokers make is pasting that straight into a free, personal AI account without a second thought about where it then goes.
On the consumer tiers of the main assistants, the default settings often allow your conversations to be used to train the model, and to be retained for a period. On Claude, for example, that has been the default since late 2025 unless you switch the training setting off, at which point you move to a much shorter retention window and your data is not used for training. The other tools have their own equivalent. The principle is the same on all of them: find the setting, check it, do not assume.
For anything involving real client data, the cleaner answer is a business or team tier. These typically exclude your data from training and come with a data processing agreement you can actually show your network. Better still, build the habit of stripping out or anonymising client details before they go anywhere near the tool, whatever account you are on. Marketing copy, research and general content with no client data in them can sit happily on a standard account with the training setting switched off.
Line two: the advice is always yours
This is the SM&CR line, and it is the one that matters most. A named, accountable human signs off advice in your firm. That does not change because a model wrote the first draft.
AI does not assess suitability. It does not make a recommendation. It produces a draft, fast and in plain English, and then you review it, verify it and own every word that reaches the client. The useful way to picture it is a capable junior: quick, tireless, and never, under any circumstances, allowed to sign anything. If you would not let a trainee send it unchecked, do not let the AI send it unchecked either.
Line three: check the output, every time
General-purpose AI is confident, and it is sometimes wrong. It can invent a rate, misquote a lender’s criteria, or state a product detail that simply does not exist. That is not a reason to avoid it. It is a reason to verify it.
Everything factual gets checked against source before it goes near a client: the fact-find, the meeting notes, the illustration, the lender’s actual criteria on the day. A general assistant has no built-in compliance layer watching over your shoulder, unlike the purpose-built advice tools. So you build that checking step yourself, and you make it routine rather than an afterthought.
The suitability letter question
This one comes up more than any other, because it is the strongest use case and the highest-risk one at the same time. Drafting suitability letters with AI is legitimate, and it saves real time. The guardrails are simply the three lines above, applied with care. You own the assessment. The figures and recommendations trace back to the fact-find and the client’s circumstances. Nothing sits in that letter because the model decided it should. Done that way, you get a faster process and a clean audit trail. Done carelessly, you get a Consumer Duty problem with your name on it.
Three questions before you paste anything in
If you remember nothing else, remember these. Before any client-related task, ask:
None of this is complicated. It is the same discipline you already apply to everything else that carries your accountability. AI does not lower the bar on Consumer Duty or SM&CR. Handle the data correctly, keep a human owning every decision, and it simply helps you clear that bar faster.
Dan Reddish trains UK mortgage and protection brokers on using AI in their businesses, with a particular focus on doing it compliantly. He is co-founder of AI Wealth Partners.
