This case study examines how AI tools respond to regulatory questions relevant to Marketing & Communications teams working at Retail Banking firms in the United Kingdom. It draws on findings across one regulation — Consumer Duty (PS22/9 + PRIN 2A), issued by the Financial Conduct Authority — covering one aggregated question where AI tools produced materially incorrect answers. The errors identified are not minor inaccuracies: in each case the AI gave a confident, detailed response that directly contradicts what the regulator's published text actually says.
For a Marketing & Communications team that relies on AI tools to navigate Consumer Duty obligations, the gap between what the AI says and what the rules require can translate directly into compliance failures at firm level.
Marketing & Communications teams at Retail Banking firms routinely encounter Consumer Duty questions in the normal course of their work. When refreshing customer-facing communications — product brochures, digital journeys, rate change letters, fee disclosures — the team must satisfy itself that the content meets the good consumer understanding outcome under PRIN 2A. When launching a new product or repricing an existing one, Marketing leads are often asked to confirm or document how communications have been designed and, critically, whether any testing is required before go-live.
AI tools are increasingly used at this scoping stage: a team member asks the AI what the rules require, uses that answer to build an internal checklist or a briefing note, and work proceeds from there.
The corporate use-cases sitting on top of these questions are consequential. Internal governance documents — Consumer Duty implementation plans, communications testing frameworks, sign-off matrices — are shaped by how the team understands the rules. Training content delivered to junior marketers, compliance attestations submitted to the Board or to the Risk function, and supplier briefs issued to creative agencies all carry forward whatever regulatory interpretation the team started with. If that interpretation was generated by an AI tool and was wrong, the error propagates into every downstream work-product before anyone with regulatory expertise has reviewed it.
When the AI's answer is wrong, the firm bears the consequences. The Financial Conduct Authority has broad supervisory and enforcement powers under Consumer Duty: it can require remediation, issue public censures, impose financial penalties, and in serious cases pursue senior individuals through the Senior Managers and Certification Regime. If a firm's communications were not tested because the Marketing team was told by an AI tool that testing was a binding rule only in specific narrow circumstances — when in fact the obligation is structured differently — the FCA's review of that firm's Consumer Duty implementation may find a systematic gap.
The reputational cost of an FCA finding on consumer communications can exceed the direct financial penalty, and the operational cost of retrospective remediation across a large retail book can be substantial.
The finding in this case study reflects a pattern that Marketing & Communications teams should treat as structurally significant: AI tools misrepresent the distinction between binding rules and non-binding guidance. In the Consumer Duty context, the AI invented a specific rule reference — attaching a rule number and marking it as binding — and simultaneously elevated guidance that explicitly uses the word "should" into something described as "effectively mandatory". These are not vague paraphrasing errors.
The AI cited real documents at real URLs, applied real FCA Handbook formatting conventions, and produced an answer that reads, on its surface, like competent regulatory analysis. A non-specialist reader has no obvious signal that anything is wrong.
All of the errors in this case study cluster on a single regulator and a single regulation: the Financial Conduct Authority's Consumer Duty framework. This concentration matters for Retail Banking firms. Consumer Duty is not a peripheral obligation — it sits at the centre of how the FCA supervises retail conduct, and the Marketing & Communications function is one of the firm's primary points of delivery for the good consumer understanding outcome.
A team that uses AI tools to answer Consumer Duty questions is operating in precisely the area where, on the evidence here, AI tools are most likely to give a wrong and convincing answer.
The systemic risk compounds quickly. If a single AI-generated answer about communications testing enters a firm's Consumer Duty implementation framework, it shapes every communications review that framework governs — potentially covering hundreds of customer-facing documents across the retail book. The cost of unwinding that error once identified, verifying compliance across affected materials, and reporting the gap to the Risk or Compliance function is not trivial.
For larger Retail Banking firms with complex product ranges and high volumes of customer communications, the operational remediation cost of a single incorrect AI answer embedded early in a governance framework can run into significant resource and time commitments, entirely aside from any regulatory consequence.
1 finding in this case study. Click any to see its full evidence card.
The starting position for any Marketing & Communications team in Retail Banking should be that AI tools are a research prompt, not a regulatory source. An AI answer that cites FCA Handbook rule numbers, uses correct Handbook notation, and references real guidance documents can still be materially wrong — as the finding in this case study demonstrates.
The team should treat any AI-generated answer about Consumer Duty obligations the same way it would treat an informal verbal briefing from a colleague: useful as a pointer to where to look, but never sufficient on its own to inform a compliance decision or a governance document.
Practical firm-level safeguards should be built into the team's workflow rather than left to individual judgement. A regulatory-verification policy should explicitly name AI tools as unreliable primary sources for Consumer Duty rule questions, and should require that any AI output which influences a firm work-product — a briefing note, a framework document, a training slide, a supplier brief — is verified against the regulator's published text before use.
Audit trails matter: if a piece of work was informed at any point by an AI-generated regulatory interpretation, that should be recorded so the firm can trace and correct the error if it later turns out to be wrong. Sign-off requirements before AI output enters firm-wide use are not bureaucratic overhead; they are the control that prevents a single wrong answer from propagating across an entire communications programme.
There are parts of the Marketing & Communications workflow where AI tools add genuine value and carry low regulatory risk. Drafting non-regulatory copy — creative briefings, tone-of-voice guidelines, internal announcements — does not depend on the AI being correct about specific rule obligations. Summarising long regulatory documents that the team then reads and verifies is a legitimate time-saving use, provided the team treats the summary as a reading guide rather than a substitute for the source.
Generating first-draft questions for a compliance workshop, or building a list of topics to raise with the Legal or Compliance function, are also appropriate uses. The boundary to maintain is between AI as a tool for organising and accelerating work, and AI as a source of authoritative regulatory interpretation — the latter role it cannot reliably fill.
RegLeg's published Hallucination Research gives Marketing & Communications teams a free, accessible reference point before relying on any AI answer in areas covered by our research. If your team is about to use an AI-generated summary of Consumer Duty communications obligations to inform a governance document or a training programme, the research lets you check quickly whether that rule area is one where AI tools have a documented pattern of error. It is not a substitute for reading the regulator's text — but it is a practical signal that can prompt the right level of scrutiny before work proceeds.
For firms that want to go further, RegLeg offers bespoke regulator deep-dives tailored to Retail Banking workflows. These map which AI-supported tasks in the Marketing & Communications function carry the highest hallucination exposure — distinguishing, for example, between tasks where AI error is low-stakes and easily caught, and tasks where an incorrect AI answer about a binding rule could shape a firm's Consumer Duty framework before anyone with specialist knowledge reviews it.
This kind of mapping gives the team and its leadership a clear view of where controls need to be strongest, without requiring the entire team to become regulatory specialists.
We also offer a confidential review of your firm's existing AI-use policy against RegLeg's catalogue of documented failure modes, with prioritised recommendations for where the policy needs to be tightened or where new verification steps should be added. Alongside this, RegLeg produces training material and CPD-aligned content that the Marketing & Communications team can use internally — helping the team develop a working understanding of how AI errors present in regulatory contexts, what the warning signs look like, and how to build habits of verification into day-to-day work without slowing down the team's output.
These are practical, team-level tools, not abstract compliance frameworks, and they are designed to be used by people whose primary expertise is communications, not regulation.