AI-Generated Content Compliance: What the FCA Expects for Financial Promotions in 2026

The FCA's position on AI-generated content is an extension of a principle that has existed since the Financial Services and Markets Act 2000: the authorised firm is responsible for the content and approval of every financial promotion it communicates or approves. Section 21 of FSMA doesn't contain the words "artificial intelligence" because it doesn't need to. The obligation falls on the firm, not on the tool used to create the promotion.

In practical terms, this means that using AI to generate marketing content does not introduce any new regulatory permissions or exemptions. It also doesn't introduce any new offences. The obligations are the same as they've always been: financial promotions must be clear, fair, and not misleading (COBS 4.2). They must be identifiable as promotions (COBS 4.3). They must contain appropriate risk warnings. They must be approved by an authorised person. And under Consumer Duty, they must support good customer outcomes and be designed for the target market.

What the FCA has added, through speeches by senior officials and Dear CEO letters during 2025 and early 2026, is explicit acknowledgement that AI tools are being used and specific expectations about governance. The three key expectations are: first, that firms must have processes in place to review and approve AI-generated content before publication, with the same rigour applied to any other financial promotion. Second, that firms must be able to demonstrate to the FCA, if asked, that AI-generated content was subject to appropriate human oversight. Third, that the use of AI does not reduce the firm's obligation to ensure factual accuracy — the FCA specifically noted that AI tools can generate plausible but incorrect information and that firms cannot rely on the tool's output as a substitute for fact-checking.

For firms operating under a network, the network's compliance team will have their own position on AI-generated content, and it's typically stricter than the FCA's baseline. Several major networks now require firms to declare whether AI tools were used in creating submitted marketing materials. Some networks have issued specific guidance on which types of content can be AI-assisted and which must be human-written. If you're under a network, check their current compliance manual before establishing your AI workflow — the rules may have been updated recently.

For directly authorised firms, the responsibility sits with your compliance officer or outsourced compliance consultant. The practical question is: does your compliance review process need to change because AI was involved in content creation? The answer is nuanced. If your existing compliance review process is thorough — if you check every factual claim, verify every statistic, ensure every risk warning is present and appropriate, and confirm the promotion is suitable for the target audience — then the process doesn't necessarily need to change. The content still goes through the same review. What changes is the reviewer's awareness that AI-generated content may contain specific types of errors that human-written content is less likely to contain.

AI language models generate text by predicting the most likely next token in a sequence. They don't retrieve facts from a verified database. This fundamental architecture creates specific risk categories that are particularly dangerous in financial promotions.

Hallucinated statistics are the most common and most dangerous risk. Ask an AI tool to write about pension transfer values and it may cite specific figures — "the average pension pot at age 55 is £127,000" — that sound authoritative but are fabricated. The model isn't lying; it's generating a plausible number based on patterns in its training data. But a financial promotion containing an invented statistic is misleading, and "the AI made it up" is not a defence the FCA will accept. Every numerical claim in AI-generated financial content must be independently verified against a primary source. This includes percentages, pound figures, dates, thresholds, allowances, and any other quantitative assertion.

Outdated regulatory references are equally problematic. AI models have training data cutoffs, and UK financial regulation changes frequently. The annual ISA allowance, pension lifetime allowance (now abolished but still referenced in older training data), capital gains tax thresholds, inheritance tax nil-rate bands — all of these have changed in recent years, and an AI model may generate content based on superseded figures. A financial promotion stating the wrong ISA allowance isn't just embarrassing; it's a breach of the requirement to be clear, fair, and not misleading.

Generic disclaimers and risk warnings represent a subtler but equally serious risk. AI tools, when asked to include appropriate disclaimers, tend to generate boilerplate language: "The value of investments can go down as well as up. Past performance is not a guide to future performance." These phrases are not wrong, but they may not be sufficient for the specific product or service being promoted. A promotion about pension transfers requires specific risk warnings about safeguarded benefits. A promotion about investment portfolio management needs warnings appropriate to the asset classes discussed. A promotion targeting a specific demographic needs warnings calibrated to that audience's likely understanding. Generic AI-generated disclaimers create a false sense of compliance — the disclaimers are present, but they're not adequate for the specific context.

Tone and balance problems are harder to detect but equally important. AI tools are trained to be helpful and positive, which means they naturally lean toward promotional language that emphasises benefits over risks. A human copywriter with financial services experience knows that a paragraph about investment growth should be balanced by a paragraph about the possibility of loss. An AI tool will generate the growth paragraph enthusiastically and add the risk paragraph grudgingly, often as a brief afterthought. The FCA's requirement for balance — that promotions give a fair impression of the product and don't emphasise benefits to the exclusion of risks — requires active editing of AI-generated content, not just fact-checking.

Confidentiality leakage is a newer concern. If your team inputs client scenarios, real portfolio details, or specific financial situations into a public AI tool to generate personalised content, that data may be stored and used in the model's future training. This creates a potential data protection breach and a confidentiality risk. Firm policies should explicitly prohibit inputting client-identifiable financial information into external AI tools.

Your existing compliance approval process probably doesn't need to be rebuilt from scratch, but it does need specific adaptations to handle AI-generated content effectively. The core change is that the reviewer needs to treat AI-drafted content with a different kind of scrutiny than human-drafted content — not more scrutiny necessarily, but differently focused scrutiny.

When reviewing content written by a known human copywriter — whether in-house or at an agency — the compliance reviewer can reasonably assume that factual claims are based on something, even if they need checking. The copywriter saw a statistic somewhere, referenced a regulation they'd read, or drew on professional experience. When reviewing AI-generated content, the reviewer cannot make this assumption. Every factual claim needs to be verified from scratch because there is no underlying source. The AI didn't read a report and cite it; it generated a plausible-sounding claim.

Practical adaptations to the approval workflow: first, require the content creator (the person who prompted the AI and edited the output) to provide source references for every factual claim. If the AI-generated draft states that "72% of UK adults don't have a will," the person submitting the content for compliance review should have found the actual source of that statistic (or a replacement statistic from a verified source) and included it in the submission. This shifts the fact-checking burden to where it belongs: on the person preparing the content, not on the compliance reviewer.

Second, add a specific check for regulatory currency. The compliance reviewer should confirm that any referenced allowances, thresholds, tax rates, or regulatory requirements reflect the current position, not a historical one. This check has always been good practice, but it's essential for AI-generated content where the model's training data may be months or years behind current regulation.

Third, review risk warnings and disclaimers with particular attention to specificity. Don't accept generic warnings that the AI has generated. Require warnings that are specific to the product, service, and target audience of the promotion. If the content discusses drawdown, the warnings should address drawdown-specific risks. If the content targets people approaching retirement, the language should be appropriate for that audience's likely financial literacy.

Fourth, assess balance explicitly. Read the content and ask: does it give a fair and balanced impression? AI-generated content tends toward optimism and promotional language. The compliance reviewer should check that risks are given equal prominence to benefits, that limitations of products or services are mentioned alongside advantages, and that the overall impression a reasonable reader would take away is accurate and balanced.

For firms under a network, the submission process should include disclosure that AI tools were used in content creation. Even if your network doesn't currently require this disclosure, providing it voluntarily demonstrates good governance and protects the firm if the network later introduces mandatory disclosure requirements. Include in your submission a brief note: "Initial draft generated with AI assistance. All factual claims verified against primary sources [listed]. Content reviewed and edited by [name] before submission."

The distinction between using AI to create first drafts and using AI to create publishable content is the most important operational boundary for adviser firms. When properly positioned, AI is a remarkably effective drafting tool that reduces the time between "blank page" and "workable first draft" from hours to minutes. When improperly positioned as a publishing tool — where AI output goes directly to the website, email platform, or social media with minimal human intervention — it becomes a compliance liability.

The first-draft use case works because it plays to AI's genuine strengths while keeping humans responsible for its genuine weaknesses. AI is excellent at structuring content: given a topic and a brief, it can produce a well-organised draft with logical section headings, reasonable flow between paragraphs, and comprehensive coverage of the key points. It's good at generating initial copy that a skilled editor can refine, much faster than writing from scratch. And it's useful for overcoming the "blank page" problem that stalls many marketing workflows — the most time-consuming part of content creation is often starting, not finishing.

But the first draft is just that: a starting point that requires substantial human work before it's fit for purpose. The human editor's job is not light-touch proofreading. It's a fundamental review that encompasses: verifying every factual claim against primary sources, updating any regulatory references to reflect current rules, adjusting tone and balance to meet FCA requirements, adding firm-specific context that the AI cannot know (your fee structure, your service proposition, your client demographic), removing generic language and replacing it with specific, relevant content, and ensuring the finished piece sounds like your firm rather than like a machine wrote it.

This last point matters more than most firms realise. AI-generated content has a recognisable style: fluent, competent, and generic. It lacks the specific voice, opinions, and personality that build trust with prospects. A financial adviser firm's marketing should sound like it was written by someone who actually advises real clients on real financial problems. AI output sounds like it was written by someone who read about financial advice on the internet. The editing process needs to inject that human specificity.

The publishing-tool anti-pattern typically emerges when firms are under time pressure. The marketing person needs a blog post by Friday, they're behind schedule, the AI generates something that reads well enough, and it goes live with minimal editing. This is where compliance breaches happen — not because anyone intended to publish non-compliant content, but because the review step was compressed or skipped. Establishing a firm policy that AI-generated content must go through the same approval process as any other financial promotion, regardless of time pressure, is a necessary safeguard.

Some firms have experimented with having AI check its own compliance — asking the same tool to review the content it generated for FCA compliance. This is inadequate. An AI model is not capable of reliably identifying its own hallucinations (by definition, it generated them because it couldn't distinguish them from accurate information). Self-review is circular and creates a false sense of security. Human compliance review remains non-negotiable.

Not all marketing tasks benefit equally from AI involvement. Understanding where AI adds genuine value and where it creates more work than it saves is essential for deploying it efficiently.

Content ideation and planning is one of AI's strongest applications. Ask a well-prompted model to generate 20 blog post ideas for a retirement planning specialist, and you'll get a usable list in seconds. The ideas will need filtering — some will be too generic, some will overlap with content you've already published, some will be off-strategy — but the brainstorming phase that might take a marketing team an hour takes five minutes. Similarly, asking AI to generate outline structures for articles, suggest angles for seasonal content, or identify questions your target audience is likely asking produces useful starting material quickly.

First drafts of educational content — blog posts, guide sections, FAQ pages — are the most common and most productive use case. The key qualifier is "educational." Content that explains general financial concepts, describes processes, or addresses common questions is relatively low-risk for AI-assisted drafting because the factual claims are typically well-established and easily verified. A blog post explaining how pension drawdown works, or what happens during the financial advice process, or how ISA allowances function — these topics have correct, verifiable answers and the AI generally gets the structure and flow right even if specific details need checking.

Social media post drafting is effective for generating volume. LinkedIn posts for advisers — sharing thoughts on market developments, commenting on regulatory changes, or promoting firm content — follow predictable patterns that AI handles well. The adviser or marketing person should always review and personalise the posts, but generating 10 LinkedIn post drafts in five minutes and selecting and editing the best three is far more efficient than writing three posts from scratch.

Email subject line and preview text generation benefits from AI's ability to produce multiple variations quickly. Generating 15 subject line options for your next newsletter, then selecting the strongest three for A/B testing, takes minutes and typically produces better results than a single human-written option because the variation set is wider.

SEO meta descriptions and title tags are well-suited to AI generation. These are short, formulaic pieces of text where the AI can follow a template (include the primary keyword, stay under the character limit, include a value proposition) and produce serviceable output that needs only light editing.

Ad copy drafting — for Google Ads responsive search ads and Meta ad copy — works well when the AI is given your existing high-performing copy as reference material. It can generate variations that maintain the core message while testing different angles, phrasing, and calls to action. The compliance review still applies, but the volume of testable variations increases significantly.

Client communication templates — welcome emails, review meeting preparation emails, post-meeting follow-ups — benefit from AI drafting because the structure is consistent across communications and the personalisation happens at the editing stage. Generating a template that the adviser then customises for each client is more efficient than writing each communication from scratch.

Certain types of financial marketing content carry such high regulatory risk that using AI in their creation requires either extremely thorough review or should be avoided entirely. Understanding these boundaries prevents the worst compliance outcomes.

Specific product recommendations or comparisons should never rely on AI-generated content. If your marketing compares two pension products, discusses the merits of a specific investment fund, or recommends a particular approach to tax planning, every word needs to be written or thoroughly verified by someone with the relevant technical knowledge. AI models frequently conflate product features across providers, mix up fee structures, and generate comparisons that sound authoritative but contain material inaccuracies. A misleading product comparison in a financial promotion isn't just a compliance breach — it's the kind of breach that triggers FCA enforcement action because it directly risks consumer harm.

Performance data and investment returns must never be AI-generated. Any content that references historical returns, fund performance, portfolio outcomes, or projected growth figures must be sourced directly from verified data providers (FE Analytics, Morningstar, the fund manager's own factsheets). AI tools will generate plausible performance figures that bear no relationship to reality. Publishing fabricated performance data in a financial promotion would be a serious regulatory breach that no amount of "the AI generated it" would mitigate.

Tax advice content requires extreme caution. UK tax rules are specific, they change annually (often in the Autumn Statement and Spring Budget), and the consequences of providing incorrect tax information are significant. AI models regularly generate tax content based on superseded allowances, abolished reliefs, or rules that apply in other jurisdictions. Any content discussing ISA allowances, pension tax relief, capital gains tax rates, inheritance tax thresholds, or dividend allowances must be manually verified against HMRC's current published guidance for the relevant tax year.

Regulatory status statements and firm descriptions need to be precisely correct. Your firm's FCA registration number, the exact name under which you're authorised, your regulatory status (directly authorised vs network member), the scope of your permissions — all of these must be accurate in every promotion. AI tools may generate approximate versions that contain errors. "Authorised and regulated by the Financial Conduct Authority" has a specific legal meaning and must reflect your actual regulatory status.

Complaint and cancellation information in promotions must comply with specific FCA rules about disclosure. The required wording depends on the product and the distribution method. AI-generated boilerplate may not meet the specific requirements for the promotion type in question.

Testimonials and case studies, even when fictionalised or composited, carry specific risks under FCA rules. AI-generated testimonials that describe unrealistic outcomes, imply guaranteed results, or fail to include appropriate caveats can breach COBS 4.7. If you use AI to draft case studies, the compliance review must be particularly rigorous in ensuring the depicted scenario is realistic, the outcomes are representative, and appropriate warnings are included.

The general principle: the closer the content is to specific financial advice, the less suitable it is for AI generation. General educational content about how pensions work sits at the safe end of the spectrum. Content that compares specific products, references specific performance data, or guides the reader toward specific financial decisions sits at the dangerous end. Map your content types along this spectrum and apply your review resources accordingly.

Documentation is the practical bridge between "we use AI responsibly" and "we can prove we use AI responsibly." The FCA hasn't mandated a specific documentation format for AI-assisted content creation, but the general principle of evidencing your compliance process applies: if you can't demonstrate that appropriate oversight was applied, the FCA may conclude that it wasn't.

The minimum documentation standard for AI-assisted financial promotions should cover four elements: what was generated, what was changed, what was verified, and who approved it.

What was generated: retain the original AI output before human editing. This doesn't need to be filed in your compliance records for every piece of content, but it should be retrievable if needed. A practical approach: save the AI prompt and raw output in a shared drive folder alongside the final approved version. This creates an audit trail showing the starting point and the extent of human intervention.

What was changed: the compliance submission should include a summary of material changes made to the AI draft. This doesn't mean tracking every word change — it means documenting substantive alterations: "Replaced AI-generated pension statistics with verified figures from ONS data (source: [reference]). Updated ISA allowance from £20,000 to current threshold. Removed unsupported claim about average adviser fees. Added specific risk warnings for pension drawdown." This demonstrates active human review, not rubber-stamping.

What was verified: factual claims in the final content should be annotated with their sources. This is good practice for any financial promotion, but it's essential for AI-assisted content where the reviewer needs to confirm that every claim has been checked against a primary source rather than accepted at face value from the AI. A simple approach: create a source document that lists each factual claim in the content alongside its verified source and the date it was checked.

Who approved it: the standard compliance sign-off process applies. The person approving the financial promotion (who must be appropriately qualified under your firm's compliance framework) should confirm that they reviewed the content knowing it was AI-assisted and applied appropriate scrutiny.

For firms under a network, check whether your network has specific documentation requirements for AI-generated content. Several networks introduced AI-specific submission requirements during 2025, and these may include mandatory disclosure forms, additional review checklists, or restrictions on which content types can be AI-assisted.

Record retention follows your existing policy for financial promotions. Under FCA rules, approved financial promotions and their approval records must be retained for at least three years from the date the promotion was last communicated. Apply the same retention period to AI-related documentation: the original AI output, the change summary, the source verification document, and the compliance approval record.

This documentation also protects the firm in the event of a complaint. If a client or prospect complains about the content of a financial promotion and the firm can demonstrate a thorough creation-review-approval process — including evidence that AI-generated content was subject to human verification — the firm's position is significantly stronger than if it cannot evidence its process. The documentation isn't bureaucracy for its own sake; it's the evidence that your firm took its responsibilities seriously.

Bringing all of these principles together, here's a step-by-step workflow that adviser firms can adopt for using AI tools in campaign creation while maintaining full compliance.

Step one: briefing. The marketing person (in-house or agency) creates a content brief that specifies the topic, target audience, key messages, required compliance elements (risk warnings, regulatory disclosures), and any factual claims that must be included with their verified sources. The brief is the human-created foundation; the AI doesn't write the brief.

Step two: AI drafting. Using the brief as the prompt foundation, the marketing person generates an initial draft using the AI tool. The prompt should include specific instructions about tone (professional, accessible, balanced), structure (the sections you want covered), and constraints (do not invent statistics, do not make specific product recommendations, include placeholders for risk warnings rather than generating them). Save the prompt and the raw AI output.

Step three: human editing. This is the most time-intensive step and it should be. The editor works through the AI draft systematically: verifying every factual claim against primary sources and replacing any unverified claims, updating all regulatory references to current figures, adjusting tone to match the firm's voice and ensure FCA-compliant balance between benefits and risks, adding firm-specific content that the AI cannot know, inserting appropriate risk warnings specific to the product or service discussed (not generic boilerplate), and removing any content that sounds authoritative but lacks a verifiable basis.

Step four: source documentation. The editor creates or updates the source reference document, listing every factual claim in the edited content alongside its verified source. This document accompanies the content through the approval process.

Step five: compliance review. The content is submitted for compliance approval through your standard process. The submission includes: the edited content (not the raw AI output), the source reference document, a note disclosing that AI tools were used in the initial drafting and describing the extent of human editing, and the standard compliance review checklist (clear, fair, not misleading; appropriate risk warnings; suitable for target audience; Consumer Duty considerations).

Step six: revision and approval. The compliance reviewer provides feedback, the editor makes any required changes, and the content is resubmitted until approved. This step is identical to the non-AI process.

Step seven: publication and filing. The approved content is published through the appropriate channel. The compliance file retains: the approved final version, the compliance approval record, the source reference document, and the AI disclosure note. The raw AI output is retained separately in the marketing team's working files.

The time savings from this workflow are real but more modest than the "AI will do your marketing for you" narrative suggests. A blog post that might take four hours to research, draft, edit, and submit for approval might take two and a half hours with AI handling the initial draft. The research, editing, compliance preparation, and approval steps still require skilled human time. What AI eliminates is the slow, often frustrating process of generating the first draft from a blank page. For a firm producing regular content — weekly blog posts, monthly newsletters, ongoing social media — that time saving compounds meaningfully.

The firms getting the most value from AI in their marketing are the ones that have accepted this reality: AI is a productivity tool that makes skilled marketers faster, not a replacement for skilled marketers. The compliance framework exists to ensure that the speed gain doesn't come at the cost of quality or regulatory adherence. Done properly, AI-assisted content creation produces more content, faster, without increasing compliance risk. Done carelessly, it produces content that's faster to create and faster to trigger a regulatory problem.