A government report, Australia, October 2025
A consultancy delivered an AUD 440,000 report to a federal department containing fabricated academic references and an invented quotation from a judgment. The final instalment — A$97,000 — was refunded.
Our position
Anyone can generate copy. The question is whether you can publish it.
This is not a romantic argument about craft. It is about what happens when something goes out with your name on it and nobody checked.
A machine will give you a draft in seconds. It will also, without warning, give you a statistic that was never published, a product name that does not exist, and a citation to a paper nobody wrote. It does this fluently and with total confidence, which is precisely the problem: the failures do not look like failures.
For a lot of content that is a survivable risk. For a press release under embargo, a claim about what an ingredient does to skin, a sustainability report your investors will read, or an internal message from a chief executive — it is not.
Two things that actually happened
A syndicated newspaper insert, United States, May 2025
A summer reading list recommended books that do not exist, attributed to real authors. It came from one freelancer's unchecked AI use and ran across several major titles before anyone noticed.
Note the shape of the second one. The failure did not happen at the publisher. It happened at a freelancer, upstream, and travelled. That is exactly the shape of an agency supply chain — which is why our position is about the chain, not about the tools.
What we commit to
- 01
We tell you, in writing, before you sign
Where a tool was used and what it did. The industry standard here is consent, not notice — the IPA and ISBA principles say neither party should put AI-generated content in front of the other without agreement, and that is now written into the standard creative services framework. We work to it.
- 02
A named person writes it, and a second one reads it
Not a pool, not a rotation. The same writer on your account for years, and a review by someone who did not write it. For translated work that second read is a native speaker, against the source.
- 03
Every claim is traceable to something you gave us
In technical and scientific work the sentence has to survive a regulatory read. We write from your research and we can show you where each claim came from.
- 04
We will not put a detector score in a warranty
AI detection tools return 5–12% false positives on non-native English and on heavily edited prose. Any supplier warranting a detector score is warranting a coin toss. We warrant the process instead, because that is the thing we control.
What we are not saying
We are not saying we never touch these tools. We use them the way any competent studio uses a tool — research, first-pass structure, terminology checks — and we tell you when we have. What we do not do is hand you a draft nobody wrote and let you find out later.
There is a legal wrinkle worth knowing about too, because it affects what you actually own. The US Copyright Office holds that prompts alone confer no human authorship, so copyright may not subsist in a purely machine-generated passage at all. An assignment clause over something uncopyrightable transfers nothing. Our terms assign what can be assigned and grant an irrevocable licence over the rest — which is the drafting the market is converging on, and most suppliers have not caught up with it.
Why this is worth paying for
Because the alternative has a cost you only see once. A refunded fee, a correction notice, a regulator's letter, an apology from a chief executive who was told the copy had been checked. Set against that, the difference between a machine draft and work a person is willing to put their name to is not really a line item.
If this is why you are looking
Some of the people reading this moved their content to AI a year or two ago and are now quietly cleaning up after it. That is a conversation we have often, and there is no awkwardness in it.
Estimate within one working day.