AI product descriptions that fit your shop

What a generic AI description generator gets wrong

A standalone AI description generator is quick: you paste a product title into a chat window and get a smooth product description back. For a blog idea that is fine. For product data it often goes wrong, because the tool knows nothing about the product you actually sell. It fills the gaps with whatever sounds plausible.

Three mistakes you see again and again

• Invented specs. Ask for a text about a 20 litre bike bag and you may get a waterproof zip, reflective strips and a long warranty thrown in. If one of those is wrong, you get returns, customer questions or a complaint that your listing is misleading.
• Hype instead of information. "The ultimate", "revolutionary", "a must have": words that say nothing about size, material or use. A shopper choosing between two coffee grinders gets more from the number of grind settings than from three exclamation marks.
• The same text as your competitors. Anyone who pastes the same title into the same tool gets a text that looks a lot like yours. Your shop then stands out nowhere.

On top of that, with hundreds of products you will not check every text line by line. One invented detail turns into a pattern across your whole range.

That is why the AI in StoreLinkr AI product enrichment works differently: it only writes from facts that are known, and you decide how those facts sound. For how to structure a good text, with or without AI, read how to write product descriptions.

How StoreLinkr's AI works: facts only

Enrichment starts from the product's EAN. With it, StoreLinkr looks up data in the product catalogue, in Google Books for books (an ISBN starting with 978 or 979), and uses what your supplier or POS import already delivers. The EAN itself is never changed, and a product without an EAN cannot be enriched. Then your own AI writes the text. There are three cases.

1. A description exists: rewrite

If the supplier or source has a description, the AI rewrites it faithfully. Every fact stays, no new claims are added, and no superlatives appear that the source does not use. The length stays close to the source. So you get your own text in your tone, with the same content.

2. No description anywhere: write from facts

If you switch on the setting to write a description with AI when no source description is found, the AI writes one from the known facts: title, brand, attributes, category and condition. When there are few facts, the text is short. A short text that is correct beats a long one with inventions.

3. Look up online sources (optional)

With the option to look up online sources, the AI searches the web for that exact product and writes 150 to 200 words from reliable pages. The source URLs are kept, so you can check where a claim comes from. If the AI finds no reliable source, it writes nothing from the web and falls back to the facts-only text. This option costs more tokens and only works with models that can search.

Next to the long text, the AI creates a short description: condensed from the long text, plain text and at most 160 characters. An existing short description of an imported product is never overwritten.

Steering the text: tone, audience and own instructions

The same facts can be written down in ten ways. In the AI settings you decide which way fits your shop.

• Tone of voice. Choose professional, informal, persuasive or humorous. A bike shop for commuters sounds different from a gift shop.
• Target audience. Describe who you write for, for example "parents of young children" or "home baristas buying their first grinder". The AI then decides which facts to put first.
• Keywords. Add words the AI should emphasise, such as "water resistant" or "adjustable". For choosing keywords, read SEO product descriptions.
• Language. The preferred language can be English, Dutch, German, French or Spanish.

New: own instructions for product texts

If you want every text to follow the same structure, write it in the own instructions field. That is free text, up to 2000 characters, added to every product description request. Examples that work well:

• "End with a bulleted list of the key specifications."
• "Always mention the material and the dimensions."
• "Close with one sentence on what is in the box."
• "Do not use exclamation marks."
• "Use British spelling and address the reader as you."
• "Start with what the product is for, not with the brand name."

Why instructions never lead to invented facts

Own instructions shape structure and wording. They never outrank the rules: the AI skips any part that would need a fact that is not given. Ask for "always mention the material" when nobody knows what the bike bag is made of, and the text simply has no material in it. Ask to "close with what is in the box" for a children's book without that information, and that sentence is left out. Better a gap you can fill in yourself than a mistake that reaches your customer.

Reviewing before an AI text goes live

The AI does the writing, you keep control. You always see what changes.

• Preview per field. When you enrich a product, you see the old and the new value of each field side by side. You accept field by field: the new category yes, the description not yet.
• Marked as AI generated. Text the AI wrote gets that label. When you review, you know which texts deserve a closer look, and you can still refuse them.
• Enrichment log. The product enrichment page lists every enrichment: manual, bulk, file import or automatic. You see the value before and after, the sources, the source URLs and whether AI wrote the text. You can filter on AI-written texts.
• One click back. If a text does not fit after all, you revert the enrichment with one click.

A way of working that holds up

1. Enrich about ten products from different categories first.
2. Read the texts and note what you keep changing, for example an opening that is too long or a missing spec list.
3. Turn those corrections into rules in your own instructions.
4. Then select the rest in the product overview and enrich them in one go.
5. Filter the log on AI texts and spot check a sample.

Costs, your own API key and imported products

Your own AI, your own bill

StoreLinkr works with your own API key for Gemini (Google), OpenAI (ChatGPT) or Claude (Anthropic). You choose the model and pay the AI provider directly. A daily token limit caps the costs. Keep in mind that looking up online sources uses more tokens than rewriting or writing from facts.

Supplier products keep your text

If your products come from a supplier, wholesaler or POS import, the obvious question is: will the next import simply overwrite your new text? No. Accepted texts are stored as product rules, so the next import leaves them alone. Attributes, weight and dimensions stay as the supplier sends them.

Automatic for new products

You can also enrich automatically. Products that an import creates, from the import sources you choose, are enriched once. Only empty fields are filled, by default the name, descriptions, category and images. With the option to enrich first, then send to the shop, a new product waits for its enrichment before it goes to your sales channels, at most 15 minutes after the import. StoreLinkr then sends the data to your webshop (WooCommerce, Shopify or SitePack) and to marketplaces such as Kaufland, eBay and Allegro.

If attributes, categories or images are missing as well as texts, read product data enrichment. The full overview is on the AI product enrichment page, and everything about keeping product information in one place is under manage product data.

Frequently asked questions