Once you have more than a few dozen products, writing detailed descriptions quickly falls behind. Handing it to AI speeds things up, but in practice it sometimes gets the material wrong or adds lines like "the highest quality." With auto-generated product descriptions, the order matters more than the writing skill. Here are the three things to decide first, in order.
1. Organize your product data into a table first
AI is only accurate within the information it receives. Before handing over descriptions, check that each product has the following items in a table.
| Item | Example | What happens if it's empty |
|---|---|---|
| Material, contents | 100% cotton, body + pouch | It makes up the material |
| Size, options | S/M/L, 3 colors | It lists options that don't exist |
| Use, care | Hand wash recommended | It fills in generic advice |
| Cautions | Do not tumble dry | Caution notes go missing |
If your products are already registered in a store platform, you don't need to enter the data again. For example, Cafe24 provides admin REST API documentation in its developer center, so you can build a flow that pulls registered product information and passes it to AI. While building online stores with Cafe24 API integration and PG payments, we also saw that the cleaner the product data, the easier the next round of automation.
2. Write your sentence rules first
If you only tell the AI "write it well," the tone and length change from product to product. Set rules like the ones below and fix them in the prompt.
[Tone] Polite, friendly but without exaggeration
[Length] 2-sentence summary paragraph + 3 features (1 line each) + 1 line on care
[Basis] Don't write anything that isn't in the table. If you don't know, leave it blank.
[Banned] Expressions like best, No. 1, 100%, perfect, guaranteed"Don't write anything that isn't in the table" is especially important. Without this line, AI tries to fill blanks with plausible sentences. You don't need new rules for every product, but for categories with very different cautions, such as clothing, food and household goods, it helps to keep a separate [Banned] line.
3. Filter exaggerated advertising with a checklist
Article 3 of Korea's Act on Fair Labeling and Advertising prohibits false or exaggerated, deceptive, unfairly comparative and disparaging labeling and advertising. Whether a person or AI wrote it, a sentence on a product page is the seller's advertising. Check the following before publishing auto-generated text. (This is a practical checklist, not legal advice.)
- Did it use expressions like "best," "No. 1" or "only" without evidence?
- Are there sentences that assert efficacy or effects?
- Did a sentence comparing with other brands slip in?
- Did any number appear that isn't in the data table?
Finally, publish after a person reviews it
For the first month or two, don't post AI descriptions right away. Stick to draft, staff review, publish. Showing the original data table and the AI text side by side on the review screen makes errors much faster to spot. When you see parts the reviewer fixes often, put them back into the sentence rules. After a few rounds, the number of fixes drops noticeably.
References
- Korea Law Information Center, Act on Fair Labeling and Advertising, Article 3: https://www.law.go.kr/LSW//lsSideInfoP.do?lsiSeq=268659&joNo=0003&joBrNo=00&docCls=jo&urlMode=lsScJoRltInfoR
- Cafe24 Developers, Admin REST API documentation: https://developers.cafe24.com/docs/api/admin/
In a one-hour AX consultation we map AI candidates and similar cases for you.
We build web apps, mobile apps and business systems, then add and run the AI they need. We build and operate our own meeting notes, contract and company analysis AI.
See our projects