When a prompt disappoints, the instinct is to blame the model or to add more instructions. Usually the cause is one of five ordinary mistakes, and all of them are quick to fix.

1. You left a placeholder in

The single most common one. A prompt says [DESCRIBE YOUR IDEAL CUSTOMER], you paste it in a hurry, and the assistant dutifully writes for a generic ideal customer. It will not tell you that you forgot. It fills the gap and moves on.

Fix: before you press enter, search the prompt for [. If anything is left, either fill it or delete the line. An empty slot is worse than no slot at all.

2. You are reusing a polluted chat

Chat assistants weight everything already in the conversation. If you spent twenty messages debugging a spreadsheet and then ask for a client email, the tone of that debugging session leaks in: shorter, more clipped, more technical than you want.

Fix: start a new chat for each distinct task. It costs a second, and it removes an entire class of "why is it being weird" moments.

3. You asked for everything in one go

"Write me a marketing strategy, the campaign plan, the emails and the landing page" produces a shallow version of four things. Attention gets spread thin, and every part gets a paragraph where it needed a page.

Fix: chain instead. Get the strategy, correct it, then ask for the campaign plan based on the strategy we just agreed. Each step inherits a foundation you have already checked.

4. You are being polite instead of specific

"Could you maybe make it a bit more professional?" gives the model almost nothing to act on. Professional how: longer sentences, fewer contractions, more hedging, more data?

Fix: describe the change mechanically. "Remove all contractions, cut every sentence over 25 words in half, and replace the three adjectives in paragraph two with a number." Feedback the model can verify is feedback it can apply.

5. You accepted the first draft

The first output is a starting position, not an answer. The fastest quality gain available to anyone is one more turn:

Which part of this is weakest, and why?

Rewrite the opening three times, each with a different angle.

What did I not tell you that would have made this better?

That last question is worth asking regularly. The answer usually tells you exactly which context slot you left thin.

The pattern underneath

Four of these five are the same mistake wearing different clothes: the assistant is working with less than you think it has. It cannot see your business, your client, your last conversation or your standards unless you put them in the prompt, and it will never stop to ask. It will produce something plausible from whatever it was given.

Give it more, in a structure it can use. That is the whole game.

Our packs are built around exactly this: every prompt states the role, lists its variables explicitly so nothing is left blank, and ends with a self-critique step. See what's in them.