Making AI Product Copy Sound Like Your Shop
Generic copy comes from the way these tools work. A model that predicts the next word1 leans toward the wording it has seen most. So a plain request gets the most average description going.
You can pull it back toward your shop. It takes samples, a short voice note and an edit.
Why every draft sounds alike
Large language models are trained on a vast amount of text1. Ask for "a description of a candle" and you get the common shape of candle descriptions. Warm glow. Cosy evenings. Perfect gift.
Those phrases aren't facts. They're filler, and they can slide into claims. "Perfect for any room" says something about the product that you haven't checked.
Three ways to give it your voice
Putting a few examples inside the request is a known technique, and the model can learn from them in-context2. That's the main lever. Here are three approaches side by side.
| Voice note only | Samples only | Samples plus voice note | |
|---|---|---|---|
| What you give | A few lines on tone | Two or three of your own descriptions | Both |
| Strength | Quick to write | Shows rhythm and word choice | Most control |
| Weakness | Tools read "friendly" in their usual way | Facts from samples can leak into new drafts | Takes longer to set up |
| Suits | One-off items | Shops with writing already published | A shop with many products to write |
A voice note might say: short sentences, no exclamation marks, say "stitched" not "crafted", never say "perfect". Name the words you dislike. A tool follows a banned list better than a vague mood.
Edit for what only you know
A draft can match your tone and still read like anyone's. What lifts it is a detail the tool couldn't know. Maybe why you chose this supplier, or how you pack orders. Add one or two of those yourself, true ones only.
Then cut:
- Openers that praise the product before describing it.
- Stacks of adjectives with nothing behind them.
- Any "perfect for" line you can't back up.
- Repeated sentence shapes. Read it aloud. If three sentences in a row start the same way, change one.
The ACCC lists "use easy to understand language" among the things a business should do3. Plain beats clever.
One thing to keep out of your voice
Don't ask a tool to write customer comments, quotes or star ratings. The ACCC says reviews should be independent and reflect the genuine opinion of the person who experienced the product or service4. It also says it's against the law in Australia to create fake or misleading reviews4. Rules elsewhere differ, so check your regulator.
The same goes for invented stories. "I spent years perfecting this recipe" is a claim about you. If it isn't true, it doesn't go in.
Puffery has limits
The ACCC treats puffery as wildly exaggerated, vague claims that no one could treat seriously, and generally doesn't consider them misleading3. Don't lean on that. A line like "keeps drinks cold all day" is specific enough that people can test it. Voice is about rhythm and word choice. Facts still come from your sheet, as in [page:product-fact-sheet-template].
Cited material
- Large language model, Wikipedia
- Prompt engineering, Wikipedia
- False or misleading claims, Australian Competition and Consumer Commission (ACCC)
- Online reviews for product and services, Australian Competition and Consumer Commission (ACCC)