The UK Government’s experimental genAI chatbot for small businesses is set to move to the next stage of testing this week, with thousands of business owners able to receive extra help and advice.

Up to 15,000 people will be able to ask the GOV.UK AI for advice on business rules and support, with the chatbot linked to thirty business guides on the government’s webpages. People with access to the trial can search for terms like “set up a business” and “search for a trade mark”, as well as ask questions about tax and the support available to them. 

A team of in-house data scientists, developers and designers are building the experimental tool using OpenAI’s GPT-4o platform, aiming to help people more quickly navigate often complex business advice.

The government said that the new chatbot will offer straightforward, personalised answers that collate information that might otherwise be spread across dozens of pages.

After the first trial, which was conducted late last year, nearly 70% of users agreed that the responses provided by the chatbot were helpful, while less than 15% disagreed. 

However, the first trial also showed that more testing and development was required to improve accuracy standards for advice and information to match the quality of that already found online.

The results from this latest trial will determine the next steps which could include potential larger-scale testing, and ultimately lead to the chatbot being rolled out across the full government website, made up of 700,000 pages.

Stressing the experimental nature of the AI tool, the government made it clear in its announcement that it cannot predict or control every response, and that it is likely the chatbot will provide some inaccurate and erroneous results, among the most common concerns regarding the use of genAI technology.

To make the tool as useful, informative, and safe as possible the government said that experts on AI safety and safeguarding techniques at the AI Safety Institute (AISI) have been consulted on the work.

Following the first testing phase, guardrails have been added to help the chatbot detect which questions it should, and should not, answer. These include measures to prevent it responding to queries that might prompt an illegal answer, share sensitive financial information, or force the chatbot to take a political position.

The government issued chatbot might prove to be less useful than is being claimed, as research has shown that even the best AI models are prone to hallucinations, especially on niche topics, with a team from the US finding that popular LLMs struggle significantly with topics related to people and finance.

The new trial comes as the science secretary’s department looks to shape a new ‘digital centre’ of government to boost technology adoption across the public sector, taking a more experimental approach with emerging technology.

Recommended reading

• Scottish Government Announce £33m Digital Evidence Pilot
• Scots Police Receive Notice on Cloud System
• Comment | Realising Scotland’s Digital Ambitions

“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,” said Peter Kyle, science secretary.

“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. 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.” 

Despite those claims to be leading the charge, the government is slightly behind the times, with recent reports showing that nearly half of UK customer service teams already use AI-powered software, and cite cost savings, faster response times, and improved customer satisfaction as a result.

It might also be that the government is overestimating the appetite for yet another genAI customer service chatbot to add to the list. A report from AI business service platform ServiceNow revealed in August that consumers are actively avoiding chatbots and show a strong preference for human interaction, with four out of five people looking up tips and ‘hacks’ online to avoid speaking to a chatbot.