You ask an AI to write the product page for something you know the name of and little else. And it does: fluent, convincing, list of benefits included. The problem is that you have no record of half of what it claims. It made it up. Welcome to "AI slop": content that sounds good and says little, or says things that are false.
On a personal blog, a made-up sentence is no big deal. In a product catalog, it's a return, a complaint or a legal problem. That's why the step that really separates a good AI-written product page from a bad one isn't the prompt: it's researching before writing.
An AI doesn't know; it predicts
A language model isn't a database of facts: it's a machine for predicting the most likely text. When you give it rich information, it predicts from that and gets it right. When you give it little — a supplier title and an EAN — it fills the gaps with what "usually goes there": typical ingredients, common materials, plausible compatibilities. It sounds perfect. And it can be completely false.
Why that's dangerous in a catalog
A product page isn't decoration: it's a promise. If it states something, the customer buys believing it. A made-up detail turns into:
- Returns and complaints: the product isn't what the page promised.
- Legal risk: allergens, regulations, health claims, compatibility.
- Lost trust: anyone who spots one error stops trusting the rest of the catalog.
- Worse SEO, not better: generic or duplicate descriptions don't rank; Google rewards useful, specific content.
The fix isn't a better prompt. It's researching first
The temptation is to polish the prompt: "don't make things up", "be precise", "stick to the facts". It doesn't fully work, because the model can't tell what it knows from what it assumes. The real fix is to change the order: first gather verified facts, and only then ask the model to write from them. The model stops imagining and starts drafting.
It's the difference between asking someone to write about a product they've never seen and handing them the spec sheet before they start. The second person writes better because they're writing about something real.
How Noobit does it
Noobit can research each product before writing a single sentence. You turn research on for the whole catalog in «Preferencias» (Preferences), or run it yourself from a single product record. It gathers what's known about the product from real sources: the data that comes from the supplier, barcode databases (for example, Open Food Facts for food products) and the web. Each fact is saved with the page it comes from, and with those facts on the table the product page gets written, building on them and on your brand voice. And some data is never written without a source: an ingredient list that doesn't appear anywhere in the product's data is removed from the page.
First, the manufacturer's product page
A search engine doesn't always bring back the right page: sometimes it cites the brand's catalog, but not the product page. So when the manufacturer's website is among the sources, the research opens that site's pages and looks for the link to the exact product page. If it's sure which one it is, it puts it first; when in doubt, it adds nothing. A gap is better than a neighboring product's page.
That page is read on its own. The AI gets only its text, and every fact it returns is checked against the page before it's accepted: the quote has to be on the page, the figures in the fact have to be in the quote, and the fact has to come from it. Anything that fails the check doesn't get in.
What's learned also adds up: whatever a new research run doesn't find again is kept instead of lost. If you know the right page, you can paste it in yourself and it's read first. And for each product you see what was consulted, page by page, and you can exclude sources, starting with your own store if it carries errors.
How to tell the difference
You don't need to be an expert to spot slop. Signs that an AI is making up the product page:
- Empty superlatives with no facts behind them ("the highest quality", "revolutionary technology").
- Detailed specs that don't appear in any of the product's sources.
- The same description, copied word for word, on different products.
- Claims about compatibility or composition that nobody has verified.
A well-made product page is exactly the opposite: specific, consistent with the spec sheet and verifiable. If a fact is there, it's because it came from somewhere.
The golden rule
If your AI can't say where a fact comes from, it shouldn't state it. That's the whole philosophy. AI is an extraordinary tool for writing, structuring and scaling, but from facts, not in place of them. Research first, write second, and your catalog will stop being a collection of nice-sounding text and become a collection of product pages you can trust.