AI, Bias and the Power of Questions: How to Get Better Answers With Smart Prompts

How objective are AI answers really? 

The sobering truth: Not at all. Every answer is based on a prompt – and every prompt already carries a perspective. This is where the problem begins. But also the solution.

In our new blog article (and white paper), we show that to use AI effectively, you need to be able to ask questions – smart, conscious, bias-sensitive questions. Or, in short: a good prompt is the best bias filter.

Why bias cannot simply be “trained away”

Bias in AI does not arise at the output stage, but deep down inside: through the training data, through the model's objective – and, last but not least, through the questions we ask. This shows time and again that:

👉 AI does not reflect the world – it condenses probabilities.

Whether you work in UX research, communication or strategy, if you only ask plausible questions, you will usually get a very pleasing but often too smooth answer. The result? Reproduction instead of reflection.

A systematic look at bias: 13 types, many pitfalls 

In the white paper, you will find a tabular overview of 13 types of bias – from training data bias to prompt bias to persona bias. For each type, you will find:

• a clear description,

• a practical example,

• an avoidance strategy

• and a specific prompt for application.

Here's a little preview:

🧠 Prompt bias

Description: The wording of the question itself suggests a particular direction.

Example: ‘Why doesn't gender mainstreaming work?’

Better prompt: ‘Please check my question for hidden assumptions.’

🌍 Representation Bias

Description: Groups are portrayed in a stereotypical manner or not at all.

Example: Trans issues are only addressed from a medical or legal perspective.

Better prompt: ‘What do queer or neurodivergent people have to say about this?’

🗓️ Temporal bias

Description: AI knows nothing about current developments.

Example: The legal situation regarding trans rights is from 2022.

Better prompt: ‘Please state your knowledge date – what could have happened since then?’

Your cheat sheet: 7 strategies for bias-sensitive prompting

Here are seven universally applicable strategies you can use to make AI responses more informed, inclusive and reflective:

1. Use neutral wording:

→ ‘How well is UX typically anchored in companies – and what could be the reasons for this?’

2. Demand diversity of perspectives:

   → ‘Please give me three perspectives: conservative, progressive and activist.’

3. Actively seek opposing opinions:

   → ‘What are the arguments against my thesis – even if it sounds plausible?’

4. Be aware of cultural differences:

   → ‘How would this be viewed in Germany?’

5. Encourage self-reflection:

   → ‘What implicit assumptions does my question contain?’

6. Set roles consciously:

   → ‘Answer as an investigative journalist.’

7. Address bias directly:

   → ‘What biases could arise in this topic – at the level of data, language, perspectives and goals?’

Three application scenarios – and suitable prompts

We have selected three typical use cases and formulated suitable prompt templates:

🔍 Interview guidelines (UX/research/transformation)

Objective: Identify leading questions and bias

👉 Prompt:

💬 Corporate communication / LinkedIn / Thought leadership

Objective: Reflect on narratives instead of reproducing them

👉 Prompt:

📈 Strategy / Policy / Management Consulting

Goal: Think from multiple perspectives

👉 Prompt:

Conclusion: Better questions lead to smarter AI 

There is no such thing as perfect neutrality – but there are better questions.

Bias in AI cannot be eliminated, but its impact can be significantly reduced with good prompting. The ability to not only ask questions, but to ask them consciously, is key.

Because every good answer starts with a good question. And that question comes from you.

💌 Not enough? Then read on – in our newsletter. It comes four times a year. Sticks in your mind longer. To subscribe: https://www.uintent.com/newsletter