Banking and Finance: When AI Agents Need to Read Your Product Pages
When someone asks an AI agent "What's the best savings account right now?", the agent needs to compare interest rates, access restrictions, minimum deposits, and bonus conditions across multiple providers. On many banking product pages the agent cannot extract this information accurately from a single page, which puts the provider at a disadvantage before a human is ever involved.
The Problem in Detail
Banking product pages are often designed around storytelling. A savings account page opens with a lifestyle image, followed by a couple of paragraphs about financial goals, then a prominently displayed headline rate. The headline rate is frequently rendered as an SVG graphic, not text. Below it sits a "Key Facts" section that is actually a series of expandable accordion panels, each requiring a JavaScript click event to reveal its contents.
Test several AI agents against a page like this and the failures are consistent. None extract the correct interest rate, because it is an SVG image. Several cannot access the key facts at all, because the accordion panels are collapsed by default and the agents cannot trigger the expansion. The few that manage to expand some panels extract partial information but miss critical details about tiered rates and bonus conditions.
Comparison is worse still. Ask agents to compare three savings products on a site like this and the output is inaccurate across the board, because the agents cannot reliably extract data from any individual page. An agent might confidently state a Fixed Rate ISA offers one AER when the real rate is materially different. For a regulated product, that is a serious problem.
What to Change
Make interest rates text. Convert every rate on every product page from SVG graphics and styled spans to plain HTML text with Schema.org FinancialProduct markup. The AER, gross rate, and any tiered or bonus conditions should all be present in the page source as readable text.
Make key facts visible by default. Remove the accordion pattern. Display all product information in full on page load, organised in a clear HTML table: eligibility criteria, minimum and maximum deposits, access restrictions, withdrawal penalties, interest payment frequency, and FSCS protection status. Use th elements for labels and td elements for values, with appropriate scope attributes.
Add a product comparison endpoint. Build a simple JSON API that returns structured data for any product or set of products, following an API-first design approach. Agents that can make API calls then get clean, reliable data without parsing HTML. Keep the endpoint publicly accessible and documented on a developer page linked from the site footer.
Use a consistent page structure. Give every product page the same template: product name as h2, a one-sentence summary, the rate table, the key facts table, eligibility details, then the apply link. Agents that learn the structure of one product page can then reliably parse the rest.
The Results You Can Expect
With rates as text and key facts visible by default, agents can finally extract interest rates, access terms, and minimum deposits accurately, and agent-referred applications can grow as a result. Fixed-rate products tend to benefit most, because their straightforward, numerical nature makes them the most commonly compared by agents.
The accuracy of agent-generated descriptions improves dramatically. Comparison accuracy improves too: agents can produce correct side-by-side tables that were simply not possible when the underlying data was locked in images and accordions.
The Compliance Angle
Financial services have a specific reason to care about agent accuracy beyond commercial benefit. If an AI agent misrepresents a financial product, and a customer makes a decision based on that misrepresentation, the regulatory situation is unclear but the reputational risk is not. For many compliance teams this is a risk exercise as much as a growth exercise.
By providing clear, structured, machine-readable product data, you reduce the likelihood that agents present inaccurate information about your products. You cannot control what agents do with the data, but you can make sure the source data is unambiguous.
Takeaways for Financial Services
If your interest rates are displayed as images, SVGs, or styled graphics rather than text, AI agents cannot read them. This is the single most common problem in financial services websites. Fix it first.
If your product details are hidden behind accordions, tabs, or click-to-reveal interactions, agents may not access them. Display key information by default, using agent-friendly form patterns for any application flows.
If you have more than a handful of products, consider a structured data API. It costs little to build and gives agents a reliable path to accurate information. This protects your brand as much as it drives applications.