FCA Financial Promotions and AI Content: A Pre-Publication Checklist

Before publishing AI-written financial promotions, an advice or mortgage firm should check that each one is fair, clear and not misleading, that every claim traces to a source, that the right approver has signed it off, and that the final text and approval are recorded. The FCA applies the same standard whoever, or whatever, wrote the words.

In practice that becomes a sign-off list attached to every piece of AI-assisted content: website copy, LinkedIn posts, email campaigns, video scripts and leaflets. The list below is a practical aid, not legal advice. Your compliance officer or consultant decides what the rules require for your permissions.

Follow me on Instagram@sagnikteaches

What the regulator has said about AI, in plain terms

The FCA's page on its approach to AI says it does not plan to introduce extra regulation for AI and will rely on existing frameworks, with a principles-based, outcomes-focused approach. Its 2024 AI Update set out how current rules already apply, and the page stresses accountability for senior managers, naming the Senior Managers and Certification Regime and the Consumer Duty among the existing rules that matter for AI. Its financial promotions pages say all promotions must be clear, fair and not misleading regardless of media type, so that consumers can make informed decisions, and its finalised social media guidance (FG24/1) says the rules are technology neutral and apply to posts as much as to print.

Connect on LinkedInSagnik Bhattacharya

Put those together and the position for a small firm is blunt: there is no "AI exemption" and no "AI rulebook" to read. The firm that publishes the promotion owns it. If a language model writes a sentence implying guaranteed returns, the firm made that statement. The rest of this checklist is about making that ownership visible and defensible. For the bigger governance picture, read the AI governance checklist for small advice firms.

Subscribe on YouTube@sagnikteaches

Is it a financial promotion at all?

Do this first, because it decides the route. A financial promotion is a communication that invites or induces someone to engage in a regulated activity, such as taking out a mortgage, buying an investment or entering an advice relationship. A generic educational post about how budgeting works may not be one. A post saying "switch your pension to us for better returns" is.

AI makes the line easier to cross by accident. Ask a model for an "engaging post about remortgaging" and it may add a call to action, a rate claim or "speak to us today". Each addition can turn education into a promotion. The first checklist question is therefore not about the AI at all: does the final text invite or induce? If you are unsure, treat it as a promotion.

The pre-publication checklist

Work through these in order. A "no" or "unsure" at any step stops publication until it is resolved and recorded.

Step 1: Origin and scope

• Is the draft logged as AI-assisted, with the tool, the date and the prompt or brief used?
• Is the audience stated (retail clients, existing clients, professional contacts)? Fair, clear and not misleading is judged against who is likely to read it.
• Does the text promote a product, service or regulated activity? If yes, it enters the promotions route below.

Step 2: Fair, clear and not misleading

• Balance: if a benefit is mentioned, are the risks, costs and limits given similar prominence?
• Absolutes: remove "guaranteed", "risk-free", "best", "always" and "no fees" unless provably true and approved.
• Performance and rates: any figure needs a source, a date and the conditions it depends on.
• Tone: no urgency pressure ("rates will rise tomorrow") and no exploitation of worry.
• Plain language: could the intended reader explain the message back after one read, well enough to make an informed decision? That is the purpose the FCA gives for the standard.
• Omissions: what would a reader need to know that the draft leaves out?

Step 3: Facts and sources

• Is every number, date, product name and legal reference matched to a primary source you have opened, not to the AI's memory?
• Are there claims about the firm (qualifications, awards, "whole of market", independence) that match your permissions and how you actually operate?
• Are there any invented testimonials, statistics or "studies show" phrases? Delete them.

Step 4: Identity and permissions

• Does the promotion identify the firm correctly, with the regulatory status wording your compliance policy requires?
• If the firm is an appointed representative or network member, does the principal need to approve it first?
• For mortgage content, has someone checked it against the mortgage conduct rules that apply to your permissions?

Step 5: Approval and record

• Has a named, competent approver signed off the exact final text, not an earlier draft?
• Are the final version, approver, date, channel and prompt stored together?
• Is a review or withdrawal date set, so stale rates or rules do not stay live?

Worked example: a remortgage email campaign

Say a five-adviser mortgage and protection firm asks an AI tool for an email to existing clients whose fixed rates end within six months. The brief: friendly, short, prompts a call. The first draft opens "Don't let your rate jump: switch now and save thousands", says "we search the whole market for you" and ends "rates are rising fast, act today".

Run it through the checklist. "Save thousands" is an unsupported claim and implies a result no adviser can promise. "Whole market" may not match the firm's permissions or panel; the compliance officer confirms the firm is not whole of market, so the phrase goes. "Rates are rising fast, act today" is urgency pressure and an unsupported prediction. "Switch now" assumes switching is right, when the client may be better on a product transfer or by doing nothing. The edited version: "Your fixed rate ends in the next six months. We can review your options with you, including whether switching or staying put makes sense. Early repayment charges and fees may apply. Book a call." The approver checks the risk wording, the firm's status line and that the audience is existing clients, then signs it off. The record holds the AI draft, the edited text, the approver and the date.

Prompts that reduce the risk before the human check

A good prompt does not replace the checklist, but it lowers the number of edits. Put your firm's rules in the brief, and save it in a ChatGPT or Claude Project, or a Gemini Gem, so every drafter uses the same one. Google is moving Gems to skills: personal accounts in November 2026, Workspace business accounts in March 2027.

You are drafting marketing copy for a regulated financial advice firm. The text will be checked by a compliance officer before it is used.
Rules:
- Do not state or imply guaranteed returns, savings, approvals or outcomes.
- Do not use: guaranteed, risk-free, best, cheapest, always, never, whole of market, no fees.
- Do not make predictions about rates, markets or house prices.
- Do not invent statistics, testimonials, awards or qualifications.
- If you mention a benefit, mention a relevant risk, cost or limit in the next sentence.
- Where a fact or figure is needed, insert [SOURCE NEEDED: description] instead of guessing.
- End with a plain invitation to talk, no urgency language.
Audience: [existing clients / first-time buyers / ...]. Topic: [topic]. Length: [words].

The "[SOURCE NEEDED]" instruction is the single most useful line. It converts the model's habit of filling gaps with plausible numbers into a visible to-do list. Count those markers in the output: every one must be resolved from a primary source, or the sentence removed.

A sample output and what to fix

For a social post on "why review your protection cover", the model might return: "Life changes fast. A new baby or a bigger mortgage means your cover may no longer fit. [SOURCE NEEDED: typical review triggers from firm's own guide] Talk to us about whether your protection still matches your needs. Costs and exclusions apply and cover is subject to underwriting." That is a decent first draft. The fixes: replace the marker with the firm's own list of review triggers, check that "costs and exclusions apply" is accurate for the products you arrange, and add your status line. Without the marker rule, the model would have invented a statistic on how many families are underinsured.

Second example: a protection landing page rewritten in three passes

A solo adviser asks an AI tool for a landing page offering a "protection health check". Pass one, the raw draft, includes: "Over a third of families have no cover at all" (invented statistic), "Our policies pay out when others won't" (unsupported comparison) and "Get covered in minutes" (implies certainty of acceptance). Pass two, after the Step 3 and Step 2 checks, deletes the statistic, replaces the comparison with a factual description of what a review covers, and changes the last line to "Book a review to see what options may be available, subject to underwriting". Pass three adds the status line, the risk and cost wording, and a note that the page will be reviewed in six months. The register entry records all three versions, because the rejected claims are evidence that the process works.

Channel notes: where the same sentence needs a different check

• Website pages: long-lived, so the review date matters most. A rate or rule reference that was true at launch is the usual problem a year later.
• Email campaigns: the audience is known, so check that the message suits those clients. A switch prompt sent to a client with a large early repayment charge is a risk the AI cannot see.
• LinkedIn and other social posts: character limits push writers to cut caveats. Put key risk wording in the post itself, not only behind a link.
• Video and audio scripts: spoken words are promotions too. Approve the script and the final cut, because the presenter may ad-lib.
• Replies to comments and enquiries: an AI-drafted reply to "is this a good time to remortgage?" can slide into advice. Route it to an adviser.

Before and after: three phrases through the checklist

Notice that none of the approved lines is longer than the draft. Compliance rewrites usually remove claims rather than add legal text.

Where AI drafting goes wrong in this sector

• Confident, specific numbers. Models produce rates, percentages and thresholds that look authoritative and may be stale or wrong. Treat any number as unverified.
• Dropped risk wording. When asked to shorten, models cut caveats first. Ask for "same length, caveats unchanged" or add the wording after editing.
• Advice by the back door. A promotion that tells a reader what to do ("you should remortgage") can drift into personal recommendation. Keep it general and invite a conversation.
• Tone drift on social media. Short, punchy copy rewards absolutes. Check threads and captions as carefully as long pages.
• Recycled content. The same prompt repeated across posts yields near-identical claims. One bad phrase can then appear in twenty places.

A realistic mistake and how it showed up

Imagine an adviser asks an AI tool to summarise a regulator's guidance into a client newsletter. The summary says a rule "requires firms to" do something that the guidance only says firms "should consider". The error survives because the adviser reads the newsletter for readability, not against the source. The fix is a Step 3 habit: for every statement about a rule, open the source and match the verb. "Must", "should" and "may" are not interchangeable.

Approval route: who signs what

The route depends on your permissions, whether the firm is directly authorised or an appointed representative, and the channel. Rather than guessing, write a one-page matrix and get your compliance contact to confirm it.

If you pay or brief a third party, such as an influencer, to post about the firm, read the FCA's social media guidance first. It warns that unauthorised people, influencers included, who promote a regulated financial product without approval from an authorised person may be committing a criminal offence. Ask your compliance contact how that applies to your arrangement before you commission anything, and keep the brief, the approved wording and the contract in the register.

Records that make the sign-off provable

If the regulator, your principal or a complaint ever asks "who approved this and on what basis?", you need an answer within minutes. A simple register does it, one row per promotion:

• Title, channel and publication date.
• Audience.
• Tool and version used, and where the prompt is stored.
• Source list for every fact and figure.
• Checklist outcome (all steps passed, or what was changed).
• Approver's name and date, and the link to the final approved text.
• Review or withdrawal date.
• Take-down record, if removed.

Keep the register for as long as your firm's record-keeping policy and permissions require; ask your compliance contact for the period, because it varies with the type of business and content. The tutorial on AI note takers for advisers shows the same record-first habit applied to client meetings.

A second look: use AI as a checker, never as the approver

Once the draft is final, you can ask a second AI session to review it against the checklist. This catches some absolutes and missing caveats cheaply. Paste the final text and your five-step list, and ask for a table of issues with the sentence quoted. Treat the result as a reviewer's note for the human, not a pass.

Review the text below as a sceptical compliance reviewer. Quote each sentence that (1) states or implies a guaranteed or likely outcome, (2) gives a figure without a source, (3) omits a relevant risk or cost, (4) sounds like personal advice, or (5) uses urgency. For each, explain the concern in one line and suggest neutral wording. If a category has no issues, say "none found". Do not say the text is compliant.
Text: [paste]

The last instruction matters: a model that says "this looks compliant" tempts a tired approver to stop. Keep the verdict with a person. The tutorial on whether advisers can use ChatGPT covers what compliance teams usually accept on tool use and client data.

What to do when a published promotion turns out to be wrong

Sooner or later a promotion will go live with a problem: a rate that changed, a claim nobody noticed or a post that reads differently in a feed than in the draft. Decide the response before it happens. Take the item down or amend it the same day, note the time in the register, and tell your compliance contact. Then work out how it passed: was the checklist skipped, did the approver sign an earlier draft, or did the AI add a line after approval? The answer usually points to one fix, such as locking the approved text in a read-only file and publishing only from that copy.

If clients may have acted on the wrong information, your compliance contact decides what further steps are needed. Do not ask the AI to write the correction notice without that decision, because the wording itself is a statement the firm must stand behind.

Training the people, not only the prompt

The checklist only works if the person drafting understands why each line exists. Run a thirty-minute session using three real past promotions: one clean, one with a missing risk statement and one with an unsupported claim. Ask each adviser to find the problems with the checklist, then compare notes. People who have found a "best rates" claim themselves stop writing them. Repeat it when your tool or prompt changes.

How to check the process is working

Every quarter, sample five published promotions from the register and test them. Can you find the approver, date and sources within two minutes? Does the live version match the approved text? Are any rate or rule references now out of date? Did any item skip the checklist because "it was only a post"? Any "no" is a process gap to fix before it becomes a finding.

Count how often the approver changes AI drafts. If almost every draft needs the same edit, put that rule into the prompt. If the approver changes nothing, check they are really reading it. For the wider conversation about where AI belongs in an advice firm, see the comparison of advice-specific AI and ChatGPT, and if you would rather build this with someone, my AI implementation consultation covers mapping the workflow and handing over a routine your team can run.

Questions firms ask about AI and financial promotions

Does the FCA ban AI-written promotions?

No. The FCA has said it does not plan extra rules for AI and relies on its existing frameworks. A promotion written by AI must meet the same standard as one written by a person: fair, clear and not misleading. Responsibility stays with the firm and its senior managers, not with the tool.

Do social media posts count as financial promotions?

They can. The FCA's guidance on social media says its standards apply whatever the medium. If a post invites or induces someone to engage in a regulated activity, treat it as a promotion and run it through the same sign-off, including posts drafted by AI.

Can the AI tool approve its own output?

No. Asking one model to check another model's wording can catch some slips, but it is not an approval. A competent, named person must sign off each promotion and the firm must be able to show who, when and against which checks.

Do I need to tell readers the text was written by AI?

The FCA standard is about whether the promotion is fair, clear and not misleading, not about how it was drafted. Check your own compliance policy and any platform rules. Do not claim a person wrote or personally reviewed something if they did not.

Further reads

• AI Readiness Checklist for Accountants, Solicitors, Consultants — Check the firm is ready for AI before drafting client-facing content.
• How to Stop AI Inventing Case Law: A Small-Firm Checking Routine — A checking routine for invented facts, useful beyond law.
• How to Automate Insurance Renewals With AI and Keep More Clients — Renewal messaging is another promotion-adjacent workflow.
• What Does AI Cost a Small Insurance Brokerage in Year One? — What AI costs a small brokerage once governance is included.
• How Small Law Firms Use AI to Draft Letters and Routine Documents — Another regulated firm's drafting-with-review routine.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the FCA's pages on financial promotions and adverts, FG24/1 finalised guidance on financial promotions on social media, and its AI approach page (last updated October 2026), all checked November 2026. Check the current Handbook rules for your permissions before relying on this list.