AI accusations in NZ politics - and what writers should take from it
After Chris Hipkins' "state of the nation" speech, Nicola Willis called it "jelly" and said ChatGPT could have written it. That kicked off a bigger question than point-scoring: how do you spot AI writing, and what does it mean for people who write for a living?
Here's what happened, what AI detectors actually said, and a simple playbook writers can use to keep their work sharp, original, and unmistakably human.
What actually happened
Hipkins delivered a broad speech with few firm promises. Willis mocked it as something an AI could churn out. When the text was run through an AI detector, the results were mixed across several political speeches-and revealing for anyone who writes.
What the detectors found
- Hipkins: GPTZero leaned human overall (about 79% "human"). Some parts read "robotic," especially a section about AI and productivity.
- Christopher Luxon: Detector said fully human.
- David Seymour: Detector said fully human.
- Nicola Willis: Detector flagged "a mix of AI and human" across her NZ Economic Forum speech, citing text that was very grammatically correct, formal, long, and impersonal. Phrases like "invest, build, and create opportunity" were marked as common and unoriginal.
Willis denied AI authorship. She said the speech was developed by staff using government material and confirmed Microsoft Copilot was used to check phrasing and grammar-not to write content.
The real lesson for writers
AI detectors don't prove authorship. They guess based on patterns. "Too clean," "too generic," and "too symmetrical" can trip alarms-ironically the same things editors reward.
If you're a speechwriter, journalist, or content lead, build a process that blends polish with unmistakable human voice. That's your moat.
How to keep your writing clearly human
- Open with a lived detail: a place, a number, a moment.
- Mix sentence lengths. Short. Then build. Then short again.
- Use concrete nouns and active verbs. Cut filler.
- Add one sharp opinion or personal stake per section.
- Name sources and include specific figures where possible.
- Swap clichés for fresh phrasing you'd actually say out loud.
Using AI without losing your voice
- Let AI be your line editor, not your ghostwriter: grammar, clarity passes, alt headlines.
- Keep the thinking, structure, and examples human. Your taste is the value.
- Fact-check manually. Tools can smooth grammar and still invent details.
- Disclose AI assistance when required by your publication or client.
A clean workflow you can copy
- Outline from research and interviews. Lock your POV and promise to the reader.
- Draft fast in your voice. Don't self-censor in the first pass.
- Run an AI edit for grammar and phrasing only. Reject anything that dulls tone.
- Add texture: names, dates, quotes, stats, and a line that only you would write.
- Read it aloud. If it feels stiff, it will read stiff.
- Optional: test with a detector, but never treat it as proof-just another signal.
What this means for teams
- Create a short policy: where AI is allowed (editing), where it isn't (analysis, claims, sensitive topics), and how to disclose.
- Track sources and drafts. If questioned, you can show your work.
- Coach juniors on voice. The goal isn't "perfect"-it's "distinct."
Resources
- GPTZero - an example of an AI-writing detector. Useful signal, not a verdict.
- Microsoft Copilot - grammar and phrasing assistance inside Office tools.
- AI for Writers - practical ways to use AI for drafting, editing, and ethical workflows.
- Office Tools - tips for integrating AI features like Copilot into your daily writing stack.
Bottom line
AI detectors flagged "polish," not proof. If you write for a living, keep your process human-led, use AI as a careful editor, and double down on specifics and voice. That's how your work stands up-to readers and to the tools.
Your membership also unlocks:
