How Yigo’s search works, and why it doesn’t use AI
When you search Yigo for “articulated skeleton frog”, you get frogs, not every articulated model on the site. There’s no AI behind that: it’s the kind of ranking classic search engines were built on. Here’s what happens between pressing Enter and seeing results.
1. Your words
First, Yigo tidies up your search. Words that appear in almost every model, such as “3D”, “model” or “STL”, are set aside. Small typos are fixed from the words that really appear in the catalogue, so “dragn” becomes “dragon”, and Yigo tells you when it has done that. Each word also matches a few others that mean the same thing: a search for “headphone holder” finds headphone stands too.
2. Finding candidates
Yigo looks for models with every one of your words in their title or tags, matching words by their root, so “frogs” finds “frog”. If that finds only a few, models with some of your words are added, leaving out words so common they wouldn’t help.
3. Putting them in order
Each candidate gets a score from a handful of simple, well-tested ideas:
- Rare words count for more. In our catalogue, “articulated” is in hundreds of models and “frog” in a few dozen, so a model that has “frog” scores much higher than one that only has “articulated”. The last word, usually the thing you want, counts a little extra.
- More of your words, in the title, close together. A model titled “Articulated Skeleton Frog” beats one that only mentions a frog in its tags.
- Missing the key word costs a lot. When plenty of models have your rarest word, ones without it drop down, which is why a frog search doesn’t fill up with sharks and crocodiles.
- Popularity breaks ties. Among similar matches, models with more likes on their own site come first, compared only with others from the same site, since sites count likes very differently.
4. Leaving out poor matches
Results that score far below the best one are left out. You may see 12 results instead of 40, but they’ll be the ones worth looking at. When nothing matches well, Yigo says so rather than showing you something random.
Why no AI?
We tried it. AI search can find models from a description, even with none of the same words. But it was also confidently wrong in ways that are hard to spot (it once ranked Yoda above an articulated fox for “fox”), every search needed a separate AI service, and it depended on monthly limits that could run out. Classic ranking is predictable: a good match is always there for a reason you can see. It’s fast, and your search words go nowhere but Yigo’s own database.
What about recommendations?
The feed and the home page’s “For you” learn which kinds of models you like, from what you search, open, save and look at for longer. That’s simple arithmetic too, done in your own browser, and never sent to Yigo. You can see what it has learned, delete it, or turn it off on the Favorites page.
Try it
A few searches that show it off: articulated skeleton frog, headphone holder and dragn. For tips on writing your own, read how to search for 3D models.