Writing an AI policy people actually follow

Most AI policies are written to protect the company, not to help the people who use the tools. They end up long, vague and full of “don’t”, so people either ignore them or stop experimenting altogether.

Why most AI policies fail

A policy only works if people read it, understand it and can apply it on a Tuesday afternoon with a deadline. Anything that needs a lawyer to interpret will quietly be skipped.

Three principles that worked

I tried to write the policy for the colleague who simply wants to get their work done:

• Say clearly which tools are approved, and why.
• Show which data must never go into an AI tool, with real examples.
• Keep it short enough to read in five minutes.

A policy nobody reads protects nobody.

Training beats documents

The document is only the starting point. What really changed behaviour were short, practical sessions: real prompts, real mistakes and real questions from the team.