Most people who are unhappy with their AI output think the problem is the AI. Usually the problem is the prompt. Not because the prompt is wrong, but because it's incomplete — it's missing one or two things that would have changed the output significantly.

The checklist below takes under five minutes to run. Use it on any prompt before you send it. Each item corresponds to one of the five most common reasons AI output comes back unusable. Check off the ones your prompt already handles. Fix the ones it doesn't.

The 5-Minute Prompt Audit
Check each item — click to mark complete
✓
1. Does your prompt tell the AI who it is?
A prompt with a defined role produces fundamentally different output than the same prompt without one. "You are a senior copywriter" changes the vocabulary, the assumptions, and the approach — even for a simple task.
Fix: Add "You are a [role] with [experience] in [domain]." at the start.
2. Does it start with a specific action verb?
Prompts that begin with a clear verb — Write, Analyze, Summarize, Explain, List, Draft, Compare — produce more focused output than prompts that describe what you want in abstract terms.
Fix: Start the task instruction with a verb. Not "I need help with an email" — "Write an email to..."
3. Does it include enough context?
The AI only knows what you tell it. If the output seems off-base or generic, it's usually because the model was guessing at context you assumed was obvious. Who is the audience? What is the purpose? What does the reader already know?
Fix: Add one sentence of background that answers: who is this for, and why does it matter to them?
4. Does it specify output format?
Without format instructions, the model guesses — and it often guesses wrong. You ask for a quick summary and get four paragraphs. You ask for ideas and get a numbered list with lengthy explanations for each. Specify format and you get what you actually need.
Fix: Add: "Output as [bullet points / a table / an email with subject line / under X words / a step-by-step list]."
5. Does it include at least one constraint?
Constraints are the most underused part of a prompt and often produce the biggest improvement. Telling the AI what to avoid — certain phrases, a maximum length, a tone to stay away from — forces it off its statistical defaults toward more specific, useful output.
Fix: Add: "Avoid [X]. Do not include [Y]. Keep it under [Z] words."
Checked: 1 / 5 Check each item that applies to your prompt

How to use this in practice

You don't need to hit all five on every prompt. A simple question — "what's the capital of France?" — doesn't need a role, a format specification, or constraints. But for any prompt where you're asking for a draft, an analysis, a summary, or anything where the output quality actually matters to you, running through these five items before sending will consistently produce better first drafts.

The most common gap: In our testing, the item most often missing from beginner prompts is the constraint (item 5). Most people tell the AI what they want but never tell it what they don't want. Adding one negative instruction — even something as simple as "under 150 words" or "no jargon" — consistently improves output quality more than any other single addition.

What to do when the output is still wrong

If you've addressed all five items and the output is still not what you need, the issue is usually one of these three things:

The context was too vague. "A B2B product" tells the AI almost nothing. "A project management tool for remote engineering teams at companies with 50 to 200 employees" gives it something to work with. The more specific the context, the more accurate the output.

The format was under-specified. "A short email" and "an email under 100 words with a subject line, one sentence of context, one sentence of value proposition, and a single low-friction call to action" are not the same instruction. The second one produces a usable first draft. The first one produces something that needs significant editing.

You need to iterate, not restart. The best use of AI for writing is not a single perfect prompt — it's a conversation. Get a first draft, then ask for specific changes: "The second paragraph is too long. Cut it to two sentences." "The tone is too formal. Make it sound more like a colleague, not a consultant." Iteration is faster than trying to write the perfect prompt upfront.

The faster version

If five questions feels like too much overhead, here's the one-question version: If I sent this prompt to a freelance writer instead of an AI, would they have everything they needed to do the job?

If the answer is no — if a human writer would need to ask you follow-up questions before starting — your prompt needs more information. Add what a writer would ask for, and the AI output will improve immediately.