A friend of ours spent an hour looking for the "right" ChatGPT prompt to write a project update. She found 12 templates. None fit. She wrote the email herself.
Search "how to use ChatGPT for work" and you'll find hundreds of prompt lists. They all share the same assumption: ChatGPT is a machine where you insert the right input and get the right output.
The people who get value from ChatGPT don't work that way. They've developed a feel for steering a conversation toward what they want. That feel comes from three behaviors, not 50 prompts.
Behavior 1: Ask (Roughly)
It's easy to overthink your first message.
Start rough instead. Tell ChatGPT what you want in plain language, the same way you'd ask a colleague who just joined your team.
"I need to write a project update for my manager. The project is two weeks behind. Can you help me draft something?"
That's a fine first message. No role assignment or format specification. Just the request, in the words you'd use.
The goal of the first message isn't to get perfect output. It's to start the conversation. You'll refine from there.
Behavior 2: Loop (Iterate)
This is where it's tempting to stop. You get a response, decide it's not quite right, and either accept it or go back to your prompt list.
The skill that matters is iteration. Get a response, evaluate it, give feedback, go again. Each round gets closer to what you want.
Round 1: "I need to write a project update for my manager. The project is two weeks behind. Can you help me draft something?"
ChatGPT gives you a draft. It's okay. A little generic.
Round 2: "This is too formal. My manager is pretty casual. Also, the delay is because a vendor shipped materials late. Can you add that?"
Better. Closer to your voice.
Round 3: "Good. Can you lead with the impact on the deadline instead of the reason? My manager cares about the timeline first."
Now it's useful.
Three rounds. About 90 seconds. The final version is better than what you'd produce in 20 minutes of staring at a blank document.
The prompt is just the opening move. Each round of feedback is where the work happens. People skip the loop because they expect AI to work like a search engine: type a query, get an answer. But ChatGPT generates responses based on patterns, and the first pattern it finds might not be the one you need. Each round of feedback narrows it down.
Behavior 3: Give Context
The single biggest reason people get bad output from ChatGPT isn't bad prompting. It's missing context.
ChatGPT doesn't know your job. It doesn't know your company. It doesn't know your manager's preferences, your team's culture, or the political dynamics of your project. It knows patterns from training data. It doesn't know you.
Context doesn't mean a longer prompt. It means the specific details that shape what good output looks like for your situation.
Bad context: "Write a project update."
Good context: "Write a project update for my manager, who is direct and wants bullet points. The project is a website redesign. We're two weeks behind because our design vendor delivered late. I need to communicate the delay, the new timeline, and what we're doing to catch up."
The second version takes 10 more seconds to write. It produces a better result.
Context to include when it matters:
- Who the output is for — your manager, your team, a client, a broader audience
- What tone fits — formal, casual, urgent, informative
- What happened — the situation itself, not a hypothetical
- What format works — bullet points, a narrative, a short email, a slide outline
- What you don't want — not too long, not too technical, no jargon
You don't need all five every time. Each one you add improves the output. And you can add them in the loop, not all at once. If the first draft is too formal, mention your manager's communication style. If it's too long, say you need it under 200 words. Context is cumulative.
What This Looks Like for Different Roles
The three behaviors are the same regardless of what you do. The applications differ.
Managers
You're preparing for a 1:1 with a direct report who's been underperforming. You want to be supportive but clear.
Ask: "Can you help me prepare for a 1:1 with someone on my team who's been missing deadlines? I want to be supportive but also set clear expectations."
Loop: "The first draft is too harsh. Soften the opening. I want to start by acknowledging what's been going well. And add a section on specific support I can offer."
Context: "She's been dealing with a family situation for the past month. Before that, she was a strong performer. I don't want to come across as punitive."
Individual Contributors
You need to write a quarterly self-review and you're not sure how to frame your contributions.
Ask: "I need to write a quarterly self-review. Can you help me structure it?"
Loop: "This feels too boastful. Make it more matter-of-fact. I want to state what I did without sounding like I'm bragging."
Context: "My role is in operations. This quarter I streamlined our onboarding process, which reduced new hire ramp time from 6 weeks to 4. My manager values measurable impact over effort."
Sales and Marketing
You're drafting a cold outreach email and want it to sound human.
Ask: "Help me write a cold email to a VP of Operations at a mid-sized logistics company."
Loop: "Too salesy. Cut the second paragraph. Make the opening more specific to their situation."
Context: "Our product is an inventory tracking tool. The specific pain point we solve is stockouts caused by manual tracking. The email should be under 100 words. No buzzwords. Just a clear, specific offer."
Things That Trip People Up
Accepting the first output. ChatGPT's first response is a draft. If you take it as-is, you're getting a first draft. Loop at least once.
Overloading the first message. Don't try to include every detail in your opening prompt. Start rough and add context as you go. The loop handles complexity better than a single mega-prompt.
Treating it like a search engine. ChatGPT doesn't retrieve answers. It generates responses based on patterns. It can be wrong, generic, or miss the point. Your job is to steer it.
Sharing sensitive information. Both the free tier and ChatGPT Plus use your conversations to train future models by default. You can opt out in settings, but that's not obvious. Don't paste confidential company data, personal information, or anything you wouldn't want stored on someone else's server. Team and Enterprise plans don't train on your data by default. We cover this in more detail in our AI literacy guide.
FAQ
How do I use ChatGPT for work without sounding like AI?
Iterate. The first draft usually sounds generic because ChatGPT defaults to a neutral, formal tone. In the loop, tell it to match your voice: "Make this more casual" or "Shorten the sentences" or "Use contractions." Give it an example of how you write. The more you steer, the less it sounds like AI. For more on developing judgment about AI output, see our beginner's guide to AI literacy.
Is ChatGPT free to use for work?
The free tier gives you access to GPT-5.6 Luna, which handles most work tasks well. ChatGPT Plus ($20/month) adds more advanced models, higher message limits, and features like file analysis and web browsing. For most beginners, the free tier is enough to start.
What should I not use ChatGPT for at work?
Don't use ChatGPT for tasks where being wrong is expensive and you can't verify the output. That includes legal advice, financial calculations without checking the math, and anything involving confidential data on a free tier. ChatGPT is a thinking partner, not a source of truth. We wrote about this broader principle in what AI is good for.
Do I need to learn prompt engineering?
No. Prompt engineering as a formal discipline is useful for people building AI applications. For everyday work, the three behaviors in this post cover what you need. If you can hold a conversation with a colleague, you can use ChatGPT effectively.
Want to go further?
Try one task you need done this week. Not a tutorial. Not a demo. Something on your list: an email, a document draft, a brainstorm.
Use the three behaviors: ask roughly, loop at least twice, give context when the output isn't right. You'll learn more from one task that matters than from any prompt list.
If you want a structured starting point, take the free GWT survey at greatwithtech.com. It takes 2 minutes and maps your AI readiness across practical skills.