Erica Farmer, AI and Future Skills Speaker, Trainer and Author
AI is not neutral; it learns from data created by humans. And humans are wonderfully intelligent, creative, flawed and biased. So as a result, AI models can inherit and amplify assumptions and blind spots that already exist in society.
One of the clearest examples is gender bias and the research consistently shows that women are less likely to adopt AI tools than men, with studies often reporting a gap of around 20–30%. This is important to know because the people using AI influence the data, examples, feedback and use cases that shape future AI systems.
But bias can affect age, race, disability, socioeconomic background, geography, education and countless other factors. This means better AI requires broader participation and better prompting from all of us.
Start with Persona: The Most Powerful Part of the Prompt
Many people jump straight into asking AI a question. Instead, tell it who you want it to be first, including role, personality, communication style and experience.
The Persona element of the Persona-Context-Outcome (PCO) framework helps AI adopt a perspective that encourages challenge rather than compliance.
So, let’s compare these two prompts:
Weak Prompt:
“Write me a hiring advert for a sales manager.”
Stronger Prompt:
“You are an inclusion specialist, occupational psychologist and recruitment expert. Your role is to identify potential bias, challenge assumptions and create inclusive hiring content for a sales manager role in the UK which will attract the widest pool of talent.”
The second prompt immediately increases the likelihood that AI will check for biased language before producing an answer.
Ask AI to Audit Its Own Thinking
Then try asking AI to critique its own response by adding:
“Before providing your final answer, identify any assumptions, stereotypes or biases that may exist in your recommendation.”
Or:
“Review your response from the perspective of someone who may have been underrepresented in AI data training.”
This creates a second layer of thinking rather than accepting the first answer generated.
Use Multiple Perspectives to Challenge AI bias
Bias often appears when only one viewpoint is considered, so therefore a useful PCO prompt might look like this:
Persona:
“You are a diverse panel consisting of an HR Director, a disability advocate, a young employee, an experienced worker over 60, a team member diagnosed with ADHD and a data ethics specialist.”
Context:
“We are designing an AI adoption strategy for a large organisation in the UK. The organisation has 10,000 people and operates in the car manufacturing sector.”
Outcome:
“Identify potential blind spots, unintended consequences and groups who may be disadvantaged by this approach and suggest alternative wording, which is more aligned to our values and ethics.”
This technique often surfaces risks that would otherwise remain hidden.
Ask the AI What Might Be Missing
One of my favourite prompts to use is incredibly simple:
“What perspectives, voices or experiences might be missing from this analysis?”
And as a neurodivergent person myself, I wish more people would use something like this. This works particularly well when reviewing recruitment processes, leadership communications, learning programmes, employee surveys, change management plans and diversity and inclusion initiatives.
AI won’t always be right, but it can help highlight areas worthy of further investigation and raise your attention to potential issues that you can nip in the bud. But remember, bias isn’t always hidden in statistics; it appears in language. So, when creating content, ask:
“Review this document for potentially exclusionary, gendered or stereotypical language. Suggest more inclusive alternatives and explain why.”
This is particularly valuable for job adverts, policies and learning content.
Don’t Let AI Mark Its Own Homework
Here’s the critical thinking part. AI can help identify bias, but it can’t be the final judge. You must always be accountable for your work with AI and have a ‘human in the loop’ process. Always ask:
- What evidence supports this answer?
- What assumptions is the AI making?
- Who benefits from this recommendation (and who doesn’t)?
- What real-world data validates this conclusion?
Critical thinking remains one of the most important skills in the age of AI and we all need it on our development plans.
In Summary: Better Prompts mean Reduced AI Bias
By using strong personas, clear context and thoughtful outcomes, we can create better prompts, make better decisions and reduce the risk of accepting flawed outputs. AI is a powerful thinking partner, but only if we challenge its thinking as much as we challenge our own.
That’s where better prompting becomes better judgement.
Try the prompt framework I have shared today and think about the difference in outcomes you are getting, and more importantly how it is impacting others.
