Report finds AI ‘widespread’ in newsrooms. Here’s how I use it
Tuesday, 3 February 2026
OPINION AI tools are now widespread in the country’s newsrooms, but most people won’t know how AI is used to assist in news production, according to a study by Auckland University of Technology.
Researcher Merja Myllylahti said that while mainstream media outlets, including Stuff (owner of The Post), NZME, RNZ and TVNZ all had principles and ethics that emphasised “human oversight and transparency”, these weren’t prominently displayed.
“News organisations don’t disclose how exactly AI is used in all content production. Many editors feel that ‘the ship has sailed’ in terms of tagging or labelling AI content,” she adds.
One of her conclusions: New Zealanders are consuming more AI-assisted news than they realise. I’ll admit that sounds bad.
Read more:
Why the AI age could supercharge New Zealand’s leisure economy
Victoria University breaks into global top 50 for artificial intelligence
My first thought is that many media organisations are, in my experience, a lot less hierarchical than they used to be.
Organising hundreds of reporters to produce a bulky broadsheet newspaper from scratch every 24 hours required an almost military chain of command.
Journalists did what they were told and there were layers of management to tell them what to do.
That has loosened a lot as print pages have thinned, anytime and all-day publishing on the internet has become the norm, resources have moved from tail to teeth, and directions from above have become few and far between.
Individual journalists are more empowered than they used to be to determine their own reporting agenda and figure out themselves how to do the job and what tools to use.
That means I can’t pretend to know how most of my colleagues are using AI.
But here’s what I do, and where I think the dilemmas may lie.
Myllylahti says AI is “used to assist in searching and researching for news stories, summarising stories of large documents, transcribing interviews, checking spelling and grammar, and generating audio/video from text or vice versa”.
It is the least controversial tools in that list that are currently the most useful, and I suspect the most widespread.
The AI tool I value the most is a tool called Otter that can be used to record and transcribe interviews.
The transcriptions aren’t completely reliable, but the text is searchable.
The value of the tool is that it lets reporters quickly find the quotes they know are important during a conversation and report them accurately — without going through a whole tape recording or spending an interview double-tasking by asking questions and scribbling down shorthand (which I don’t have).
Once I’ve written a story, I’ll usually run it through a free AI engine and ask it to list any errors.
The main purpose is to check for spelling mistakes, missing words (which are my personal forte) and grammatical issues. But occasionally it will also call out a literal error such as incorrect date or misspelt name, usually with a lot of false positives.
Technically, of course, the spell-check itself won’t be ‘AI’, any more than the ‘chatbots’ of the early 2000s were really AI.
Some technical functions are still more efficiently done using a rules-based engine — in this case this will be an AI engine just checking words against a dictionary.
I can confirm that another common use of AI is to make headline suggestions.
Occasionally when writing columns, I will ask an AI engine to check them for any political bias. I’ll admit to being surprised by the sophistication of some insights, which I’ll take or leave — usually leave.
Although Myllylahti puts it first on her list, I rarely use AI for research, with a few exceptions.
If an organisation like the Treasury puts out a 300-page report and there is a need for a story in an hour, I might use an AI engine to compare it with a draft report I had already written about to try to get a quick sense on whether much of importance had changed.
But only as a back-up to my own skim-reading. Myllylahti puts “summarising stories of large documents” second on her list of uses for AI, but I don’t think the reliability is there yet.
On a couple of occasions — as an experiment — I have used AI to do a first draft of a story where that has been largely from a single source; in one case a long select committee hearing.
On both occasions I have needed to totally rewrite them.
I reckon it can be a useful technique to get some ideas on story structure on the few occasions where one doesn’t leap out from the material, or as an extra check on not missing an important angle, but I’m not convinced it’s a time-saver.
In the same vein, I have sometimes asked an AI engine to check whether I have missed any major angles in a story I have completed, from the transcripts of interviews and documents.
Fortunately the answer, to the best of my recollection, has always been ‘no’, but it could be another quality control check.
A bottom line with all these uses of AI is that it is the journalist’s byline on the story and what it says is their responsibility.
That is a better protection than it might seem. As a reporter, individual reputation — with the public, colleagues and sources — is everything.
Myllylahti does make clear, but perhaps doesn’t stress enough, that — with a few exceptions, mostly in the specialist business press — media organisations are not using AI to actually generate stories from material.
It’s good not to be complacent though. There could be trouble ahead.
The thorny question, I believe, is whether it would ever be appropriate to use AI as a tool to help decide what topics to report on, or what questions to ask of people.
At its heart, reporting is a simple job. We do the questions. Someone else does the answers. And I should add we have no magic powers to enable us to elicit information. We are just members of the public who are lucky enough to get paid to be a pain in the neck.
The skill, such as it is, is asking the right questions of the right people, at the right time and in the right way.
Maybe technology could help us do that better.
But is that something we could ever hand over to AI, or ask it to assist with? Would we still be using AI as a tool, or relinquishing control? We are all learning at this. All I can tell you is that’s my own red line.