How to Write a ChatGPT Prompt That Actually Works
Most people write prompts the same way they'd type a Google search. They get disappointing results and conclude that AI isn't useful for their work. The problem isn't the model — it's that ChatGPT needs context, structure, and constraints to produce expert output. A well-built prompt is closer to a design brief than a search query. This guide gives you the 5-part framework that consistently produces output worth using.
Why most prompts fail
The most common failure mode is ambiguity. "Write me a marketing email" could mean a cold outreach email, a re-engagement campaign, a product launch announcement, or a newsletter. ChatGPT will make assumptions to fill the gaps — but those assumptions won't match your situation. The model isn't bad at marketing emails; it just doesn't know who your customer is, what they care about, or what you want them to do.
The second failure mode is not specifying the expert role. ChatGPT can reason from the perspective of a senior copywriter, a junior intern, a brand manager, or a generic "helpful assistant." Without direction, it defaults to the middle — competent but undistinctive. Telling it to be a "B2B SaaS copywriter who specializes in email sequences for enterprise sales" produces a fundamentally different output.
The 5-part prompt framework
1. Role
Tell ChatGPT who it is. Not "you are an expert" — that's too vague. Be specific: "You are a senior UX writer who has worked on onboarding flows for B2B SaaS products." The role sets the model's frame of reference for every word it produces. Think of it as telling a consultant their domain of expertise before they open their mouth.
2. Context
Provide the situation. Who is the audience? What problem is being solved? What has already happened? What does the reader already know? Context is where most prompts are too thin. The more specific your context — "our users are logistics managers at mid-size trucking companies who are evaluating 3 competing platforms" — the more calibrated the output.
3. Task
State exactly what you want produced. Use active verbs and be specific about scope: "Write a 4-email drip sequence" not "write some emails." If you want a specific structure — an intro, three bullet points, a CTA — describe it. ChatGPT takes instructions literally, which is an advantage when you use it correctly.
4. Format
Specify the output format. Do you want prose, a numbered list, a table, a script with labeled sections, or bullet points with sub-bullets? Should it use markdown headers? Should it be a complete document or just the body copy? If you're pasting the output into a tool that doesn't support markdown, say so. Format instructions eliminate a lot of the post-generation editing.
5. Constraints
Tell it what not to do. Word limits, tone exclusions, topics to avoid, phrases that are off-brand — constraints are what turn a generic output into one that fits your specific situation. "Under 150 words," "no bullet points," "do not use the word 'seamless,'" and "do not make any claims about ROI without a source" are all legitimate constraints that meaningfully change the output.
5 Example Prompts Using the Full Framework
B2B cold email sequence
You are a senior B2B copywriter who specializes in outbound email sequences for enterprise software sales. You write emails that get responses — not because they're clever, but because they're specific and relevant. I'm writing a 3-email cold outreach sequence targeting VP of Operations at mid-size logistics companies (50–500 employees). Our product is a route optimization platform that reduces fuel costs by an average of 18% and is typically implemented in under 4 weeks. The pain point we're addressing: manual route planning that takes dispatchers 2+ hours per day and often produces suboptimal routes. Write a 3-email sequence: - Email 1 (Day 1): Hook with a specific pain point, introduce the outcome (not the product), one CTA - Email 2 (Day 5): Follow-up that adds a new insight or data point — not just "checking in" - Email 3 (Day 12): A soft close or a different angle that either gets a response or a clear no Format each email with: Subject line, Email body (under 150 words each), and a note on timing/send day. Constraints: No buzzwords (disruptive, game-changing, revolutionary). No fake personalization tokens like [FIRST NAME] — write as if addressing Marcus specifically. No pressure tactics.
Explaining a technical decision to non-technical stakeholders
You are a senior software architect who is also a skilled communicator. You've spent years translating complex technical decisions into language that business stakeholders can evaluate and approve. I need to explain why we're migrating from a monolithic backend architecture to microservices. Context: I'm presenting this to a board of directors that includes no engineers. They're concerned about cost (the migration will require 6–8 months of additional engineering time before they see benefits) and risk (what happens if the migration goes wrong mid-product). Our primary reasons: current architecture can't scale to handle our expected 10x user growth without significant downtime, and it's causing 3–4 engineering weeks per month in incident response. Write an executive briefing document (not a slide deck script) that: - Opens with the business problem, not the technical decision - Uses an analogy to explain monolithic vs. microservices (make it intuitive, not condescending) - States the risks honestly and how we're mitigating each one - Quantifies the cost of not migrating (current incident cost + growth ceiling) - Ends with a clear recommendation and the specific approval we're asking for No jargon. Max 500 words. Formatted with clear headings.
Performance review self-assessment
You are a career coach and former engineering manager who helps individual contributors write self-assessments that accurately represent their impact — not just list what they did, but connect their work to business outcomes.
I'm writing my annual self-assessment. I'm a mid-level software engineer (L4 at a tech company). Here's what I worked on this year:
- Led the backend implementation of our new payments API (shipped 2 weeks early, zero P1 incidents post-launch)
- Mentored 2 junior engineers — one was promoted after 8 months
- Wrote internal documentation for our authentication system (previously undocumented, now used by 4 other teams)
- Participated in on-call rotation (avg 2 hours/week incident response time, down from 5 hours/week after I rewrote 3 alert thresholds)
Write a self-assessment that:
- Frames each contribution in terms of impact, not just activity ("led the payments API implementation" → "delivered the payments API 2 weeks ahead of schedule with no post-launch incidents, enabling our Q3 revenue goal")
- Highlights both technical and collaborative contributions
- Is written in first person, confident but not arrogant
- Ends with 2–3 goals for next year that are specific and ambitious
Under 600 words. No filler phrases like "I believe I have shown" or "I am passionate about."Research summary for a decision-maker
You are a research analyst who synthesizes complex information into clear, decision-ready summaries. You understand that executives don't want to read everything you read — they want to know what it means and what to do. I've been researching whether our SaaS company should expand into the European market in the next 18 months. I need a research summary memo covering 4 dimensions: market size, regulatory requirements (especially GDPR), competitive landscape (are our main US competitors already there?), and operational requirements (data residency, payment localization, language/support). I don't have research to paste in — generate this from your training data, but be specific about what you know confidently vs. what I should verify independently. Format as a decision memo with: - Executive summary: 3 bullet points (opportunity, main risk, recommended next step) - One section per dimension (market, regulatory, competitive, operational) - A "What we need to know before deciding" section that lists the gaps in this analysis - A clear recommendation at the end: proceed to deeper due diligence, delay, or don't pursue Flag anything in this memo that may have changed since your training data cutoff.
Workshop facilitation guide
You are an experienced workshop facilitator who designs structured sessions for leadership teams. You believe most workshops fail because the agenda is too packed, the goals are unclear, and people spend the first 40 minutes getting aligned on what they're even trying to decide. I need to facilitate a 3-hour strategy session with 8 senior leaders (VP-level) at a $50M B2B SaaS company. The goal: align on 3 strategic priorities for next year. Current situation: there are 11 initiatives on the table, the leadership team has different views on what matters most, and the last strategy session ended without a decision. This time, we need to leave with a ranked list of 3 priorities and clear ownership. Design a facilitation guide that includes: - Pre-work to send to participants 48 hours before (30 minutes max — they won't do more) - Minute-by-minute agenda for the 3 hours - Specific facilitation techniques for the prioritization exercise (dot voting, forced ranking, or something better for this situation) - How to handle the moment when two VPs disagree and the conversation stalls - A decision documentation template to capture output during the session - What to send as a follow-up within 24 hours to lock in alignment Be specific. Don't just say "facilitate a discussion" — tell me exactly what to do.
Tips for Writing Effective ChatGPT Prompts
ChatGPT handles long, detailed prompts very well — don't be afraid of length. A 300-word prompt that fully specifies role, context, task, format, and constraints will almost always outperform a 20-word prompt, even if the short prompt feels faster to write. The time you invest in the prompt pays dividends in how much editing you don't have to do afterward.
ChatGPT works especially well in a conversational back-and-forth. If the first output is 80% right, don't restart — refine it. "Make this more direct," "shorten this by half," or "the second section is too formal — rewrite it in the same tone as the first" are faster than starting a new prompt. The model remembers the full conversation context, so follow-up instructions don't need to repeat everything you said in the first message.
For creative work, consider telling ChatGPT to generate multiple variations before you choose one. "Write 3 different versions of this headline — vary the emotional angle" is often more productive than iterating on a single direction. The model is fast enough that generating options costs you nothing except reading time, and having 3 options often reveals a direction you wouldn't have thought to ask for directly.
Why use PromptBro to write ChatGPT prompts?
PromptBro's 6-step flow is a guided implementation of the 5-part framework above. It walks you through role, context, task, format, and constraints before assembling a complete, structured prompt — so you don't have to keep the framework in your head while also thinking about your specific problem. Speak your goal out loud, answer a few guided questions, and get a production-ready ChatGPT prompt in under 60 seconds.
Try PromptBro free — build your first prompt in 60 seconds →