AI for E-commerce: Product Pages That Sell (and the London Goldsmiths' Mark)

You have three hundred products online and half the pages say "High-quality product, perfect for any occasion." AI can rewrite them all in an afternoon. The trouble is that it can also invent, just as elegantly, a material the product doesn't have. This guide shows how to get the speed without losing the truth.

The leopard's head mark

London, 1300. A statute of Edward I requires silver objects to meet the "sterling" standard and, before leaving the workshop, to be assayed by the wardens of the craft: only then do they receive the leopard's head mark. In 1363, under Edward III, a second rule follows: every goldsmith must also strike his own personal mark, so it is known who made the piece. This is the origin of the hallmarking system that London's Goldsmiths' Company still runs today.

Notice the logic. The mark said nothing about how beautiful the cup was. It said two things: the content is what is declared, and this is the name of whoever answers for it. The buyer, who couldn't melt the cup down to check, trusted the mark.

Your product page is that mark. The customer can't touch the sweater, measure the bag or read the cosmetic's label. They trust what you write. If AI writes "100% organic cotton" and the supplier says "cotton and polyester," you don't just have a wrong sentence: you have a return, a bad review and, at worst, a misleading commercial practice.

What AI does well in a catalog (and what it doesn't)

AI is excellent at turning technical data into readable copy, harmonizing tone across hundreds of pages, creating variants (short, long, bullets, social), translating (with native review for key markets) and flagging missing attributes.

It is risky when it has to "complete" information it doesn't have (sizes, certifications, composition), make promises it can't verify (durability, health benefits), or produce near-identical text for different products.

The method in five steps

1. Build the "assay sheet"

Before opening any tool, put only verified data in a spreadsheet for each product: name, SKU, materials, dimensions, weight, colors, certifications, care instructions, who it's for, three real selling points. This is your single source of truth. If a field is empty, it stays empty.

2. Write a prompt with explicit rules

Write the product description using only the data provided. Do not add features, certifications or benefits that are not in the data. If useful information is missing, end with "MISSING DATA:" and a list. Tone: direct, concrete, second person. Structure: one opening line on who the product is for, three bullet points on benefits, one paragraph on materials and dimensions, one line on care. Maximum 120 words.

The "MISSING DATA" line is your assayer: it shows where AI would otherwise have had to invent.

3. Generate in batches, sample-check, then check everything

Start with ten products, read them carefully, fix the prompt, then scale. Before publishing, someone who knows the product must reread every page, at least the risky parts: composition, sizes, compatibility, promises.

4. Optimize for search without cutting corners

• Title with product type, brand or line, and a distinguishing attribute.
• A unique description per product, answering real customer questions.
• Product structured data (price, availability, reviews) consistent with what's on the page.
• Descriptive image alt text, reviewed.

Google's guidance on AI-generated content asks you to check the accuracy of text, metadata and structured data, and warns that generating many pages without added value may violate its scaled content abuse policy.

5. Label where required

If you use Google Merchant Center for Shopping, the same guidance states that AI-generated titles and descriptions must be specified separately and labeled as AI-generated, and AI-generated images must carry the appropriate IPTC metadata. Check your feed settings before uploading the rewritten catalog.

Which tools to use

Check current prices on official sites before choosing.

Mistakes to avoid

1. Starting from photos or product names alone.
2. Bulk publishing without proofreading.
3. The same text on every variant.
4. Undocumented health or "green" claims.
5. Forgetting old pages: update first the products with the most traffic and fewest conversions.

A typical scenario

Picture an online homeware store with 400 products run by two people. Today a new product page takes half an hour. With the assay-sheet method, time shifts to collecting verified data (needed anyway), and writing becomes a draft to review. The real gain depends on supplier data quality and should be measured on the first fifty pages.

Read next

This guide is part of our dossier on AI for small business marketing. To get your pages cited in Google's AI answers, read SEO and AI Overviews; to get the right products to the right customers, see AI email marketing.

London's goldsmiths spent centuries building trust in a mark. You can lose it with one rushed page. If you run an online store, see what we do for e-commerce and book a free intro call.