What happens when an AI assistant quotes insurance without a live data connection?
When a consumer asks ChatGPT or Claude about an insurance product, AI generates a response based on training data and any indexed web content — enter every CMO's SEO, AEO, and GEO strategies.
The issue, however, is that it doesn't connect directly to an insurer's pricing engine or retrieve real-time policy docs. The figures it returns for a premium aren't guaranteed to be accurate.
In an unregulated context, that lack of clarity adds up to a minor inconvenience. But in a regulated financial services context, it opens the potential to genuinely misleading interactions and even mis-selling — a definite can of worms.
Whilst the regulation hasn't quite caught up with 2026's channels, on a principle basis it's evident that a consumer who purchases a policy based on inaccurate information about its terms has been materially misled, regardless of whether the source of that misinformation was a human, a website, or an AI model.
Gartner has forecast that more than 2,000 legal claims linked to AI will materialise in the near term, reflecting a broader recognition that the liability exposure created by uncontrolled AI outputs in regulated sectors is very real. Insurance is a sector where every output carries a potential contractual commitment, and as such sits at the sharp end of this risk.
Why does the FCA's Consumer Duty make AI distribution more complicated?
Consumer Duty, which came into force in July 2023, requires firms to demonstrate they're delivering good outcomes for customers, over and above simply following a documented process. For any insurer exploring AI as a distribution channel, this changes the question from "did we show the customer the right information?" to "did the customer actually understand what they were buying, and was the product right for them?" Big difference.
An AI assistant operating without a structured compliance layer simply can't satisfy that standard. It'd be unable to verify that the product it describes matches the customer's stated needs, nor would it be capable of confirming that the price it surfaces reflects the current underwriting logic. Last, but certainly not least, it wouldn't be able to produce the auditable record of the interaction that Consumer Duty requires as evidence of a good outcome.
Can insurers not just connect an AI assistant to their existing quote-and-buy journey?
This is the most common question insurers raise when exploring AI distribution. Unfortunately, a direct connection void of any middle-layer generally creates more problems than it solves.
Existing quote-and-buy journeys are built for a specific type of interaction: an individual on a screen filling in a structured form with predominantly static information displayed alongside. An AI conversation behaves quite differently. The user might ask about coverage in an order that the pre-set process doesn't anticipate, or ask comparative questions across products, or even providers, mid-conversation. The language they use, and the language people generally use when interacting with the likes of ChatGPT, is often informal and maps imprecisely to what can often be quite static underwriting categories.
Pointing an AI assistant at a quote-and-buy journey without a layer that translates between conversational inputs and structured underwriting logic produces unreliable outputs and opens up potential compliance risks. McKinsey's research on AI in insurance notes that processes need to be revamped end-to-end to extract value from AI, rather than simply layering AI on top of existing processes.
The same principle applies to distribution: the compliance architecture has to be rebuilt for the channel and can't simply be retrofitted from one that was designed for a different interaction entirely.
What is the difference between an insurer being visible in AI and being transactable in AI?
Visibility in this scenario means AI mentions your product or brand when a consumer asks a relevant question. Transactability means the consumer can receive an accurate, live quote and purchase a policy without leaving the AI environment. This is a major distribution unlock where early-movers win.
Most insurers that have any AI presence today have visibility only. Their products appear in AI-generated answers because their web content is indexed well and retrievable (see things like Malcolm's GEO Leaderboard for Insurers to get a glimpse into how this is stacking up in 2026). But the answer AI produces is assembled from that static content, not from a live connection to their pricing engine or other underwriting systems. The quote a consumer sees is therefore an approximation, and not always a well-grounded one at that. The winner of this next wave of distribution in insurance will be the ones that combine GEO, AEO and, importantly, AIO:
- GEO: Generative Engine Optimisation — optimising your brand and content to appear in AI-generated answers across tools like ChatGPT, Claude, & Perplexity.
- AEO: Answer Engine Optimisation — structuring said content so search engines & AI assistants surface it as the direct answer to a specific user question.
- AIO: Artificial Intelligence Optimisation — the broader practice of making your products, data, and infrastructure readable, queryable, and actionable by AI systems, end to end.
GEO and AEO get you found and recommended inside, say, ChatGPT. But without AIO, the journey breaks the moment a consumer tries to get an actual quote or purchase a policy. An insurer can rank top in every AI answer but still lose the sale because their pricing engine isn't queryable, their product data isn't structured for AI consumption, and there's no compliant path to purchase inside the conversation. To close the sale, the insurer needs to hope that the user clicks on the right link to their site from ChatGPT and somehow navigates to the right product…by which point, all context and user insight from the initial conversation within ChatGPT would have been lost. You can see how this introduces unnecessary friction, drop-off, and frustration on both sides.
As we explored in our recent article, ChatGPT is now a distribution channel, and consumers are already asking AI assistants insurance questions at a significant volume.
You might appear top of the charts from a GEO and AEO standpoint, but if your competitor plugs the AIO gap, ChatGPT could surface your competitor to the customer in question since they're able to continue their quoting process directly in chat, without ever leaving ChatGPT.
Malcolm is the InsurTech built specifically to plug that critical gap for insurers.
What does a compliant AI distribution architecture actually look like?
McKinsey recently described a near-future model in which AI agents handle customer onboarding in insurance, with separate agents handling intake, risk profiling, pricing, compliance review, and decision orchestration.
The architecture to execute on this, where we come in, and the compliance layer in a well-structured AI distribution setup work similarly. The AI interfaces the conversation while a middleware layer, Malcolm, retrieves live pricing data, applies underwriting rules to the customer's stated circumstances, enforces disclosure obligations, and returns an accurate, current quote. The insurer's core systems remain the source of truth throughout. Nothing is approximated, and nothing is generated that hasn't been validated against live product data.
Why solve this now?
As we wrote in The CMO Playbook for the AI Era, the pattern in insurance distribution is consistent: every major channel innovation was dismissed as marginal by incumbents until it wasn't. The window between early mover and laggard closes faster than anyone anticipates, and the advantages built by early movers compound over time.
McKinsey's analysis finds that only a small number of insurers have so far extracted significant competitive value from AI (in all segments, not just distribution). Insurers who treat compliance as the reason not to engage with AI distribution, rather than a major opportunity, are at risk of making the same category of mistake that comparison site laggards made in the early 2010s.
If you want to understand what this all looks like in practice for your product lines, get in touch.
You can also explore which insurers are currently appearing in AI responses across different lines of business.
Original LinkedIn Article here.
Frequently asked questions
Why most insurers can't accept AI traffic directly
The majority of insurers can't accept AI traffic directly because large language models generate responses probabilistically by default, rather than by accurately retrieving live policy data. This simply means that quotes, coverage limits, and exclusions surfaced to an individual during a conversation with ChatGPT, without a compliant middleware layer, are very likely to be inaccurate and could constitute a mis-sale under FCA rules, including Consumer Duty obligations that came into force in 2023.
What is trymalcolm.com, and what does it do for insurers?
Malcolm is a compliant middleware layer that connects legacy insurer pricing engines and underwriting systems to modern AI distribution channels, including ChatGPT. It allows insurers to surface accurate, live quotes inside AI conversations without hallucination risk, without storing personal data, and without requiring insurers to rebuild their existing systems.
Why can AI models not simply retrieve accurate insurance information from insurer websites or APIs?
Product pricing for insurance products is dynamic. Complex underwriting rules can change, products are continuously updated, new products are released, and risk appetites for certain lines are often adjusted. If AI pulls information from a website, based on the last time it was indexed, there's a high risk that the information surfaced to a user is outdated. The only way to solve this problem is by having a compliant, secure middle-layer that connects pricing APIs to AI conversations.
What does Consumer Duty require from insurers using AI distribution?
Consumer Duty requires firms to demonstrate good outcomes for customers, including that customers receive accurate information, that products are appropriate for their needs, and that firms can evidence the quality of the interaction. Any conversations that revolve around insurance quotes are required to produce an auditable record of what was said, quoted, and on what basis. Without the necessary middleware, insurers simply cannot offer this.
Does being mentioned in AI responses generate insurance sales?
In theory, it can help from a visibility standpoint. But visibility itself does not generate sales directly if there's no AI-native means in which to do so. Consumers asking insurance questions inside AI assistants need to receive an accurate quote and complete a transaction in that environment to convert. The full value of AI distribution is unlocked when customers can interact, from discovery to purchase, directly in chat.
Is ChatGPT or Claude retrieving information about my insurance products right now?
If your brand and products are being indexed by Bing, Google, or equivalent platforms, ChatGPT and Claude are likely retrieving and referencing certain content to prospective customers. Whether those responses are accurate depends on how well your content is structured for AI retrieval and whether your pricing information is connected directly with the AI in question.
Malcolm's GEO rankings tool tracks which insurers are appearing in AI responses and in which lines of business.
For a full list of all Malcolm's FAQs covering all things AI & agentic insurance, compliance and regulation, distribution and integrations, visit our dedicated FAQs page here.
Malcolm is the compliant infrastructure layer for AI insurance distribution. It connects insurer pricing and underwriting to AI-native channels without hallucination risk and without FCA exposure. Get in touch to see how it works.