Advertising veteran Jane Evans has launched Teaching The Machine, an initiative spotlighting how large language models erase and stereotype women. For marketers now shipping AI-generated content at scale, the bias in the tools is a brand problem.
The machine does not invent a worldview. It repeats the one it was trained on, and ships it into your brand at scale.
Advertising figure Jane Evans has formally launched Teaching The Machine, an initiative to spotlight how large language models erase and stereotype women. Evans has spent a career on the visibility of women in advertising, and her argument is that the same blind spots are now baked into the tools writing a growing share of the world's content.
This is not a fringe worry. UNESCO research found large language models regularly associate women with words like home, family and children, and men with business, executive, salary and career. The model is not neutral. It reflects the material it learned from, and that material carries decades of skew.
For marketers the point is uncomfortable and practical. If AI is generating your copy, your images and your product descriptions, it is also carrying its defaults into your brand, unless someone is checking.
Why it matters
More brands are producing content with AI at volume, and volume is exactly where bias goes unnoticed. One ad gets a human eye. A thousand auto-generated variations do not. The stereotypes ride along in the defaults, in who the model pictures as the boss, who it pictures at home, whose story it tells and whose it skips.
That is a brand risk and a market risk. Content that quietly narrows who your brand speaks to, or reproduces a tired stereotype, alienates the customers you did not mean to lose. In a country as diverse as Australia, a tool trained mostly on someone else's material will not represent your audience unless you make it.
UNESCO found LLMs regularly link women to words like home, family and children, and men to business, executive and career. That skew ships into AI-generated content by default.
What to do about it
Look at who your AI content pictures. Run your usual prompts and check who shows up as the leader, the expert, the customer. The defaults tell you what the tool assumes.
Write representation into the brief. If you leave it to the model, you get its bias. Specify who your brand speaks to and for.
Keep a human review on anything public. The point of a review step is to catch what the tool does not know it is doing. Do not automate that away.
Check how AI describes your category. If the machine paints your whole sector with a stereotype, that is both a risk and an opening to stand apart.
Treat this as brand safety, not politics. Content that misrepresents your audience costs you customers. That is a commercial problem before it is anything else.
The tools are useful. They are also not neutral. Ship what they produce without looking, and you are handing your brand voice to defaults you never chose.