A store owner with three hundred products and no descriptions once had two options: write them all by hand over several months of evenings, or use the text provided by the supplier. Most chose the supplier text. A third option now exists, and many businesses are adopting it at the same time.
Shopify added a description generator to its platform in February, under the name Shopify Magic. WooCommerce stores have access to numerous plugins that do the same through the OpenAI API. Many people also simply open ChatGPT in another tab and paste in product names one at a time.
The approach works, with qualifications. We have been testing it on real catalogues this spring, and the results range from excellent to embarrassing, depending almost entirely on the information provided.
Where do AI product descriptions actually help an online store?
The clearest benefit is speed on repetitive work. A good first draft in thirty seconds rather than fifteen minutes, multiplied by a few hundred products, represents weeks of work saved for a small store. The gain is real.
It also helps overcome the blank page. Many small business owners know their products thoroughly but struggle when they need to write about them. An AI draft gives them something to respond to. "No, that is incorrect: it is not waterproof, it is water-resistant" is a much easier sentence to write than an entire paragraph from scratch.
It is also useful for the variations no one wants to write: the short version for the category page, the meta description, the social media post and the marketplace listing. The facts remain the same; only the length changes.
It helps less than expected with French content. In Quebec, a commercial website must be available in French; this is a legal requirement. The models write in French, and the quality has improved over the past year, but the result often reads like European French or like a translation. "Courriel" becomes "e-mail," the tone is inappropriate, and Quebec customers notice. We recommend planning for more extensive editing of the French than of the English, or entrusting it to someone who writes Quebec French. Professional bilingual copywriting remains, for now, a task for people.
What information should be provided to the AI?
Real product facts, not a bare name. This is the decisive factor.
Typing "write a product description for a merino wool toque" produces an attractive paragraph about warmth, softness and "the perfect companion for your winter adventures." Every store selling a merino toque will receive nearly the same paragraph, which says nothing about the specific product.
Instead, provide what you actually know:
- Material and blend, exactly as stated on the label.
- Where it is made, and by whom if this matters to your customers.
- Sizes and measurements.
- Care instructions.
- What customers have said about it, in their own words, from your reviews or emails.
- Who it is for, and who it is not for.
- Anything that distinguishes it from the version sold at a big-box store.
The output changes completely: it now describes a product rather than a category. We use a simple spreadsheet for this purpose, with one row per product, one column per fact and a prompt template that draws on each row. This requires more time initially, but it is the only method we have found that produces descriptions worth keeping.
An additional benefit is that completing the spreadsheet reveals how little information is available for some products, which is valuable to know in any case.
Why should AI drafts be edited to reflect your own voice?
Because AI-generated copy has a recognizable style. Once identified, it is difficult to overlook: everything is "perfect for," "take your kitchen to the next level" or "for beginners and experts alike," with three adjectives in a row and an exclamation mark at the end.
Customers may not be able to name it, but they perceive it, and the store begins to sound like every other store.
We recommend reading each draft and asking whether you would say this aloud to a customer at your counter. If not, revise it. Remove unnecessary adjectives. Add the specific detail only you know ("the grey runs a bit darker than the photo"). Use your own words. If you have a particular way of describing your products, including humour or a regional expression, keep it. This is what a generator cannot provide, and it is what makes customers remember a small store.
In our view, fifty products with descriptions that sound like a person are preferable to three hundred that sound like a brochure.
Why must every claim be verified?
Because inaccurate claims can cause real harm.
These tools invent information, not often, but with confidence. We have seen generated descriptions add "machine washable" to an item that is not, invent a certification, round off a size or describe a feature belonging to a different product with a similar name. The text reads perfectly well, which is precisely the problem: no one notices until a customer returns a shrunken sweater.
Incorrect claims on a product page lead to returns, negative reviews and, in some cases, legal issues. Misleading product information remains misleading, regardless of who or what wrote it.
Every fact in every description must therefore be checked against the label or the supplier's specification sheet: materials, dimensions, weight, care, origin, compatibility and anything involving a number. The process is tedious, but it remains much faster than writing from scratch, so the net result is still positive.
What is Google's position on AI-written content?
Store owners frequently ask whether Google will penalize a store for using AI. Google addressed this question in February in a post on Google Search's guidance about AI-generated content. In summary, Google evaluates whether content is helpful, not how it was produced. Using automation primarily to manipulate rankings remains spam, as before.
For product pages, we interpret this as a caution rather than an endorsement. Three hundred generic descriptions resembling those of every competitor are unlikely to result in a penalty, but they will not improve rankings either, since they add nothing. The pages that perform well tend to contain specific information people actually search for: measurements, real use cases and answers to questions customers ask.
The qualities that make a description sell also make it rank: accurate facts, expressed in your own voice.
For stores still relying on supplier text and considering a thorough revision, this is also an appropriate time to review the rest of the setup, including categories, photos and checkout. Descriptions are one component of an effective online store. We recommend starting with the twenty best-selling products and completing those thoroughly before addressing the rest.