Most people use ChatGPT the same way they'd use Google — type something short, read what comes back, accept or reject it. That works, but it leaves most of the model's capability untouched. These seven changes take under a minute each to implement and consistently produce dramatically better output.
1. Tell It Who to Be
Before every prompt, add a role. "You are a financial adviser with 20 years of experience" produces very different output from the same question asked without context. The role tells the model what style, vocabulary, and level of expertise to apply.
Keep it specific: "You are a conversion copywriter who has written landing pages for SaaS companies" is better than "you are a copywriter."
2. Add a "Banned Phrases" List
ChatGPT has stylistic defaults — phrases it reaches for when it doesn't have more specific guidance. "It's important to note," "In today's fast-paced world," "Game-changing." These are the tells of generic AI output.
Add a one-line rule at the end of your prompt: "Banned phrases: [LIST]." This alone significantly improves output quality for writing tasks.
3. Specify the Exact Format
Without format guidance, ChatGPT chooses its own — which may not match what you need. Be explicit: "Return a numbered list of 5 items," or "Respond in a single paragraph under 100 words," or "Format as a table with columns: [COLUMN 1], [COLUMN 2], [COLUMN 3]."
4. Say What You're Going to Do with the Output
Context about end use changes the output. "Write a summary of this article" is vague. "Write a 3-sentence summary of this article I'll use as the introduction email to send to my newsletter subscribers" gives the model context about tone, length, and what matters.
5. Ask It to Critique Before You Accept
After getting a response, ask: "What are the weaknesses of what you just wrote? What would a critic say is missing?" This forces the model to surface its own limitations — and often identifies exactly what needs fixing before you've spotted it.
6. Push Back When You Disagree
ChatGPT is sometimes too agreeable. If you think the first answer is wrong, say so — and say why. "That's not quite right. The problem with that approach is [REASON]. Try again with this constraint in mind."
The model doesn't know you've pushed back unnecessarily versus correctly. But stating your reasoning when you do gives it more to work with.
7. Use "Give Me 3 Options"
For any output where you're not sure exactly what you want — a headline, a subject line, an opening paragraph — ask for three versions with different approaches. "Give me 3 options: one benefit-led, one problem-led, one curiosity-driven."
Three options gives you a range to choose from and often surfaces angles you wouldn't have specified in advance.
The Compounding Effect
None of these individually is revolutionary. Combined — role + banned phrases + format + end-use context + three options — they transform the quality of what you get back. The model has everything it needs to do the task well, rather than filling every gap with the statistically average answer.
Copy any prompt from the Promptzio library to see all seven of these principles applied to specific tasks across marketing, writing, coding, and more.