Why your AI prompts aren't working (and how to fix them)

1. You are being too vague

"Write a marketing plan" gets you generic slop because the model has to guess everything. The fix is specificity: who it is for, what the goal is, and what the constraints are.

You are a head of marketing at a 12-person SaaS startup. Write a 6-week launch plan for a $99/mo product. Include weekly deliverables, an owner, and a KPI for each. Keep it under 400 words.

Same model, completely different result.

2. You skipped the role and context

Models behave differently depending on who you tell them to be. "You are a senior copy editor" produces sharper edits than no role at all. Add the context it needs, too — the audience, the prior decisions, the format of the source material. Without context, the model invents it, and that is where wrong answers come from.

3. You didn't specify the output format

If you do not say what the output should look like, you get a wall of prose. Specify it as a contract: a table, a bulleted list, JSON, a three-sentence summary. "Return only a markdown table with columns Task, Owner, Due" stops the model from rambling and makes the output something you can actually use.

4. You accepted the first draft

The first draft is the worst draft. The single highest-leverage move is to make the model critique and improve its own answer: "List three weaknesses of that draft, then rewrite to fix them." It costs a few seconds and almost always improves the result.

5. You gave no examples

For anything where style or format matters, show — do not tell. Two or three examples that bracket the range you want teach the model faster than a paragraph of instructions. This is the few-shot pattern, and it is the fastest fix for "close, but not quite right."

Putting it together

Most failed prompts break one of those five rules. A reliable opener that covers most of them: role, goal, context, constraints, and output format — then a self-critique pass. That structure is the backbone of practical prompt engineering.

The fastest way to internalise it is reps. Iro AI's Prompt Lab grades your real prompts and shows you exactly which of these mistakes you are making; the ChatGPT path applies the same moves inside the tool.