An aide to Andy Burnham was photographed seeking ChatGPT’s feedback on a Welsh Labour conference speech, prompting the chatbot to suggest making a passage about industrial communities “a little more emotive”.

Number 10 defended the use of the tool, saying the aide had submitted an existing draft for feedback. The chatbot’s suggestions were reportedly not used in the final speech, delivered on Saturday.

The photograph appears to show the aide drinking Prosecco while working on an afternoon train from London Euston to Holyhead. ChatGPT reportedly compared two passages and favoured the second because it moved from miners to their families and the wider community, describing how industrial employment created bonds beyond the workplace.

A Number 10 spokesperson said: “The PM writes most of his own speeches. An aide sought feedback on a speech draft that had already been written.”

The speech’s subject gave the episode its political edge. Burnham spoke about industry’s role in giving people a “deep sense of belonging”, describing Welsh workplaces that “created a way of life, the social infrastructure that we took for granted, the working men’s clubs, the sports teams, the choirs, the trade unions”.

Criticising Margaret Thatcher’s legacy, he argued that decisions in the 1980s had shattered communities. He linked the privatisation of economic power and centralisation of political power to a loss of local agency, saying communities had spent 40 years without the power they needed to recover from the closure of mines, factories and steelworks.

Reform UK chairman Lee Anderson mocked the apparent drink, joking that writing the Prime Minister’s speeches must be “thirsty work”. He added: “AI Andy’s man-of-the-people act has been exposed as a sham.”

Anderson’s attack concerns the authenticity of Burnham’s message; Number 10’s defence concerns how the speech was prepared. Those are distinct arguments. Seeking automated feedback does not establish that the chatbot wrote the address, and the reported omission of its suggestions limits what can be concluded about its influence on the speech delivered.