Search the shape,
not just the words.

AI prior art search for product design means searching registered designs, design patents and published products by image as well as text, so a design team finds similar existing shapes before launch and can take them to IP counsel early.

Registered designs in the EU (now EU designs at the EUIPO), UK designs, and US design patents at the USPTO are indexed by Locarno classes and short product indications. Two kettles that look nearly identical can sit under different descriptions, and a keyword search will miss one of them. WIPO’s Global Design Database and the national registers offer some image search, but they cover their own collections and rarely your competitors’ unregistered products or your own past work.

So we build one index across the sources that matter to you: the public registers you are entitled to query, product imagery you collect, and your own archive of past projects, which is often the first place a near-duplicate turns up. Images and renders are embedded so that similar shapes sit near each other, and every result carries its source and registration details.

Recall is what we measure, because a missed design is the expensive failure. We build a test set from known pairs, for example designs cited in past oppositions or invalidity actions, and confirm the search returns the earlier design near the top when given the later one. Where a register cannot be queried systematically under its terms, we say so and leave it out rather than scrape it.

The same index is useful upstream: a concept from generative concepting can be searched the moment it is chosen, which is when changing it costs least. For utility patents and freedom-to-operate work across technology, the structure is similar but the judgement is different, and that is a conversation for your patent attorney.