Generic prompts produce generic copy. “Write a product description for X” gets you three adjectives and a call to action that could belong to any business on earth. These four patterns produce something closer to usable — and every single output still gets edited before it ships.
1. The constraint-first pattern
State the audience, the length limit, and the one thing to avoid, before you ask for the copy itself. “Write for a first-time buyer who has never used this category of product. Keep it under 60 words. Do not use the words revolutionary or cutting-edge.” Constraints do more work than adjectives.
2. The real-detail pattern
Feed the model one real, specific fact about the product — a material, a measurement, a use case a real customer mentioned — and ask it to write around that fact rather than inventing its own. Specific beats generic every time, and it is the fastest way to stop output from sounding like every other AI-written page.
3. The competitor-contrast pattern
Ask the model to write the copy twice: once describing what makes this product different from the obvious alternative, and once as a plain feature list. The contrast version is almost always the stronger draft, because it forces a specific claim instead of a vague one.
4. The read-aloud pattern
Ask directly: “rewrite this so it sounds like a person said it out loud, not like marketing copy.” This single instruction catches more of the generic AI tells — em dashes stacked three deep, empty phrases like “in today’s fast-paced world” — than almost any other prompt trick.
What we always edit afterward
- Any claim that sounds impressive but is not actually true for this specific product.
- Adjectives doing the work a real detail should be doing.
- Any sentence that would read the same on a competitor’s page with the noun swapped out.
The prompt gets you a draft. The edit is what makes it usable. Skipping the edit is the difference between AI-assisted copy and AI-generated copy — and customers can tell.