Skip to course content
Free prompt workflow course

Prompting and AI Workflow Design

Bias, Unfair Assumptions, and Tone Risk

Unit ID: PAW-M11-U02 Estimated active time: 20-25 minutes

What this looks like in everyday work

Not dramatic. Usually small and cumulative:

Where it comes from

Patterns in the material these tools learned from, and patterns in your prompt.

Your prompt matters more than people expect. A request that says "write to the team about the change" will make assumptions about that team, and the fix is describing them.

Checks that catch it

FAIRNESS CHECKS

Swap test        Change the name, gender, age, or place. Does the tone
                 change? If yes, why?
Group test       Is any group described in different terms from another
                 doing the same thing?
Assumption test  What has this assumed about who the reader is?
Absence test     Who is not mentioned who should be?
Receiving test   How would this read to the person it is about, rather
                 than the person it is addressed to?
Example test     If examples are used, are they varied?

The swap test is the quickest and catches the most.

Tone risk in difficult messages

Rejections, complaints, apologies and decisions that affect someone are where tone matters most and where generated text is most likely to be subtly wrong — too breezy, too formal, or hedged in a way that reads as evasive.

For these, write the draft yourself and use AI for critique. This is the same rule as high-stakes writing in Module 9, for a different reason.

A person, not a checklist, decides this

Fairness checks find candidates. Whether something is actually unfair in your context is a human judgement, and often one that needs the affected person's perspective rather than a simulation of it.

Mini practice

Take a message that affects someone.

Run the swap test and the receiving test.

If either produced a change you would not defend, redraft it yourself.