When a team says an AI tool produced something generic, the brief is usually the problem. Models fill gaps with the average of everything they have seen, so anything you leave out gets replaced by the most common answer. The fix is a short, structured brief you reuse every time.
What the brief needs
- Audience: who this is for, in one line, including what they already know.
- Job: what the piece must make the reader think, feel or do.
- Voice: three words that describe the tone and three that do not.
- Facts: the specific details that must appear: product names, numbers, dates, names.
- Constraints: length, format, platform, anything legally required.
- Reference: one example you like and one you do not, with a sentence on why.
How to use it
Paste the brief at the top of every prompt, then ask for several variations rather than one perfect answer. Choose, edit, and feed the edited version back as the new reference. Over a few rounds the tool learns what you mean by good, and the human time shifts from writing to judging, which is where it belongs.
For images and video
The same structure holds. Replace voice with visual style, add framing and aspect ratio to constraints, and always include a reference image. Describing what you do not want, for example no stock-photo smiles, is as useful as describing what you do want.
