Custom AI instructions: how to steer your product description generation

AI writes to the average of what it has seen unless you tell it otherwise. Without direction, a lounge chair and a protein powder get described in the same neutral voice. A short instruction pulls the output toward your brand, your audience, and your category.

In AI Product Optimizer, this lives in the extra instructions field on the Products screen. Whatever you type there is applied across every product in your selection, so one good instruction improves an entire batch.

Tone is the easiest win. Compare "Write in a refined, aspirational tone, emphasize craftsmanship and materials, avoid hype words" with "Use a friendly, upbeat voice with short sentences and speak directly to the reader as you".

Same product, completely different copy. Pick the voice that matches your store and keep it consistent across your catalog.

Telling the AI who is buying sharpens everything. "This is for first-time buyers who are price-conscious" produces different copy than "This is for professionals who care about specifications".

For fashion, that might mean specifying women, men, or unisex. For electronics, it might mean hobbyist versus professional. The more specific the audience, the more relevant the benefits the AI leads with.

You can shape the format too. "Keep the description under 90 words, one short paragraph plus a few bullet points" is great for mobile-heavy stores. "Open with the main benefit, then supporting details, then a soft call to action" gives you a consistent, conversion-friendly structure.

For SEO, an instruction like "place the focus keyword in the first sentence and one subheading, write naturally, target 120 to 160 words" keeps the copy optimized without keyword stuffing.

Some categories need guardrails. For supplements, cosmetics, or anything health-related, an instruction like "state ingredients and usage factually, avoid medical or health claims" keeps your copy on the right side of the rules.

Because everything still goes through review before publishing, you get a second chance to catch anything the instruction missed. For a full set of ready-made snippets, see the prompt library.