A customer opens ChatGPT and asks: "What is the minimum deposit for a Halifax first-time buyer mortgage?" The AI answers confidently. The figure it gives is outdated. The customer plans their finances around it. Later, they are surprised at the application stage.

This scenario is not hypothetical. AI systems are actively generating answers to financial product questions using information they have indexed, synthesised, and summarised — without real-time access to your current documentation, terms, or rates. And unlike a Google search result, the AI presents its answer as a direct statement, not a link to verify.

The compliance exposure most firms haven't considered

For FCA-regulated firms, the regulatory framework around financial promotions is well established. A communication that is misleading, unfair, or not clear about a financial product is a compliance matter — regardless of who produced it.

The harder question is whether an AI-generated description of your product, which you did not produce and cannot fully control, constitutes a financial promotion for regulatory purposes. Legal opinion on this is evolving. But the spirit of Consumer Duty — requiring firms to ensure customers receive accurate information and are not harmed by it — provides a clearer direction of travel.

The distinction matters. Regulators have historically taken a dim view of systemic gaps that a firm should reasonably have identified and addressed. AI-generated misrepresentation is increasingly within that territory.

What AI systems are actually saying about financial products

Our analysis of major AI platforms — including ChatGPT, Google's AI Overview, Perplexity, and Microsoft Copilot — reveals several consistent patterns in how financial products are described.

First, AI systems frequently use outdated information. Rates, thresholds, and eligibility criteria change regularly, but AI training data does not update in real time. A savings rate from 18 months ago may still be cited as current.

Second, AI systems often conflate products across providers. A general description of "how ISAs work" may blend information from multiple sources, creating a composite that accurately describes no specific product.

Third, nuance is frequently lost. Eligibility conditions, exceptions, and product-specific terms are often omitted in favour of a cleaner, simpler answer. For financial products, these are precisely the details that matter most.

What responsible firms should do now

The honest answer is that full control over what AI systems say about your products is not yet achievable. But the gap between doing nothing and taking reasonable steps is significant — and it is likely to become more significant as regulatory expectations develop.

Three things are within reach for most regulated firms today.

Understand your current AI-generated presence. Systematically review what major AI platforms are saying about your key products. This is not a one-time exercise — it requires ongoing monitoring as AI systems update.

Improve the quality of the information AI systems are reading. AI systems synthesise information from the sources they can access. Firms that provide clear, structured, schema-marked content give AI systems better material to work from. This does not guarantee accuracy, but it improves the probability of accurate representation.

Put an escalation framework in place. When a material misrepresentation is identified, who is responsible for responding? What is the process? Having this documented is both good practice and defensible if the question is ever asked by a regulator.

A customer asks an AI platform: "What is the minimum deposit for a first-time buyer mortgage?" The platform answers — confidently, immediately, and in detail. The information it provides may be accurate. It may be outdated. It may conflate criteria across multiple lenders. The customer, receiving what appears to be a direct and authoritative answer, plans accordingly.

This is not a hypothetical scenario. It is the current operating reality across every major AI platform — ChatGPT, Google AI Overview, Perplexity, Microsoft Copilot. Customers are receiving AI-generated answers to financial product questions at scale. Most FCA-regulated firms have not assessed what those answers say.

The governance gap

The information AI platforms generate about financial products was not submitted by the firms whose products are being described. It was not reviewed by compliance. It was not approved as a financial promotion. It is drawn from training data — a synthesis of web content, product documentation, and third-party sources — that may be months or years out of date.

For most regulated products, this creates an immediate governance question: who is responsible for the accuracy of AI-generated descriptions of your products?

The FCA's Consumer Duty framework requires firms to ensure customers receive information that is fair, clear, and not misleading — and to take reasonable steps to prevent harm. It does not carve out an exception for information generated by third-party AI systems. A firm that knows its products are being misrepresented in AI-generated search and takes no action is in a materially different regulatory position to one that has assessed the issue and implemented a response.

What misrepresentation looks like in practice

Analysis of AI-generated responses about regulated financial products reveals several consistent failure modes.

Outdated information. AI training data has a cutoff. Rate changes, updated eligibility criteria, and revised product terms introduced after that cutoff are not reflected. A savings rate from eighteen months ago may still be cited as current. A mortgage product that has been withdrawn may still appear as available.

Cross-provider conflation. AI platforms frequently synthesise information from multiple sources. A response about "fixed-rate mortgage eligibility" may blend criteria from several lenders into a description that accurately represents none of them.

Omission of material conditions. Eligibility thresholds, exclusion criteria, and product-specific conditions are frequently absent from AI-generated summaries. For regulated products, these are precisely the details that carry most compliance weight.

The escalation question

Most regulated firms do not yet have a defined owner for AI search representation. Marketing is often focused on visibility rather than accuracy. Compliance is often unaware that AI-generated descriptions exist, let alone that they may be inconsistent with approved product documentation. Legal has typically not been asked to assess the question.

The absence of ownership means there is also no escalation path. When a material misrepresentation is identified — an incorrect rate, an inaccurate eligibility criterion, a withdrawn product still described as available — there is no defined process for responding. This is the governance gap that most regulated firms currently have.

Most FCA-regulated firms have some form of media monitoring in place. Press coverage, social media, review platforms — these are tracked, logged, and in many cases reviewed by compliance before action is taken. AI-generated search is absent from almost every monitoring framework we have encountered.

This is a material gap. AI platforms are generating answers to financial product questions at a scale and frequency that dwarfs most other channels. A customer who types a mortgage question into ChatGPT or asks Google's AI Overview about ISA rates receives an immediate, confident, synthesised answer — drawn from sources the firm has no visibility over and did not approve.

What to monitor, and why

Product descriptions and feature claims. The most fundamental monitoring requirement is accuracy of product description — rates, terms, features, and eligibility criteria. These change regularly. AI training data does not update in real time. The gap between what is currently true and what AI platforms are saying can be significant, and it accumulates over time without intervention.

Eligibility criteria and exclusions. Eligibility conditions are among the highest-risk elements of AI-generated financial product descriptions. A response that omits a material exclusion criterion — income thresholds, residency requirements, credit history conditions — may lead a customer to apply for a product they do not qualify for, or to make financial decisions based on incorrect assumptions about their eligibility.

Comparative positioning. AI platforms frequently generate responses that implicitly or explicitly compare providers. How your firm's products are positioned relative to competitors — on rate, on features, on accessibility — is determined by AI synthesis, not by your marketing or communications function. Understanding this positioning is a prerequisite to influencing it.

Regulatory and disclosure language. FCA-regulated products carry specific disclosure and regulatory language requirements. AI-generated descriptions frequently strip these out in favour of simplicity, creating summaries that may not meet the standards expected of approved financial promotions — even though they are not technically promotions in the traditional sense.

How to structure a monitoring framework

Effective monitoring of AI-generated search representations requires three components.

First, a defined query set — the specific questions customers are likely to ask about your products on AI platforms. These should be drawn from customer research, search data, and product knowledge, and reviewed regularly as product ranges and customer behaviour evolve.

Second, a systematic review process — regular sampling of AI platform responses to those queries, with a structured framework for assessing accuracy against current approved product documentation.

Third, an escalation path — a defined internal process for what happens when a material misrepresentation is identified. Who is informed? Who is responsible for the response? What remediation options exist? Without this, monitoring produces findings that sit in a report without consequence.

Who owns this?

In most regulated firms, AI search monitoring currently sits in no one's remit. Marketing is focused on visibility and acquisition. Compliance is focused on approved communications and financial promotions. Digital teams are focused on conventional search and owned channels.

Establishing ownership is the necessary first step. In practice, this is most effectively approached as a cross-functional responsibility — with compliance setting the accuracy and disclosure standards, marketing owning the visibility and representation strategy, and digital owning the technical infrastructure that influences what AI platforms can read and synthesise.

For the past two decades, the dominant question in search strategy has been: how do we rank higher in Google? The answer to that question has been refined continuously — keyword optimisation, backlink strategies, technical site performance, content quality signals. An entire industry has developed around it.

AI-generated search asks a fundamentally different question: which sources does this AI system trust enough to synthesise answers from — and what does it say when it does?

These are not the same question. The skills, tools, and methodologies that address one do not address the other. Firms and agencies that treat AI search as an extension of conventional SEO are starting from an incorrect model of the problem.

How conventional search works

Conventional search engines return a ranked list of URLs. The ranking is determined by a complex set of signals — relevance, authority, technical performance, user behaviour — that have been the subject of optimisation for years.

The user clicks a link. They visit a page. They form their own judgment about the content. The search engine's job is to surface relevant pages, not to interpret or summarise them.

How AI-generated search works

AI search platforms do not return a ranked list of URLs. They return a synthesised answer — a direct response to the user's question, drawn from multiple sources, presented as a single coherent output.

The sources the AI draws on are not determined by the same signals that drive conventional search rankings. They are determined by what the AI system can read, parse, and trust — which depends on how information is structured, attributed, and made available to the system's indexing and training processes.

A firm with strong conventional search rankings may have poor AI search presence. A firm with weaker conventional rankings may be well-represented in AI-generated answers because its content is structured in a way that AI systems can synthesise with confidence. The correlation between the two is weak and unreliable.

What AI search visibility actually requires

Entity optimisation. AI systems understand the world through entities — named things, with defined attributes and relationships. A financial services firm that is clearly defined as an entity — with consistent, structured information about what it is, what it offers, and how it relates to other entities in the financial services landscape — is more reliably represented in AI-generated answers.

Schema and structured data. Machine-readable markup tells AI systems what information on a page means, not just what it says. Product terms, eligibility criteria, rates, and features expressed in structured schema formats are more reliably read, understood, and synthesised by AI systems than the same information presented in unstructured prose.

Knowledge graph presence. AI systems draw on knowledge graph data — structured information about entities and their relationships — as a high-confidence source. Firms that are well-represented in knowledge graphs are more consistently included in AI-generated answers, and the information presented tends to be more accurate.

Source authority and indexability. AI systems apply their own assessment of source authority and reliability. This is not identical to conventional domain authority, and it does not respond to the same optimisation signals. Understanding what makes a source trusted by AI systems — and ensuring your firm's content meets those criteria — requires a different analytical approach.

The agency problem

Most digital agencies, including those with strong SEO capabilities, have not yet built the methodologies to address AI search visibility. This is not a criticism — the field is genuinely new, and the tools and frameworks are still developing. But it does mean that delegating AI search strategy to an existing agency relationship, without verifying that the agency has specific AI search capability, is unlikely to produce meaningful results.

The first step is to understand your current position accurately — what AI platforms are saying about your firm, where the gaps and inaccuracies are, and what structural factors are driving them. That understanding is the prerequisite for any meaningful strategy.