A confident answer can feel like regulated advice. That does not make it so.
On 27 August 2026, the Financial Conduct Authority published research exposing misconceptions about AI and investment protection. Among respondents, 44% incorrectly regarded AI-generated financial information as regulated; 38% considered it acceptable to base an investment decision entirely on AI outputs; and 32% mistakenly expected FSCS or Financial Ombudsman compensation if AI advice went wrong.
The survey covered 666 UK adults aged 18–40 who held investments or would consider investing within the following year. It was conducted on 24 July. These figures describe that sample, rather than all UK consumers.
There is a revealing tension: 73% recognised that AI can produce inaccurate information. Awareness of possible errors therefore coexisted with mistaken expectations about protection. The FCA distinguishes general-purpose chatbots from tools specifically deployed to provide financial advice, which would likely fall within its remit. FCA announcement, 27 August 2026
What this means for customer communications
For compliance, operations and communications teams, the practical question is how to respond when a customer arrives with an apparently convincing explanation from somewhere else.
Consider a hypothetical customer who asks a chatbot to summarise an investment’s risks. The summary sounds reassuring but leaves out an important restriction. The customer then reads the provider’s email through that earlier interpretation.
The firm may have accurately described the restriction. Even so, its team may need to establish whether the customer has understood it and correct the misunderstanding during the interaction.
This scenario is an editorial illustration, not a finding from the survey. It highlights a useful review question: does the communication make the information needed for a decision easy to find, especially where customers may already hold incorrect assumptions?
The existing Consumer Duty connection
For firms and communications within scope, the Consumer Duty’s consumer understanding rules provide the relevant framework. PRIN 2A.5 requires communications to meet customers’ information needs, be likely to be understood and support effective, timely and properly informed decisions.
The rules also require firms to test communications where appropriate, monitor their impact and investigate and correct identified deficiencies. The approach takes account of the communication’s purpose and the customers receiving it. FCA Handbook, PRIN 2A.5
The August announcement reports research and gives consumer guidance. It should not be read as introducing a new requirement to use a particular AI approval system.
An operational response can start with the questions customers ask. Repeated confusion about guarantees, eligibility, protection or withdrawal rights can help a team identify which explanations deserve closer attention.
When AI helps write the firm’s own explanation
There is a separate issue when AI participates in preparing the firm’s communications.
An assistant can help a writer simplify a paragraph, organise an explanation or suggest alternative wording. Each suggestion still needs to be assessed against the source material and the purpose of the communication.
For example, imagine an assistant shortening a paragraph about compensation arrangements. The original explains that eligibility depends on specified conditions. The revision removes those conditions and retains only a reassuring statement about protection.
The revised sentence may be easier to read while conveying a materially different meaning. A review limited to grammar or tone would not address that change.
This hypothetical example illustrates why a useful approval process lets reviewers compare the proposed text with the original and inspect the supporting information.
Make human review a specific task
A practical review can ask four questions:
- Accuracy: Does the wording preserve the relevant facts, conditions and limitations?
- Meaning: Could a reader reasonably interpret the explanation as a guarantee or personal recommendation?
- Presentation: Are material qualifications visible where the customer needs them?
- Authority: Has the appropriate person approved this particular version for its intended use?
These are suggested operational checks, rather than a checklist prescribed by the FCA announcement.
Teams can apply them through arrangements proportionate to their work. A controlled document with recorded review decisions may suit a small volume of communications. Higher volumes, frequent revisions or multiple contributors may justify a more structured workflow.
Whichever approach is used, the reviewer needs enough context and authority to amend or reject the suggestion.
Keep the AI contribution reviewable
A useful record connects the relevant AI-assisted change with the human decision and the resulting communication version.
Depending on the use case, that can include what was proposed, what was accepted or amended, who reviewed it and which version was approved. Records should have appropriate access controls and retention arrangements, with unnecessary personal information excluded.
This makes later questions easier to investigate. If someone challenges a sentence, the team can examine its origin and the decision behind it.
CommsPliant applies this approach in its working platform. Its AI Enhance feature works on templates containing placeholders; it has no access to the generated, personalised customer content. Users can accept, amend or reject AI suggestions, and AI activity is recorded and available for review. The AI cannot approve or send communications. Version control and approval records support the team’s assessment of how wording progressed through the workflow.
Those records establish part of the process. Evidence that customers understood the communication still requires attention to their experience.
Check what happens after publication
A sensible next step is to select one important communication and follow it through the customer journey.
Check the source information, any AI-assisted revisions, the approved wording and the version used. Then examine relevant customer questions, complaints or comprehension feedback.
If a misunderstanding persists, consider whether the wording, layout or timing needs to change. Record the reason for the revision and assess its effect.
That gives teams a concrete way to connect responsible AI use with the purpose of customer communication: helping people understand the decision in front of them.