When ChatGPT gives you a bland, generic or slightly-off answer, the instinct is to blame the model. Most of the time the problem is the prompt. The model fills every gap you leave with the safest, most average guess — and average is rarely what you wanted.
The good news is that the gaps are predictable. These seven fixes cover most of them, and they work the same way in Claude, Gemini and other assistants.
1. Say who it is for
The same explanation reads very differently for a CEO, a new hire or a ten-year-old. If you don't name the audience, you get something pitched at nobody in particular.
Before: "Explain what an API is."
After: "Explain what an API is to a marketing manager with no technical background, using one everyday analogy, in under 150 words."
2. Describe what "done" looks like
Format is the most common missing piece. A list, a table, an email, three options, a single recommendation — if you know the shape you need, say it.
Before: "Give me ideas for our team offsite."
After: "Give me five ideas for a one-day team offsite for 12 people with a budget of $1,500. Present them as a table with columns for activity, cost estimate and why it builds team trust."
3. Put the context in the first message
ChatGPT can't see your situation. Drip-feeding details across follow-ups means the first answers are built on guesses, and every follow-up resends the whole conversation.
Before: "Write a reply to this customer." (then three follow-ups explaining the order delay, the refund policy and the tone you want)
After: "Write a reply to the customer email below. Context: their order is delayed by 5 days because of a supplier issue; we can offer free express shipping but not a refund. Tone: apologetic but confident, no corporate jargon. Under 120 words."
4. Set limits on length
Without a limit, the model tends to write more than you need. Long answers are slower to read and, on usage-based plans, cost more.
Before: "Summarize this report."
After: "Summarize this report in five bullet points, one sentence each, focused on decisions the board needs to make."
5. Ask for one thing at a time — or number the parts
Prompts that bundle several requests usually get one of them done well and the rest rushed. Either split them, or make the structure explicit.
Before: "Review my CV and tell me what jobs to apply for and write a cover letter."
After: "Do three things, in order. 1) List the three biggest weaknesses in the CV below. 2) Suggest two job titles it is best suited for. 3) Wait for my confirmation before writing a cover letter."
6. Give an example of what good looks like
If you have a sample of the style you want — a past email, a paragraph you like — include it. One example often does more than a paragraph of instructions.
Before: "Write a LinkedIn post about our new hire."
After: "Write a LinkedIn post welcoming our new Head of Finance. Match the tone and length of this earlier post: [paste example]. No emojis, no hashtags."
7. Tell it what to avoid
Negative instructions are underrated. If you keep getting clichés, filler intros or made-up figures, say so up front.
Before: "Write an introduction for my report on water access."
After: "Write a two-paragraph introduction for a report on rural water access. Do not start with a rhetorical question, do not use statistics unless I provide them, and avoid the words 'crucial' and 'landscape'."
A simple checklist before you hit Enter
- Audience: who will read the answer?
- Format: what shape should it take?
- Context: what does the model need to know about your situation?
- Length: how long should it be?
- Example: do you have a sample of good output?
- Avoid: what keeps going wrong?
You don't need all six every time. For a quick question, one or two is enough. For anything you'll send to someone else, run through the whole list.
Let a tool do the restructuring
Writing prompts this way becomes habit with practice, but it takes effort when you're busy. IacuWise does the restructuring for you: paste a rough prompt, pick the AI you use, and get back a version with the audience, format, context and limits filled in — ready to send. With the Chrome extension, the Optimize button sits right inside ChatGPT, Claude and Gemini.