Support us
Editorial desks across Australia are quietly adapting to a new kind of inbox noise. Beyond ordinary press releases and tip-offs, senior editors now wade through submissions written or entirely fabricated by artificial intelligence. The shift has been gradual but persistent, and it is reshaping how newsrooms approach verification before commissioning a freelance piece.
In Sydney and Melbourne newsrooms, the problem is no longer limited to opinion columns. Investigative desks, regional correspondents in Queensland, and community outlets in Perth have all flagged pitches containing plausible-sounding sources that turn out to be invented. The Free Press Foundation, an independent nonprofit supporting rigorous journalism on Ukraine and global affairs, has responded by designing hands-on training for editors. More about that work is available on the foundation's homepage.
The programme draws on the foundation's long experience in conflict-affected regions, where misinformation campaigns are a daily hazard. By bringing those field habits to newsrooms further afield, the trainers aim to give Australian editors a sharper instinct for distinguishing the synthetic from the genuinely worth publishing.
The training itself is layered. It blends linguistic analysis with technical forensics, ethical standards, and the trauma-aware reporting that protects both subjects and readers. Each component is designed to feel less like a lecture and more like a routine editorial check.
Editors in the workshops are shown samples of pitches generated by large language tools. The text often reads cleanly, with confident headlines and tidy paragraph breaks. Sources are named, dates are supplied, and the structure mimics what a seasoned freelancer would send to a commissioning editor.
What gives these drafts away is usually in the details. A Brisbane-based editor in one training cohort noticed that a pitch about a fictional council vote named streets that did not exist in the suburb. Another submission claimed to quote a rugby league player who had retired four years earlier. The workshops walk editors through the habit of running every named fact through a quick sanity check.
Beyond factual slips, synthetic prose carries a particular rhythm. Sentences tend toward even length, transitions feel smoothed over, and the language is unusually neutral. There is often a faint bureaucratic tone, even in features that should carry a human voice. Editors are trained to read for this cadence before they read for content.
Trainers also flag how machine-written text handles local idiom. Australian English is full of small variations, from colloquial spellings to phrasing that differs from North American usage. A piece that mixes these up, or that feels slightly translated, is worth a closer look.
The training is delivered as a series of short drills, each mimicking the pace of a real editorial meeting. Editors are given a stack of submissions and asked to flag anything suspicious within a tight window. Trainers then walk the group through the reasoning behind each call.
Sessions have been held with regional outlets in Townsville, public-interest teams in Adelaide, and investigative editors at major city titles. The format is deliberately fast because the working environment is fast, and the lessons stick better under pressure.
Technical forensics form the backbone of the verification stage. Editors learn to run reverse image searches, examine metadata for signs of manipulation, and cross-reference named sources against public records. This is where many synthetic pitches collapse under their own weight.
The drills also surface the stranger corners of automated content. Trainers use examples such as the San Diego weather city comparison page, which illustrates how quickly machines fill content gaps with loosely-related comparisons that no human editor would commission. Spotting this kind of filler is a skill in itself.
A piece that handles grief, displacement, or destruction with care is rarely produced by a machine. The foundation's training spends time on covering rebuilding without ignoring trauma, because genuine reporters working with affected communities write with an attention that synthetic text cannot replicate. Editors are encouraged to use this emotional register as one of several quality checks.
Australian coverage of recent bushfires and flood recoveries has shown how nuanced this work can be. When a submission captures that nuance, a real journalist spent time listening. When it does not, the gap is often the first hint that no one was ever on the ground.
The training ties verification habits back to formal frameworks, including guidelines from the Australian Press Council and standards observed by outlets such as the ABC. Editors learn to document their reasoning so that any decision to spike a submission can be defended later if challenged by a freelancer or a source.
Transparency matters as much at this stage as it does in published corrections. A newsroom that openly explains why a piece was rejected signals to readers that its filter is active. That visible discipline is part of why public trust in serious outlets has held up even as the wider information environment has become murkier.
Synthetic content keeps evolving, and the foundation's training is built to evolve with it. Trainers refresh drills each quarter, drawing on fresh examples drawn from pitch inboxes and partner newsrooms. The aim is not paranoia but preparedness, a working rhythm in which verification becomes as automatic as spell-checking once did.
By sharing these habits across editorial teams in Australia and beyond, the foundation hopes that the next wave of machine-generated submissions meets a newsroom that is ready for them.
