The FCA’s good and poor practice paper on AI, published on 19 October 2026, makes a point that every advice firm using AI should take seriously. Having a human in the loop isn’t a control in itself. Human review only counts if the reviewer can genuinely challenge and change the output, and if your firm can show that this actually happens. For AI-drafted suitability reports, that means your paraplanners need 3 things: access to the source documents, the expertise and authority to override the draft, and a review record that shows what they checked and what they changed. A signature at the end of a report the reviewer clearly skimmed doesn’t prove much.
This article explains what the FCA said, why automation bias is a real risk in report writing, and how to build a review process you can actually evidence.
What the FCA actually said
The short answer: The FCA’s paper doesn’t introduce new rules. It uses examples from its supervisory work to show how firms should apply the existing framework to AI, including the Consumer Duty, SM&CR and the systems and controls rules. On human oversight, it warns about automation bias and review becoming a box-ticking exercise, and expects firms to be able to evidence meaningful review and challenge.
The paper, AI in financial services: good practice and areas for improvement, covers 3 broad areas: consumer outcomes, governance and accountability, and operational resilience. It includes boxes setting out what the examples mean for smaller firms, which makes it more practical than most regulatory publications.
A few messages stand out for advice firms.
Human review has to be real. The FCA’s concern isn’t that firms lack a reviewer. It’s that the reviewer may not have the information, expertise or authority to challenge the output, or that review gradually becomes a formality. The level of oversight should be proportionate to the risk of the decision.
Governance isn’t a one-off approval. Signing off a tool when you introduce it isn’t enough. The FCA expects firms to keep monitoring how AI performs in live use and what outcomes it produces.
Accountability sits with named people. Senior managers don’t need to be AI experts, but they do need to understand the AI risks in their area and take reasonable steps. Under SM&CR, that’s a named individual, not the software or the vendor.
Vendor changes matter. Good practice included contracts requiring advance notice of material changes to the underlying models, so firms aren’t caught out when a tool’s behaviour shifts.
The timing isn’t a coincidence. A week later, the joint Bank of England and FCA survey of AI in UK financial services found that 86% of responding firms now use AI, up from 75% in 2024. Using AI is now normal. The question the FCA is asking is whether firms are controlling it properly.
Why automation bias is a real risk in report writing
The short answer: Automation bias is the tendency to trust a confident, well-written output more than you should. AI-drafted suitability reports are especially prone to it because they read fluently and look finished. When a paraplanner is reviewing their 6th draft of the day, it’s easy to check that it reads well rather than whether it’s right.
The risk isn’t laziness. It’s how review works under time pressure. A badly written human draft invites scrutiny. A polished AI draft invites a quick read and a signature. The errors that matter tend to be quiet ones: a capacity for loss conclusion that doesn’t match the fact find, a disadvantage that’s generic rather than specific, a figure pulled from the wrong page of a provider document or a guarantee that was never mentioned.
Picture a paraplanner who approves 20 AI-drafted review letters in an afternoon with no edits to any of them. Maybe all 20 were perfect. But if the firm can’t show what was checked, a reviewer looking at that file has every reason to doubt it. That’s exactly the pattern the FCA is warning about.
Tick-box review vs genuine review
| Signal | Tick-box review | Genuine review |
| Source documents | Not opened | Checked against key figures and facts |
| Edits | None, or only cosmetic | Substantive changes where needed |
| Time spent | The same few minutes for every case | Varies with complexity |
| Record of review | A name and a date | What was checked and what was changed |
| Reviewer’s authority | Feels unable to reject the draft | Can override or send back |
| Escalation | No route for doubts | Clear route to adviser or compliance |
| Monitoring | Never looked at | Correction rates and patterns reviewed |
None of these signals proves anything on its own. Taken together, they show whether review is a control or a formality.
How to evidence genuine review of an AI-drafted report
Give reviewers what they need to challenge the draft
The FCA’s point about information is practical. A paraplanner can’t check a figure they can’t see. Every review should happen with the source material to hand: the fact find, meeting note, LOA responses, cashflow output and risk assessment. If your workflow makes the paraplanner hunt for the originals, the review will be thinner than it should be.
This is where the design of the tool matters. Planbot lets users point the AI at specific pages and sections of source documents, and its chatbot lets reviewers question the draft directly. A reviewer can ask “where did this transfer value come from?” and check the answer against the source.
Record what was checked, not just that it was checked
A review record should capture:
- Key Figures Checked Against Source
- Objectives Matched To The File
- Risk And Capacity For Loss Confirmed
- Disadvantages Specific To The Case
- Vulnerability Considered
- Changes Made And Why
- Reviewer Name And Date
That doesn’t need to be a long form. A short structured checklist built into the report workflow is enough, as long as it’s completed for each report rather than signed off in bulk.
Keep the edit trail
The most convincing evidence of genuine review is the difference between the AI draft and the final report. If your system keeps versions, you can show what the reviewer changed. If it doesn’t, the reviewer should note the main corrections. A file where the final report shows considered changes from the draft tells a much stronger story than one where the 2 are identical.
Make sure reviewers can say no
Authority matters as much as information. If a paraplanner believes the draft is wrong but feels they can’t hold up the case, the review isn’t working. Make it explicit that reviewers can reject a draft, send it back or escalate it to the adviser or compliance. Then make sure that route is actually used.
Watch the patterns, not just individual files
This is the monitoring the FCA expects, and it’s where small firms often fall short. Look at your review data every month or quarter. What proportion of drafts are corrected? What kinds of corrections come up again and again? Does any reviewer always approve without changes?
A correction rate of 0 across hundreds of reports isn’t a sign of a perfect tool. It’s a sign that review may have become automatic. On the other hand, the same correction appearing again and again usually means a template or prompt needs fixing. Fix it at the source rather than relying on reviewers to catch it every time.
Test the review now and again
Some firms check whether review is working by having compliance sample AI-drafted reports and recheck them independently against the file. If compliance regularly finds errors the reviewer missed, that’s useful information. It’s far better to find that gap internally than to have it found in a complaint.
Who is accountable when an AI-drafted report is wrong
The short answer: Your firm is, and so is the named senior manager responsible for that area under SM&CR. The paraplanner who reviewed the report and the adviser who gave the advice remain responsible for their roles. The software provider isn’t accountable to your client or the regulator, whatever your contract says about liability between you.
That’s the main reason the review record matters. If a report is challenged, your defence is the evidence that a competent person checked it properly. Our guide to FCA expectations on AI in financial advice covers how SM&CR and the Consumer Duty apply to AI-drafted documents in more detail.
What this means if you’re choosing or reviewing an AI tool
The FCA’s paper gives you a useful set of questions to put to any vendor of AI software for financial advisers:
- Can reviewers see the source behind every figure and statement?
- Does the tool keep a record of what the reviewer changed?
- Will the vendor tell you in advance about material changes to the underlying model?
- Does the tool work within your firm’s approved templates, or does it generate free text?
- Where is client data processed, and is it stored or used for training?
Template-based tools have an advantage here. When suitability report software works within your firm’s approved structure and wording, reviewers know exactly where to look and what should be there. That makes genuine review faster and easier than checking free-form AI text. An AI suitability report generator that pulls figures from specified source pages also makes the “check against source” step quicker to do and easier to evidence.
We designed Planbot around the idea that the paraplanner stays in control of every document. That’s the same principle the FCA is now spelling out. AI for paraplanners should take the drafting and formatting work away, so that the time saved goes into the checks that only a person can do. Specialist paraplanning software should make those checks easier, not tempt reviewers to skip them.
If you’re building the case for AI internally, it’s worth budgeting review time properly. Our guide to the business case for AI in your advice firm explains how to count the savings honestly, including the time genuine review still takes.
Frequently asked questions
Does the FCA have specific rules on AI?
No. The FCA’s 19 October 2026 paper confirms that it continues to apply its existing framework, including the Consumer Duty, SM&CR and the systems and controls rules, rather than introducing AI-specific rules. The paper gives illustrative examples of good practice and areas for improvement, which firms should expect to inform supervisory discussions.
What is automation bias?
Automation bias is the tendency to accept the output of an automated system without enough scrutiny, especially when it looks confident and polished. In advice firms, it shows up when reviewers approve AI-drafted reports because they read well, without checking them properly against the client file.
Who is responsible if an AI-drafted suitability report contains an error?
Your firm, and the relevant senior manager under SM&CR. The adviser and the paraplanner remain responsible for their part in the advice and the review. AI tools don’t transfer regulatory responsibility to the vendor, which is why firms need to be able to show that human review was genuine.
How can a small advice firm show that human review of AI is meaningful?
Keep the source documents with the draft, use a short structured review checklist for each report, keep a record of what was changed, and look at correction rates over time. The FCA’s paper includes notes for smaller firms, and the emphasis is on proportionate controls that actually work rather than on extra committees or paperwork.
Do advice firms need to tell clients that AI helped draft their report?
The FCA’s paper describes good practice where consumers are told when they’re using an AI-enabled tool. That mainly applies to tools clients interact with directly. For reports drafted with AI and then reviewed and approved by your team, check your privacy notice and terms of business with your compliance adviser, and be open if a client asks.
See how Planbot supports genuine review
If your firm uses AI to draft reports, or is planning to, the review step is where the FCA will look. Planbot works within your firm’s own templates, links drafts to the source documents and lets paraplanners question and edit every report before sign-off. No client data is stored or used to train AI models.
Book a demo and see how Planbot fits into a review process you can stand behind.


