Practical AI for HR: Real-World Examples in Microsoft 365 and SharePoint
"I would say use the metaphor of intern for Copilot, for AI in general."
What it was about
Copilot and AI in Microsoft 365 only deliver value when HR professionals prompt deliberately, treating the AI like a well-briefed intern rather than a search engine. It also requires IT to have the right governance and sensitivity controls, like Microsoft Purview, in place to protect sensitive HR data.
By the numbers
Key notes
- Use the CTAT framework (Context, Task, output format, Tone) to structure prompts instead of typing vague keyword-style queries.
- Treat Copilot like an intern you're mentoring, not an all-knowing expert or a search engine. Never accept its first answer without refining it.
- Ask Copilot to generate a reusable prompt from a long conversation so the same result can be reproduced later or shared with teammates.
Take this back Monday
Ask IT whether Microsoft Purview sensitivity labels are on before letting anyone use Copilot on FMLA or other sensitive HR files.
Copilot only works if you prompt it like a well-briefed intern, not a search engine, and never trust its first answer.
Watch out for
- Treating Copilot like a search engine by typing bare keywords instead of giving it context, a task, and a desired output format.
- Accepting Copilot's first answer as final without verifying it, since answers sound confident even when they may be incomplete or hallucinated.
- Mixing unrelated topics in a single conversation thread, which causes 'drift' where old (possibly wrong) context bleeds into new answers.
Jarrod Edge has spent over 25 years as a Microsoft Certified Trainer, delivering SharePoint and Microsoft 365 training around the world.
