Five Prompt Patterns That Cover Most Office Work
There is an entire cottage industry devoted to making prompting complicated: courses, hundred-item prompt libraries, threads promising the secret phrases that unlock AI. You need almost none of it.
Here is what actually matters. Output quality tracks request quality almost exactly, and good requests follow a small number of patterns. Office work runs on five of them. Five is a number a team can learn in one session and actually remember on a Tuesday, which is more than any prompt library has ever managed.
First, the four ingredients
Before the patterns, the parts. Every request that works contains some mix of four things:
Goal. What you want produced. Be concrete: a one-page summary, a reply that declines politely, a table with three columns.
Context. What the AI cannot know. The audience, the situation, the facts that live in your head or in a phone call it wasn’t on. This is the ingredient people skip, and its absence explains most disappointing output.
Constraints. Length, tone, what to include, what to leave out.
Source. What material to work from: this thread, that document, last Tuesday’s meeting. Grounding a request in specific content is what separates a paid assistant from a generic chatbot.
When output disappoints, one of those four is usually missing. That alone is a useful diagnostic.
- Catch Me Up
Compress a pile of communication into orientation plus obligations.
The skeleton: “Summarize [this thread or channel or meeting]. What was decided, what’s still open, and what needs something from me?”
That last clause is what elevates this from a book report into a working tool. A summary is nice. Knowing what you owe people is actionable. Make “what needs something from me” a standard part of your catch-up prompt and the assistant becomes a commitment detector.
- Draft From Intent
Turn a goal plus your private facts into a first draft.
The skeleton: “Draft [thing] for [audience]. Goal: [outcome]. Context: [the facts only you know]. Tone: [tone]. Keep it [length].”
The context line is the difference between a draft you can send after light edits and a draft that reads like a template. The assistant was not on yesterday’s call. Tell it what happened.
Compare “write an email to the team about the deadline,” which produces something serviceable and hollow, against “draft an email to the project team. Goal: move the deadline from the 15th to the 22nd without triggering panic. Context: the client requested the change, this is good news for QA, and the QA lead already knows. Tone: calm, brief.” The second one comes back close to sendable.
- Transform
Content exists in one shape and you need it in another.
The skeleton: “Turn [this] into [that], keeping [what matters].”
Notes into a status table. A rambling document into an executive summary. Bullet points into prose, prose into bullets, meeting chaos into an agenda for next time. Transformation is mechanically easy for AI and mechanically tedious for humans, which gives this pattern the best effort-to-payoff ratio of the five.
One upgrade worth adopting: ask for the gaps too. “Anything unclear goes in a questions list below the table” turns a formatting task into a review of what your source material failed to capture.
- Interrogate
Ask questions of content instead of reading all of it.
The skeleton: “Based on [source]: [your question].” And the more valuable variant: “What’s missing from this? What would a skeptical reader challenge?”
This is the pattern that turns an assistant from a producer into a second pair of eyes. What are the termination terms in this contract? Which action items from last month are still unresolved? What am I asserting in this proposal without evidence?
Verification stays your job. Interrogation gets you oriented in a forty-page document in two minutes so you can spend your actual attention on the four paragraphs that matter.
- Coach Me
Get feedback on your work before the audience does.
The skeleton: “Review this [thing]. How will [audience] receive it? What would you strengthen, soften, or cut?”
This is the least used pattern and the one I would defend hardest. It costs fifteen seconds, applies to everything, and fills a permanent gap in professional life: the absence of a candid reader before the moment of no return. It also compounds. Ask “what would you cut” enough times and you start pre-cutting, which means the AI improved your writing by critique rather than by replacement.
Teaching this to a team
Two things matter more than the patterns themselves.
Teach patterns, not prompts. A prompt library goes stale and nobody browses it under deadline. Five named patterns fit on one page and generate infinite specific prompts on demand.
Follow-up is not failure. The first response is a first response. “Shorter,” “more formal,” “now as a table,” “you missed the budget angle” is how the tool is meant to be used. New users often quit after one mediocre answer, acting as graders when they should be editors.
None of this changes what the assistant can do. It changes how often you get it, which turns out to be most of the gap between teams that find AI useful and teams that quietly stopped opening it. Worth knowing which zone your actual problem sits in first, though: some of what people try to prompt their way through is not an assistant problem at all, and the workflows most worth changing are usually the ones nobody is prompting yet.
These five patterns are one chapter of a free guide I wrote for teams adopting Copilot. Get in touch if you want the whole thing, or help teaching it.
Rosemarie Withee has spent thirteen years helping operations teams get real work out of their software, first Microsoft 365, now AI. She’s written six books for Wiley and builds AI products at Portal Integrators.