Understanding the bias in AI algorithms
AI is already reshaping the marketing industry, from how content is created and media is bought, to how decisions are made. As adoption accelerates, the opportunity, and the responsibility, is to ensure these systems support more inclusive and representative creative work, rather than reinforcing stereotypes or exclusions.
UN Women warns that the technology is reproducing bias and stereotypes inherent in its training data. AI does not start from a neutral place. Generative tools work by predicting what is most likely, so they default to the most common depictions in the data they were trained on, and in the references, examples and prompts we give them. Without deliberate human oversight, those defaults get reproduced at scale.
A study of 133 AI systems found that 44 per cent demonstrated gender bias, while more than a quarter showed both gender and racial bias
Of 138 countries assessed worldwide, only 24 referred to gender in their national AI strategies, and just 18 included substantive gender-responsive measures
According to UN Women data, nearly one in four surveyed women human rights defenders, activists and journalists reported experiencing AI-assisted online violence, twelve per cent said personal images had been shared without their consent, while six per cent reported being targeted by deepfakes or manipulated images and videos
Women account for only 30 per cent of the global AI workforce, based on data from the International Labour Organization
The Unstereotype Alliance has published the 3Cs Playbook: Skills for Inclusive AI, a plug-and-play resource for marketers navigating GenAI. As members of the GenAI Working Group, we are proud to have contributed to its development.
Built around Curate, Craft and Control, the Playbook gives teams a simple, repeatable framework to recognize where bias can enter the process, challenge stereotypical defaults, strengthen human oversight and build more inclusive outputs from the start. The skills apply wherever GenAI is used, from research and insight generation through to creative ideation, asset creation and media buying.
Curate. Avoid the biases that can come from the data AI draws from. So think about both the data the tool has been trained on and the data you provide: references, examples and audience insight.
Craft inclusive prompts. Avoid bias in the way we ask AI to support or create work. Be specific about context, audience, language, representation goals and watch-outs. The tool will not infer inclusion on its own.
Control stays human. The tool can support the process, but people must stay accountable for the final judgement, reviewing outputs for bias, accuracy, authenticity and brand fit.
How to use the Playbook
The playbook is designed to be worked through once and then returned to.
Start with the risks and the benefits. The opening section explains how bias enters GenAI outputs and what the technology makes possible. It includes a risk framework showing how different uses carry different levels of brand risk, so you can judge where extra care is needed on your own projects.
Then take the skills one at a time. Each of the 3Cs is introduced with an explanation of the skill, the benefits of applying it, and a set of questions to ask yourself as you do. The questions let you upskill as you go and experiment on live work.
See what good looks like. Each skill comes with a case study showing the guidance applied in practice.
Bring the tools into your workflow. Checklists and other practical tools are included so you can start applying the 3Cs to the way your team already works.
Download the 3Cs Playbook: Skills for Inclusive AI
Comments