What belongs in a brief
A solid brief has six slots: product name, platform, audience, features (as facts, not marketing claims), differentiators, and tone. Optional slots include region, locale, length ceiling, and existing copy to react to. Slots that are not used should be deleted, not left blank. The model treats blanks as soft hints and often makes them up.
Keep the brief honest. The product does not have a 'premium hand-stitched feel' — it has 12 oz waxed canvas and a YKK zipper. The model can do the framing for you. The model cannot do the facts for you.
Common brief failure modes
Dumping a whole catalog into the prompt is the most common mistake. The model cannot write 40 listings at once with the same level of detail; it averages across them. Feed it one product at a time. If you must batch, group by audience or platform so the model can stay consistent.
Asking for 'creativity' is another trap. Without a constraint, the model defaults to the median of its training data, which is exactly the kind of empty, adjectival copy you do not want. Specific constraints produce specific output. Soft asks produce soft output.
From brief to publishable copy
The AI output is the start of a two-step process: draft, then edit. The edit pass should focus on three things. Are the facts preserved? Does the title match platform conventions? Does the bullet list lead with the most important benefit, not the most obvious feature? The AI does a passable job on the first pass; the operator's value is in the second pass.
After the second pass, the listing is ready to publish. The whole loop, from brief to published copy, should take less than ten minutes per product once the team has a template. Ten minutes beats an hour-long copywriting session when the input quality is consistent.