Fill in incomplete product records from the page that publishes them
Recovers the data your catalog is missing — price, images — from the original product page, and normalises the extracted values.
Before and after, on the same record
- Before: no price, no images, "Esaurito".
- After: €129.00, three images, normalised availability.
- Fields you had already filled stay as they were.
What it actually does, step by step
What an imported catalog looks like
- Products with no price.
- Records with no images.
- Availability written five different ways.
The problem: catalogs full of gaps
An imported catalog brings along whatever it had to begin with, and that is almost always incomplete: products with no price, records with no images, availability written five different ways because it passed through three systems. A chatbot reading that catalog inherits the gaps: asked "how much is it?", it cannot answer if the price is not there, and it has no way to invent one — which is right, but does not help the customer.
Recovering data from the page that already has it
The information missing from the catalog is, almost always, published on the product page itself. This feature goes and reads it, and brings back what is needed: the price where it is missing, the images where there are none. Recovered images are merged with the ones already there rather than replacing them, and fields that already have a value are left alone — the existing catalog stays the primary source, this work fills the holes.
Normalising what arrives written ten different ways
"Esaurito", "out of stock", "unavailable", "sold out" mean one thing to a person and four things to a program. Extracted values pass through a normalisation layer that brings them back to a single form, configurable per domain when a site has conventions of its own. It matters because inconsistent availability is worse than missing availability: missing is visible, inconsistent produces confident wrong answers.
Why it shows up in the chatbot’s answers
All of this exists for a practical reason: retrieval works with the data it has. A record with no price and no images is a record the chatbot can say little about, however good the model behind it. Completing the catalog before indexing it raises answer quality more than any downstream tuning, because it acts on what the assistant has to read in the first place.
Does it overwrite the data already in my catalog?
No, it fills the gaps. Fields that already have a value stay as they are, and recovered images are added to the existing ones rather than replacing them.
Does it rewrite titles and descriptions with AI?
No, and we would rather be plain about it: there is no automatic text generation and no SEO optimisation of content. What it does is recover from the page the data your catalog is missing, and normalise the extracted values.
Does it work on any site?
It works on product pages that are reachable and readable. Normalisation rules can be configured per domain, which is exactly what sites with their own conventions for availability and pricing need.
Ready to answer in chat with your own data?
Free trial, no credit card. Set up in minutes.