Launching the UK’s first credit score inside ChatGPT

On 20 August 2026, a consumer in the UK accessed their personal credit score inside an LLM for the first time. They did it with a single conversational prompt, “@Experian, what’s my credit score?”, and their personalised 1250 Experian Credit Score appeared in seconds, alongside their score history and a breakdown of what goes into it. The interaction was powered by an agentic Model Context Protocol (MCP) integration between Experian and OpenAI’s connected apps, built so that no user data was ever accessed by ChatGPT’s model. You can read the launch announcement here.

Our strategy at Experian Consumer Services has always been to help consumers better their score, because a better score ultimately means a better story. That ambition doesn’t stop at our own app. It continues wherever the financial story our users are writing takes them, and we want to be there whenever a financial conversation is happening, offering the trusted, accurate insight that only Experian can give.

Until now, that wasn’t possible in the place more and more people are turning to first. Someone having a mortgage-readiness conversation inside ChatGPT had to jump between apps and manually piece information together to make the conversation meaningful. This launch removes that barrier at the exact moment it matters most.

Why Experian joined the FCA’s AI Live Testing programme

Experian developed and launched this experience, as part of the FCA’s AI Live Testing programme. The programme was designed to support safe and responsible AI deployment, and Experian was selected to join it at the beginning of the year.

Being the first to do something nobody has done before is risky. Having an open dialogue with the regulator and chance to explore regulatory considerations and any grey areas meant we could move faster, not slower, in our efforts to deliver a safe and reliable AI solution.

The collaboration helped us surface and discuss key issues with the FCA at an early stage, test our assumptions in controlled real-world conditions, and reflect what we learnt in a customer-centric design that delivers the right outcome. It also meant we could contribute practical experience and emerging good practice back to the regulator, helping to shape future policy on AI for the rest of financial services as the industry works through what it means to show up inside LLMs.

What we tested

Across the research we’ve done, users have told us that LLMs have become the number one place they turn to for financial guidance, but that the lack of accurate, up-to-date information is the main thing stopping them acting on it. 62% of regular ChatGPT users already turn to AI for help with money goals, and over 200 million people ask ChatGPT finance questions globally. Our own research found that nearly one in three UK adults use generative AI for financial guidance each week and 85% of Brits trust AI to educate them.

The scope was deliberately narrow: one score, one prompt, one clear outcome. That let us focus the testing on three areas.

Accuracy and accountability

Launching inside an existing third-party experience meant we needed to understand how the integration worked in practice and identify potential risks for consumers. For example, we tested the accuracy of responses, how to make it clear when the model was drawing on Experian information rather than other sources, and what happens if the model produced adverse outcomes.

Customer comprehension and privacy

We tested whether users understood what they were looking at, where the data came from, and who could see it. The experience was designed so that the score is only ever rendered inside Experian’s secure, logged-in widget, and is never exposed to or used by the model when it generates responses. Testing copy as well as design mattered here: a technically sound privacy model is of little use if the wording makes people doubt it.

Signposting and the treatment of vulnerable users

We tested how clearly users could find their way from the ChatGPT experience back to Experian’s own app and website, where the full credit report, improvement tips and human support sit.

What we learnt

Build with a clear customer segment and problem in mind

A specific, narrow use case with data minimisation at heart, focuses regulatory conversations, makes unhappy paths testable, and surfaces the real risks rather than hypothetical ones. It is also what makes an AI solution likely to see the light of production. For us, that meant starting with one score and one question, not a general financial assistant.

Use research to tell real risks from hypothetical ones

Research before, during and after launch is how you evidence whether a risk is likely to materialise or is unlikely ever to come to life. Test the copy as well as the design: wording raises early flags about misunderstanding long before anything reaches production. Be ruthless about following what users tell you. For us it meant de-scoping parts of the experience we were excited about, but that users felt less useful inside ChatGPT and would expect it elsewhere.

Be clear on what success looks like from the outset

Define it before you build, and check it at every stage: during build, at go-live and afterwards. Without that, it’s impossible to say whether the experience is delivering the right customer outcome or just working technically.

Let design and copy carry the weight when you’re a guest in someone else’s product

When you launch inside third-party infrastructure, your part of the experience has to do the heavy lifting to differentiate itself, so users always know when they’re interacting with you and when they’re interacting with the platform. Make it obvious how to transfer to your own app for the support they need. If you do it right, trust that was earned in one channel can be carried into a new one.

What happens next

Edu Castro, Managing Director of Experian Consumer Services, UK&I, put it this way: “As people increasingly seek money help on a range of different channels, it’s important the information they get is accurate and genuine. This is fundamental to our strategy of meeting people wherever they choose to engage with us and helping make credit understanding truly accessible, especially for younger consumers who are growing up with AI as their first port of call for financial questions.”

Since the app went live in August, we have received positive feedback from users, and we’re excited to see the level of engagement with it continue to grow as it scales.

We’re pleased to have contributed to the paper the FCA published titled: “AI in financial services: good practice and areas for improvement”, which draws on the AI Live Testing programme alongside the FCA’s wider supervisory work. We’ll keep sharing what we learn with the FCA and the wider industry as we extend what’s possible inside AI platforms. This is only the start of the journey