Outlook
Part of Finance content in 2027: four labels for every claim about the future
Finance creators AI applications: what reviewers check
AI drafting and retrieval for finance creators in England: the checks an editor still makes, the rules a draft must meet and when to review again.
A model can propose an outline or shorten a sentence, but it cannot decide whether wording is a financial promotion. Where finance creators use AI to draft or to retrieve sources, the reviewer's checks stay the same: compare every statement with a cited record and keep the approved version.
What to take away
- AI drafting and retrieval are editorial scenarios for finance creators, not evidence of current practice or supplier announcements.
- The reviewer compares every factual statement and product claim with the cited record before publication.
- A model does not decide whether wording is a financial promotion, whether a product suits a reader or which risk can be omitted.
- AI does not verify product claims, replace clear advertising disclosure or remove applicable financial-promotion requirements.
- A change of model, prompt, retrieval source or audience triggers a fresh review.
Where the human line sits
The line sits at judgement. A team might use a model to propose an outline, shorten a sentence or create format variants from an approved source pack, and the reviewer still compares every factual statement with the cited record and retains the approved version.
Human review before publishing AI drafts
Is every factual statement compared with the cited record?
Editor retains approved version
Keep draft out of publishing route
Do not let the system decide whether wording is a financial promotion, whether a product suits a reader or which risk can be omitted. Those are human decisions, taken against approved material and recorded with the version that was released.
Examples of tools a team might consider include ChatGPT, Gemini, Microsoft Copilot or Claude. None was tested here, and a tool name is not evidence that its output is accurate or suitable for a financial promotion. A provider release note is a supplier announcement, not evidence that errors have fallen.
What the reviewer checks on each drafting task
Each drafting task has one check the model cannot perform for itself. Pairing each task with a check and a human decision stops a fast tool becoming a fast route to an unchecked claim.
| Task given to a model | Check the reviewer makes | Decision that stays human |
|---|---|---|
| Propose an outline | Every heading maps to a statement in the approved source pack | What the post is allowed to promise |
| Shorten a sentence | Names, numbers and warnings still match the approved source | Whether a warning can be shortened at all |
| Create format variants | Each variant is compared with the approved version | Which variant is approved for each channel |
| Reformat an approved post with affiliate links | Product claims, risks and terms match approved material | Whether the post is a financial promotion |
For example, a personal-finance creator in a hypothetical paid partnership with an investment provider might use AI to reformat an approved post containing affiliate links. The creator still needs to check the product claims, risks and terms against approved material and make the commercial relationship clear. The ways finance creators get paid differ in regulatory exposure.
Retrieval from a closed source set
Retrieval is safer when the source set is closed. An internal tool could locate a passage in an approved regulator or first-party record and return the passage, URL and access date for human checking. It should not turn retrieval confidence into factual confidence.
- Fix the source set and record the access date for every page in it.
- Ask known-answer questions and compare each returned passage with the page.
- Include stale pages, conflicting dates and inaccessible documents in the set.
- Try deliberate prompt attacks and note what the tool returns.
- Keep any unverified output outside the publishing route.
For a creator checking a partner's product claim, retrieval can find a source but cannot establish that an offer is current or that a promotion complies with applicable rules. A person still reads the page and records what was read before the claim is used.
The FCA's AI Lab page describes AI Live Testing as a safe place for firms to test AI systems in real-world conditions, with regulatory support and oversight. It concerns selected financial-services systems. It does not report a universal accuracy rate or show that testing has validated tools for finance creators.
The rules a drafted post still has to meet
The FCA says its existing rules apply to AI and that it does not plan to introduce extra regulations for AI. A generated draft is therefore judged like any other communication. That regulatory position is not approval of generated content or creator partnerships.
Where content is a financial promotion, relevant FCA requirements include COBS 4, including the requirement for promotions to be fair, clear and not misleading. Check the relevant firm on the FCA Register. The Consumer Duty applies to firms within its scope; it is not a blanket creator licence.
Section 21 of the Financial Services and Markets Act restricts an unauthorised person from communicating an invitation or inducement to engage in investment activity, unless an exemption applies or an authorised person with the relevant permission communicates or approves it. For paid and affiliate partnerships, the commercial relationship must also be clear under the ASA/CAP Code.
When to run a fresh review
Run a fresh review when the model, prompt, retrieval source or intended audience changes. Each of those changes can alter what the tool returns, so an earlier approval does not carry over to the new set-up.
Retrieval needs its own triggers: a corpus update, a failed citation, a material regulatory publication or an announced FCA practice report. The quality checklist is the place to confirm the final format before release.
Statements about what AI might do for finance content in 2027 need their own labels. The 2027 outlook keeps observed evidence apart from editorial scenarios, so a draft does not present a scenario as a result.
Before you act
- Compare every factual statement and product claim with the cited record and approved partner material.
- Keep unverified output outside the publishing route.
- Make paid and affiliate relationships clear under the ASA/CAP Code.
- Where a financial promotion is involved, check the applicable COBS 4 and section 21 route and verify relevant firms on the FCA Register.
- Trigger a fresh review when the model, prompt, retrieval source or audience changes.
Common questions
Can a model decide whether a post is a financial promotion?
No. A model does not decide whether wording is a financial promotion, whether a product suits a reader or which risk can be omitted. Those decisions stay with a person and are recorded with the approved version.
What should a closed source set record for each page?
The passage, the URL and the access date, so a person can check the returned text against the page. Retrieval can find a source but cannot show that an offer is current.
What does the FCA say about AI rules?
The FCA says its existing rules apply to AI and that it does not plan to introduce extra regulations for AI. That position is not approval of generated content, and separate financial-promotion requirements may apply to a creator's partnership content.
