Trev Garrett against Stuff

Stuff published an article on February 26, 2026 titled Nicola Willis says Chris Hipkins’ big speech could have been AI – but is AI writing her speeches?

The story led with Finance Minister Nicola Willis’ comment that Labour leader Chris Hipkins speech was so unsubstantial it could have been written by AI.

A Stuff reporter ran the speech through an AI tool which gave it a human score of 79 percent. The same tool assessed major recent speeches by Christopher Luxon, David Seymour to be all human.

However, a speech by Nicola Willis was assessed as a mix of AI and human and the story reported the tool’s reasoning for that was set out in some detail. Asked to comment she said the speech was written by real people, but Microsoft Copilot was used to edit it.

Trev Garrett complained that the article was unbalanced and appeared to be written by an Opposition press secretary rather than an independent journalist. He said the same AI analysis scores for the Hipkins and Willis speeches should have been used so a fair comparison could be made.

The failure to use comparable figures meant the article was intentionally misleading and in breach of Principle (1) Accuracy, Fairness ad Balance.

In response Stuff said it used the same metric on all the speeches.

“As reported, the speeches by Luxon and Seymour that we analysed showed no signs of AI, while the speeches by Willis and Hipkins showed some signs.

“While we prominently stated that such tools are not 100 percent accurate, both Willis and Hipkins were given the opportunity to respond to the GPT's findings and did so.

“We disagree that the article reads like it was "drafted by an opposition press secretary". As above, we reported that the speeches by the current PM and deputy PM were found to have been "all human".

The Media Council saw this as a fairly straightforward piece of political reporting in which Stuff was prompted by Ms Willis’ comment to check if an AI tool could determine whether politicians’ speeches were written by humans.

This was an amusing and interesting story about what was found when speeches were tested using a far from proven artificial intelligence checker.

Stuff chose to report the “human score” figure for Mr Hipkins’ speech but not the equivalent figure for Ms Willis’ speech. It is not clear why this was done. But more importantly it reported Hipkins and Willis both appeared to have used AI.

They were then both given the opportunity to explain and take a swipe at their opponent. But having raised the issue first, there was always a risk for Ms Willis that her comment could rebound on her.

Politicians levelling charges at their opponents run the risk that tables are turned, and indeed we note that this complaint is not from Ms Willis herself.

There was no evidence to show the reporting was inaccurate, unfair or unbalanced.

Decision:  No grounds to proceed.