A product page has two kinds of text. There's the text you write — the description, the benefits, the title — and the text you copy: a cosmetic's ingredient list, a food's allergens, an appliance's warnings. The first leaves room for style. The second can't take a single extra comma: it has to say exactly what the packaging or the manufacturer says.
The problem is that, to an AI, the two look very much alike. And if it doesn't have the ingredient list, it writes one.
Why an AI makes up an ingredient list
A language model has seen countless shampoo ingredient lists. Ask it for a shampoo's product page without giving it that shampoo's list, and it fills the gap with the most likely one: water, glycerin, a couple of surfactants, fragrance. It looks perfect, and it may have nothing to do with the product. If the list leaves out an allergen the product does contain, the mistake is no longer a matter of style.
A cosmetic's ingredient list and a food's allergens are regulated information. They aren't something you can write: you either copy them or leave them out.
What gets copied, word for word
That's why Noobit separates what gets written from what gets copied. In Product Knowledge, regulated data — the ingredient list (INCI), allergens, warnings, dosage, nutrition information — is stored verbatim: exact, just as it comes from its source, with no paraphrasing or reordering. When the product page is written, this data is marked as such: it gets copied, not rewritten.
Web research handles it separately too: it looks for this data in a pass designed just for it, and it only accepts verbatim text if it matches what the page says.
What gets removed
The other half is not letting through anything that isn't backed up. Before saving a product page, Noobit looks for ingredient lists in the text and compares them with the product's data. If a list doesn't appear anywhere in that data, it's removed. Better for it to be missing than for a wrong one to go out: the gap gets filled as soon as the right data turns up.
What data each product type requires
Not every product requires the same data. A cosmetic needs its ingredient list; a food, its allergens and nutrition information; an electrical appliance, its power rating and recycling information; a garment, its fiber composition. Noobit keeps track of the regulations for each product type: it knows what data each case calls for, both mandatory and recommended, and flags what's missing from the product page. We turn this check on with you when you get started.
The underlying rule
A product page can be more or less attractive. What it can't do is say something the product doesn't. Writing with style what can be written, copying exactly what has to be copied and leaving blank what isn't known: that's the difference between generated text and a correct product page.