Why your prompts get generic answers
An AI assistant answers the question it was asked, using whatever it can guess about what you meant. "Write a cover letter" leaves it guessing your job, your experience, the employer and the tone, so it falls back on the most average letter possible.
The fix is not magic wording. It is giving the model the information a capable human assistant would need. These principles apply equally to ChatGPT, Gemini, Claude, Copilot and other chatbots.
The five-part prompt
Good prompts usually contain five ingredients. You do not always need all five, but checking for them fixes most weak prompts.
| Part | What it does | Example |
|---|---|---|
| Role | Sets the expertise and perspective | "You are an experienced recruiter in the tech industry." |
| Context | Gives the facts the model cannot know | "I have 4 years in customer support and I'm moving into product management." |
| Task | States exactly what you want | "Draft a cover letter for this job posting." |
| Constraints | Limits length, tone and content | "Under 250 words, confident but not boastful, no clichés." |
| Format | Shapes the output | "Three short paragraphs, then a one-line sign-off." |
Before and after
Example 1: an email
Weak: "Write an email asking for a deadline extension."
Strong: "Write a short email to my manager, Sara, asking to move the quarterly report deadline from Friday to next Tuesday. The delay is because the finance data arrived two days late. Offer to send a draft of the first half on Friday. The tone should be professional and matter-of-fact, without over-apologising, in under 120 words."
The second prompt gives the reason, the new date, an offer and the tone. The answer needs almost no editing.
Example 2: learning a topic
Weak: "Explain inflation."
Strong: "Explain inflation to me as if I'm a smart 16-year-old with no economics background. Use one everyday example involving grocery prices. Then give me three questions to check I understood, and wait for my answers before telling me if I'm right."
The strong version sets the level, asks for an example and turns the chat into a lesson.
Example 3: a decision
Weak: "Should I buy a laptop or a tablet?"
Strong: "Help me decide between a laptop and a tablet. I'm a university student. I mainly write essays, join video calls and watch lectures, and I sometimes do light photo editing. My budget is around $700. Compare the two for my use in a table, then give a recommendation and the one thing that would change your mind."
Techniques that make a real difference
Give an example of what you want
If you have a style or format in mind, show it. "Write product descriptions in this style:" followed by one sample is often more effective than a paragraph of instructions.
Ask it to ask you questions first
For anything personal or complex, end with: "Before you start, ask me any questions you need answered to do this well." The model will often spot missing details you had not thought about.
Break big jobs into steps
Instead of "write my business plan", work in stages: outline first, then each section, then a review. You can correct the direction early instead of rewriting everything.
Ask for options, then refine
"Give me five headline options in different styles" followed by "Make the third one shorter and more direct" is faster than trying to describe the perfect headline up front.
Ask it to check its own work
After a draft, try: "Review this for factual claims I should verify and for anything unclear." It is not a guarantee, but it catches obvious problems.
Reusable templates
Copy these and fill in the brackets.
Summarise:
Summarise the text below for [audience] in [number] bullet points. Keep any numbers and dates exactly as written. Then list anything that is unclear or missing.
Rewrite:
Rewrite this [email or paragraph] to be [clearer, shorter or more formal]. Keep my meaning and any specific details. Show me only the rewritten version.
Plan:
I want to [goal] by [date]. I have [time per week] and [constraints]. Make a week-by-week plan, and mark the steps where I'm most likely to get stuck.
Explain:
Explain [topic] at the level of [audience]. Use one concrete example. Finish with a short list of common misconceptions.
Mistakes to avoid
- Asking for facts without checking them. Models can invent statistics, quotes and references. Verify anything that matters.
- Cramming several unrelated tasks into one prompt. Separate requests get better results.
- Accepting the first answer. Treat it as a draft. Ask for changes, say what is wrong, and push back.
- Pasting sensitive data. Leave out passwords, personal identification numbers, other people's private information and confidential work files.
- Using the wrong tool. For current events, use a chatbot with web search turned on and check its sources.
Where to go next
If you are choosing between assistants, our comparison of ChatGPT and Gemini for everyday tasks covers their strengths. For a wider look at AI apps for work, see the best AI tools for productivity.