Examples and context
Show, don't tell: few-shot examples for format and voice, and how to feed the model the context it can't know.
Two upgrades take prompts from good to excellent: showing examples of what you want, and feeding context the model can't know. Both exploit the same fact — the model is a pattern-continuation engine, so the best way to get a pattern is to start it.
Examples: the fastest way to transfer 'how we do it here'
Some things are hard to describe but easy to demonstrate — your team's status-update style, how your company writes customer apologies, the exact table format your boss likes. Don't describe it. Paste one or two real examples and say match this.
Here are two product descriptions in our house style: "The Juniper Throw — wool that earns its keep. Woven in Maine, warm without weight, and guaranteed for ten winters." "The Alder Bench — four boards, no nonsense. Solid oak, hand-finished, holds two adults and every coat in the hallway." Write three more in exactly this style for: a ceramic mixing bowl, a linen curtain set, and a cast-iron trivet.
Notice you never had to define the style ('short, wry, concrete benefits, no adjectives-for-adjectives'-sake') — the examples carried it. Two examples is usually enough; five rarely beats three.
Context: close the knowledge gap
The model knows the world up to its training cutoff but nothing about your world. The single biggest quality lever in workplace prompting is pasting in the relevant context: the email thread you're replying to, the policy you're applying, the notes you're summarizing, last month's version of the report.
- Replying? Paste the whole thread, not your summary of it.
- Applying a rule? Paste the actual policy text — don't trust the model's memory of what 'standard PTO policy' says.
- Continuing a pattern? Paste the last version. 'Here is last week's report. Write this week's in identical structure. This week's facts: …'
Everything you paste leaves your machine and lands on the AI provider's servers. Before pasting customer data, financials, or anything under NDA, know your company's rules — Module 4 covers exactly what's safe to share and how to redact. When in doubt, redact names and numbers first.
Examples and context scale into a full discipline — more advanced techniques for feeding AI the right context, reliably and at scale. That's the Prompt Engineering & Context Design course — the next step in most of our tracks. This lesson is the 20% that delivers 80% of the value.