How to Write AI Prompts
AI Prompt Examples: How to Write Prompts That Actually Work
AI prompt examples, in one sentence: a good prompt gives the AI your context, a role, a specific task, and a format for the answer—then treats the first response as a draft. This guide breaks down each part, rewrites real weak prompts into strong ones, and gives you examples to copy.
1-on-1 with Pete Enestrom, on your actual work
Written by Pete Enestrom
Yale & Columbia, ex-Microsoft & Intel — 1-on-1 AI coaching for executives
Your coach’s background
A prompt is simply the message you type into an AI tool—ChatGPT, Claude, Gemini, any of them. Everything in this guide applies to all of the major tools, because the skill of writing a good prompt transfers completely between them. I want to be direct about something up front: prompting is not a technical skill. There is no syntax to memorize and no code to learn. Writing a good AI prompt is the same skill as briefing a person well—saying who you are, what you're working on, what you need, and what a good result looks like. Most people were never taught to brief anyone, which is why most prompts underperform. That is fixable in an afternoon, and this page is the fix: the anatomy of a good prompt, real before-and-after rewrites, and examples you can copy word for word.
Here is the pattern I see with nearly every new coaching client. A smart, successful founder opens ChatGPT for the first time and types something like “write a follow-up email” or “summarize this.” The tool produces something generic, and the founder concludes—reasonably, from that evidence—that AI is overhyped. The problem was never the tool. A one-line prompt is the equivalent of handing a new employee a document and saying “handle this” with no context, no audience, and no definition of done. No good hire survives that briefing, and no AI does either. The executives who get startlingly good results from AI are not using secret prompts. They are giving the AI the same quality of briefing they would give a person they trust, and then treating the first answer as the start of a conversation instead of the end of one.
A note on using the examples below. Every prompt in this guide is written for real executive work—board materials, difficult emails, actual decisions—not party tricks. Anything in [brackets] is a placeholder for your own material: paste the actual document, thread, or notes. The prompts are the easy part to copy; your context is what makes the output worth reading.
What are the four parts of a good AI prompt?
Almost every strong prompt I teach contains the same four ingredients. You won't need all four every time—but when an answer disappoints you, one of them is almost always missing.
Context
- Who you are, what your company does, and what you're working on.
- The background a new hire would need: the situation, the stakes, the audience.
- Paste the source material—the thread, the draft, the numbers. Context in, quality out.
Role
- Who you want the AI to be: “a skeptical board member,” “an executive editor,” “a direct HR advisor.”
- Optional, but it changes the lens the answer comes through.
- Skip the role for simple tasks; use it whenever perspective matters.
Task
- One specific job, stated as a verb: draft, summarize, compare, critique, list.
- Hand over materials with the job—tasks with attachments beat questions.
- Two jobs in one message? Number them. Don't make the AI guess.
Format
- What the answer should look like: length, structure, tone.
- “Under 200 words, three bullets, no jargon” beats hoping for the best.
- Name the audience and what you'll do with the output—that's the brief.
How do you write a good AI prompt, step by step?
This is the exact sequence I teach in first sessions. It works in ChatGPT, Claude, Gemini, and Copilot—the tools differ, but the briefing doesn't.
Start with a real task from this week
Skip the trivia test. Open your to-do list and pick something you actually owe someone—a summary you've been putting off, a difficult reply, a document you've been avoiding. A prompt written against real stakes teaches you more in ten minutes than an hour of experimenting with questions you don't care about.
Write two sentences of context
Before the ask, say who you are and what you're working on: “I'm the founder of a 40-person logistics company. We're preparing for our quarterly board meeting.” Then paste the source material—the thread, the draft, the numbers. Two sentences and an attachment do more for answer quality than any clever phrasing.
Assign a role when perspective matters
For anything where the lens matters—feedback, critique, negotiation—tell the AI who to be: “Act as a skeptical board member reading this cold,” or “Act as an executive editor known for cutting ruthlessly.” Skip the role for simple tasks. When you use one, make it specific; “act as an expert” adds nothing.
State the task as a job, with materials
Give the AI a verb and a deliverable: draft, summarize, compare, critique, list. “Summarize this” is a coin flip; “summarize the three numbers that moved most, and flag what will draw questions” is an assignment a competent person could execute. One job per message, or clearly numbered if there are two.
Specify the format, then push back
Close with length, structure, and tone: “Under 200 words, three bullets, plain language, no jargon.” When the first answer comes back, treat it as a draft. The most useful follow-ups are specific: “shorter,” “less formal,” “the second point is wrong—here's why.” The third version is usually where the value is.
How does a weak prompt become a strong one?
These three rewrites are composites of real prompts I've seen in first sessions, with details changed. The weak versions aren't stupid—they're just under-briefed. Watch what the four parts add.
I'm the CEO of [one line: company, size, industry]. Attached is the draft deck for Friday's board meeting: [paste]. Act as a board member reading it cold. Summarize the three numbers that moved most this quarter, flag anything likely to draw questions, and list the two decisions I need the board to make. Under 300 words, plain language.
The weak version was: “Summarize this deck.” The rewrite adds context (who's asking, what meeting), a role (a board member reading cold), a specific task (three numbers, flags, two decisions), and a format (300 words, plain language).
I run [company type]. A long-time customer just emailed to say they're considering leaving after our price increase. Their email: [paste]. Draft a reply from me that acknowledges the frustration in one sentence, explains the increase honestly in one more, and offers a call this week. Under 150 words, warm and direct—no “we value your business.”
The weak version was: “Write an email to an unhappy customer.” Without the actual email and the tone instruction, the AI writes the generic apology you'd delete. With them, it writes a first draft you can actually edit.
I'm deciding whether to [decision, e.g., promote internally or hire externally for our new COO role]. Here's the situation: [3–4 sentences: team, timeline, budget]. Act as a skeptical advisor. List the strongest argument for each option, the two risks I'm most likely underestimating, and the questions I should answer before deciding. Don't make the decision for me.
The weak version was: “Should I hire a COO?” The rewrite doesn't ask for the answer—it asks for the thinking. That last line, “don't decide for me,” keeps your judgment in the loop, which is where it belongs.
What AI prompt examples can you copy today?
Each of these follows the four parts—context, role, task, format. Copy them verbatim, replace the brackets with your material, and adjust the last line to taste.
I'm the [role] of [one line about company]. We've received a proposal from [vendor/partner] to [what they're proposing]. The proposal: [paste]. Act as a skeptical advisor on my side of the table. List the three strongest points in their favor, the three biggest risks for us, what looks missing or vague, and five questions I should ask before the next call.
Works on partnership offers, vendor contracts, and investment teasers alike—the role plus the pasted document is what turns the AI from cheerleader into a second set of eyes.
I need to give ten-minute remarks at our company all-hands on [topic]. Here are my raw talking points: [paste]. And here is a past talk of mine, for my voice: [paste]. Draft remarks that sound like me—direct, unhurried, no buzzwords. Open with the one thing I most need people to remember, and end with what happens next. About 800 words.
Pasting a past talk is the move most people skip. Without a sample of your voice, the AI writes in its voice—competent and generic.
Below are my messy notes on [project]: [paste]. Turn them into a one-page plan: the goal in one sentence, three to five workstreams with an owner and a next step for each, open questions, and anything I've apparently not decided yet. Use [initials] as placeholders where I haven't assigned an owner. No commentary—just the plan.
“Anything I've apparently not decided yet” is the sleeper line here—the gaps the AI surfaces are often the most valuable part of the output.
I'm about to sign [document type: lease, MSA, partnership agreement]. The full text: [paste]. I'm not a lawyer. Summarize what this document commits us to in plain English, flag the five clauses most worth my attention, and list questions to ask my attorney before signing. If anything is ambiguous, say so rather than guessing.
This is a reading aid, not legal advice—the questions it generates for your attorney are where the value is. “Say so rather than guessing” keeps the flags honest.
I want to hand a project to [name or role on my team]. Here's what's in my head: [describe it]. Turn this into a brief they can run with: the outcome I'm asking for, why it matters, the constraints (budget, timing, who to involve), which decisions they can make without me, and how we'll check in. Under 300 words.
Notice what's happening: the same four parts that make a good AI prompt—context, role, task, format—also make a good brief for a person. That's the point of this whole guide.
Why do good prompts still get generic answers?
A client of mine—runs a distribution company, deeply smart—kept getting answers he described as “fine, I guess.” His prompts weren't short, and they weren't lazy. When we looked together, the same handful of small omissions kept showing up. If your prompts look solid but the answers read like they were written for no one in particular, one of these is usually why.
- No audience. “Draft an announcement”—for whom? Customers, employees, and investors need three different documents for the same event, and the AI can't pick for you.
- No format instruction. Length, structure, and tone are the cheapest lines in the prompt and the first things people leave out.
- Five jobs in one message. “Summarize this, critique it, and draft the reply” blends three tasks into one mush. Run them as three messages and each answer sharpens.
- Asking for facts instead of judgment. The AI doesn't know your numbers, this quarter's pipeline, or last week's news unless you paste them. Give the AI material to reason over; keep the lookup jobs for search.
- Vague feedback. “Make it better” teaches the AI nothing. “Cut the second paragraph—it's wrong, here's why” is the edit conversation where the quality actually lives.
How do you turn good prompts into a working habit?
The executives who get real leverage from AI all do one unglamorous thing: they keep the prompts that work. When a prompt produces something genuinely useful, save the exact text in a note, organized by task—board prep, customer emails, plans. Within a month or two you'll have a small personal library, and each entry improves every time you reuse it. A prompt library you built from your own work beats any list on the internet, including this one.
Everything on this page, you can do alone, this week, for free—and many people should. The ceiling I watch executives hit around month two isn't prompting; it's prioritization: which of your workflows—board prep, customer communication, how your firm writes—should be rebuilt around these tools, and in what order. That's the work I do with clients 1-on-1 over video: we open your real workload and build the prompts and habits around your business, your voice, and your risk tolerance. This page gives you the mechanics. Working together is where the mechanics get pointed at your specific week.
1-on-1 coaching
Want a teacher, not another tutorial?
Guides get you started. A 1-on-1 session gets you fluent—built on your actual work, at your pace.
Your coach
Coached by Pete Enestrom
Yale and Columbia grad, former Microsoft and Intel, and a venture-backed exited founder. Pete has spent the last four and a half years going deep on every major AI tool, and he teaches the way operators learn: on your real work, at your pace, with nothing assumed.
Common questions
Straight answers, the way I'd give them across a table.
Keep reading
Guide
ChatGPT Prompts for Business: The 12 I Hand Every Client
The ChatGPT prompts for business I hand coaching clients: twelve copy-paste prompts for drafting, analysis, meetings, and decisions, each with the reason it works.
ReadResource
AI Prompt Library: What It Is and Why Every Executive Should Keep One
An AI prompt library is your personal collection of tested prompts, organized by the work you repeat. Why every executive should keep one, plus a starter set.
ReadTraining
AI Fluency: What It Is and How Executives Get There
AI fluency defined in plain English: what fluent AI use looks like on an executive's real work, why certificates don't create fluency, and how leaders build it.
ReadGuide
How to Use Claude: A Practical Guide for Busy Executives
A plain-English guide to Claude for founders and executives: what it is, how to start in ten minutes, and five real workflows with copy-paste prompts.
ReadBeyond coaching
When the workflow needs to be encoded, not just learned — that's what our team at Zaigo builds.