UK government trials chatbots in bid to bolster small business support
While earlier tests had problems with accuracy, the next version of GOV.UK Chat shows consistent improvements


The UK government is running a private beta of a new chatbot designed to help people set up small businesses and find support.
The generative AI chatbot, based on OpenAI’s GPT-4o technology, will be tested out by up to 15,000 business users, after earlier trials went well.
GOV.UK Chat is linked from 30 government business support pages, including 'set up a business' and 'search for a trade mark'. Users can ask questions about tax and the support available to them.
Technology secretary Peter Kyle said the use of chatbots could markedly improve efficiency and enable the government to provide timely support for small businesses across the country.
"Outdated and bulky government processes waste people’s time too often, with the average adult in the UK spending the equivalent of a working week and a half dealing with public sector bureaucracy every year," Kyle said.
"We are going to change this by experimenting with emerging technology to find new ways to save people time and make their lives easier, as we are doing with GOV.UK Chat."
If this trial goes well, it could ultimately be rolled out across the full government website, made up of 700,000 pages. This attracts over 11 million users per week and is the best-known digital service in the UK, according to YouGov.
Get the ITPro daily newsletter
Sign up today and you will receive a free copy of our Future Focus 2025 report - the leading guidance on AI, cybersecurity and other IT challenges as per 700+ senior executives
"With all new technology, it takes time to get it right so we’re taking it through extensive trials with thousands of real users before it is used more widely," said Kyle.
"This is an essential part of our ambition to use AI to improve public services in a safe and reliable way, making sure the UK government leads by example in driving innovation forward."
During the first trial, nearly seven-in-ten users said the chatbot was helpful, with only 15% disagreeing.
However, the trial did throw up some problems with accuracy, generating a number of 'hallucinations'. Some of these were “challenging, or abusive, or even seductive”, according to Paul Willmott, chair of the government’s Central Digital and Data Office
Since then, developers behind the chatbot said it's improved significantly. The team has been working with subject matter experts at HMRC to score the accuracy of the chatbot's answers, assess AI answers against example answers written by content designers, and monitor for inaccurate or inappropriate answers and investigate any they find.
There's now also a message included in the onboarding process explaining the risk of inaccurate answers to users, and a link underneath every answer so that users can check the source guidance.
"Unlike ChatGPT, GOV.UK Chat is designed to draw on GOV.UK as the source of its answers,” according to Sam Dub and Josh Davey of the development team. “This means we can ensure it’s always using the most up-to-date guidance, and users can trust the answer comes from government."
There have also been improvements aimed at preventing the chatbot from giving harmful responses such as illegal answers, the sharing of sensitive financial information, or the adoption of a political position.
Emma Woollacott is a freelance journalist writing for publications including the BBC, Private Eye, Forbes, Raconteur and specialist technology titles.
-
Bigger salaries, more burnout: Is the CISO role in crisis?
In-depth CISOs are more stressed than ever before – but why is this and what can be done?
By Kate O'Flaherty Published
-
Cheap cyber crime kits can be bought on the dark web for less than $25
News Research from NordVPN shows phishing kits are now widely available on the dark web and via messaging apps like Telegram, and are often selling for less than $25.
By Emma Woollacott Published
-
Meta executive denies hyping up Llama 4 benchmark scores – but what can users expect from the new models?
News A senior figure at Meta has denied claims that the tech giant boosted performance metrics for its new Llama 4 AI model range following rumors online.
By Nicole Kobie Published
-
DeepSeek and Anthropic have a long way to go to catch ChatGPT: OpenAI's flagship chatbot is still far and away the most popular AI tool in offices globally
News ChatGPT remains the most popular AI tool among office workers globally, research shows, despite a rising number of competitor options available to users.
By Ross Kelly Published
-
Productivity gains, strong financial returns, but no job losses – three things investors want from generative AI
News Investors are making it clear what they want from generative AI: solid financial and productivity returns, but no job cuts.
By Nicole Kobie Published
-
Legal professionals face huge risks when using AI at work
Analysis Legal professionals at a US law firm have been sanctioned over their use of AI after it was found to have created fake case law.
By Solomon Klappholz Published
-
Microsoft says AI tools such as Copilot or ChatGPT are affecting critical thinking at work – staff using the technology encounter 'long-term reliance and diminished independent problem-solving'
News Research from Microsoft suggests that the increased use of AI tools at work could impact critical thinking among employees.
By Nicole Kobie Published
-
Future focus 2025: Technologies, trends, and transformation
Whitepaper Actionable insight for IT decision-makers to drive business success today and tomorrow
By ITPro Published
-
Looking to use DeepSeek R1 in the EU? This new study shows it’s missing key criteria to comply with the EU AI Act
News The DeepSeek R1 AI model might not meet key requirements to comply with aspects of the EU AI Act, according to new research.
By Rory Bathgate Published
-
The DeepSeek bombshell has been a wakeup call for US tech giants
Opinion Ross Kelly argues that the recent DeepSeek AI model launches will prompt a rethink on AI development among US tech giants.
By Ross Kelly Published