Product descriptions with AI: templates and quality checks

Writing descriptions for three hundred products by hand is work no small online store can sustain, and the typical result is copy-pasting the supplier's sheet, identical to that of ten competitors. AI can solve the problem, but used badly it makes it worse: it generates product descriptions that all look alike, bloated with adjectives and missing the information buyers actually need. The difference between the two outcomes lies in the workflow, not the tool.

Why badly generated descriptions are instantly recognizable

If you ask an AI model to write the description of a leather bag giving it nothing but the product name, you'll get a plausible, empty text: elegance, versatility, attention to detail. Multiplied across an entire catalog, it produces nearly identical pages that don't help the customer choose and give Google no reason to prefer you: duplicate or copy-paste content tends not to rank. The principle to fix before any prompting is this: the AI doesn't know your products, it can only describe them from the data you provide. Output quality depends on input quality.

Before the prompts: the source data

For each product, prepare a fact sheet, even just in a spreadsheet: materials and composition, dimensions and weight, available variants, origin and manufacturing, who it's for and on what occasion, what sets it apart from similar products in your own catalog, any certifications, what's in the box. Add the questions customers ask most often about that type of product: you'll find them in support emails and in reviews. This work is the real bottleneck, and it can't be delegated to AI: but done once, it feeds descriptions, spec sheets and even customer service replies.

A prompt template that works

A good prompt for product descriptions always contains the same blocks:

• the role and context: who you are, what the store sells, who it's aimed at;
• the tone of voice, with two or three examples of descriptions already written in the style you want;
• the product facts, from the sheet above, with the explicit instruction not to add characteristics that aren't listed;
• the output structure: for example a benefit-oriented opening paragraph, a list of concrete features, a line about shipping or product care;
• the prohibitions: generic superlatives, stock phrases, repeating the product name in every line;
• the expected length and the format (plain text, HTML, markdown) ready for your CMS.

Save the template and reuse it varying only the facts: that's how you get style consistency across the whole catalog, with texts that differ from one another because the source data differs.

Quality checks before publishing

No generated text goes live without human review. The minimum check we apply in the eCommerce projects we manage covers four points. Accuracy: every claim must match the fact sheet; models sometimes invent plausible characteristics, and on materials, dimensions or certifications an error turns into a return or a dispute. Usefulness: does the description answer the questions the customer asks before buying? Uniqueness: sample a few descriptions and verify they aren't interchangeable between different products. Tone: the voice must stay yours across the whole catalog. For large catalogs, a sample-based process with clear thresholds works best: full review of the main categories, spot checks on the rest.

The product page isn't just text

The best description delivers little inside a slow product page with no structured data and poor photos. Title and meta description, structured data markup for Google's rich results, optimized images, visible reviews: it's the whole that ranks the page and converts the visit. That's why we treat AI-generated content as one piece of the eCommerce project, not the entire solution.

A catalog that works for you

If your store has dozens or hundreds of anonymous product pages, a well-built workflow of AI generation plus human review can turn them into pages that rank and sell. In our websites and eCommerce service we set up the complete workflow: source data, templates, quality checks and technical optimization of the product pages. Book a free call and let's look at your catalog together.