What we will cover
The anatomy of a strong prompt
Every effective prompt carry four ingredients: a goal, enough context, a source, and a clear expectation of the output. The goal says what you want done. The context explains why and for whom. The source points Copilot at the specific file, meeting, or thread it should reason over. The expectation sets the shape of the answer, whether that is a length, a tone or a format. We will take a handful of weak prompts and rebuild them ingredient by ingredient so attendees can see exactly what each one contributes to the result.
Be specific rather than generic
“Summarize this” is a wish. “Summarize this forty-page proposal into five bullets an executive can read in a minute, and flag anything that affects pricing” is an instruction. Most of the frustration people report with Copilot traces back to prompts that leave the audience, the length, and the format unstated, which forces the model to guess. We will look at paired examples side by side and discuss how naming the reader of the output changes the writing that comes back.
Grounding Copilot in the right source
Copilot's real advantage is that it can reason over your organization's own content, but it needs pointing. Referencing a specific document, meeting or colleague inside the prompt is what turns a plausible answer into a verifiable one. We will practice referencing files and meetings directly and talk about why a grounded prompt is also an auditable one: you can trace the answer back to the material it came from.