Can AI Write Product Copy and Predict Returns? For Webshops

Can AI write product descriptions and predict returns?

AI can write product descriptions from your catalogue data and predict returns from your order history, and the two together attack the most expensive line in e-commerce.

Both, and they are the same problem seen from two sides. A product description that says what the product actually is, in the customer's language, with the sizing and material facts filled in from the catalogue, is the first return-prevention tool a webshop has. A model that predicts which order will come back tells you which product pages, sizes and customers to fix first. This guide covers both, with what they cost.

By Tom Joseph · Last updated: 14 September 2026

Product descriptions that are true, in every language

Written from the catalogue, not from imagination. Material, dimensions, fit, care, compatibility and certifications come from your PIM or ERP fields; the model writes around facts, it does not invent them.

In the house voice, per channel. Your own shop, Bol.com, Amazon and Marktplaats each get the length and structure they rank on, from one source of truth.

Sizing that prevents the return. Where fit data exists (measurements, reviews saying 'runs small'), the description says so, which is the least expensive return prevention there is.

Reviewed before it is live. Every generated text is checked against the source fields, and a person approves the first batch per category before the rest is published.

Predicting which order comes back, and why

Signal	What the model uses it for
Order composition	Two sizes of the same item is a fit test, not a purchase; the order is flagged before it ships
Product history	Items with a high return rate for a stated reason (size, colour, quality) get their page fixed first
Customer history	Serial returners are handled by policy, with a person deciding, not by silent penalties
Page and photo	Descriptions and images that correlate with 'not as expected' returns get rewritten and reshot
Return reason text	Free-text reasons are classified so the buyer sees which supplier or size chart is the problem

What AI does not fix in a webshop

It does not fix a catalogue with empty fields; if the material and the measurements are not in the PIM, no model can write them truthfully, and the audit starts there. Predicting a return does not prevent it; the value is in what you change on the page, the size chart and the supplier. And return policy is a business decision: a model that flags a customer is an input for a person, not a rule, which is also what fairness under the GDPR and the AI Act's transparency duties expect.

What it costs a webshop

Three fixed steps: a €2,500 audit on a year of orders and returns, which shows where the returns come from by product, size, page and customer; a €20,000 Production-ready MVP, either product copy for one category or return prediction for one range, measured against the current numbers; and production from €50,000, integrated with the PIM, the shop platform (Shopify, Magento, Lightspeed, Shopware) and the WMS. Excl. VAT, all code handed over. See AI for e-commerce and the Bol.com bid optimisation case.

Frequently asked questions

Can AI write product descriptions for a whole catalogue?

Yes, from the fields in your PIM or ERP, in the house voice and per channel, with every text checked against its source fields. The first batch per category is approved by a person; after that it runs per new product.

Can AI predict returns before shipping?

It can flag the orders most likely to come back and why, from order composition, product and customer history. What you do with the flag is a policy decision; the model informs it, a person decides.

How much can returns be reduced?

That depends on how much of your return rate is fit and expectation, which the audit measures from a year of return reasons. In fashion, where size and fit drive 70% of returns, the page and the size chart are the lever; the honest claim is measured per category in the MVP.

Does it work with Shopify, Magento or Lightspeed?

Yes. Copy is written into the PIM or the shop platform, predictions into the order flow or the WMS; the integration is scoped in the audit.

What does it cost?

At Crux Digits: a €2,500 audit, a €20,000 Production-ready MVP for one category or range, and production from €50,000, all excl. VAT, with the code handed over.