Smarter Matches: How Semantic Search Is Changing the Way Recruiters Find Workers on Skillit

If you've ever searched for a Journeyman Electrician on Skillit and ended up with an apprentice, or typed "CAT 395" and got a list of workers who've never touched a 90-ton excavator, you already know the disconnect at times with keyword-only search. Exact words matter, but the underlying meaning is significantly more important. 

We've been quietly upgrading our Semantic Search feature alongside the existing keyword ranking system, and the results are worth talking about.

Here's the simplest version of what it does: instead of only looking for candidates who wrote your exact search terms on their profile, it understands what you're actually looking for and finds workers who fit, even when the words don't match perfectly.

Take one real example from our testing. A recruiter at a carpentry firm was searching for cabinet installers using keywords like "drawer," "pantry," and "cabinet installation." The old system ranked a moonlighting cashier in the top 25. While this person technically matches on the trade filter, they were obviously not a fit. Semantic Search pushed the worker out and promoted two experienced carpenters with back-to-back framing roles on their profile. None of them wrote "pantry" on their profile, but they were clearly the right people.

We tested this across 34 searches. The top 25 results stayed the same 86% of the time - so the system wasn’t broken, it just needed to be optimized for better performance. Out of the 14% that did change, the swaps were the right call. Workers added by the AI actually had more of the recruiter's keywords on their profiles on average (0.49 per worker) than the workers who got pushed out (0.31). In other words, it's not trading keyword precision for some fuzzy "semantic" guess, it's finding candidates who are more literally and contextually relevant at the same time.

The feature works hardest on the searches that need it most: technical jargon, model numbers, polysemous words like "Mechanic" (auto? heavy equipment? marine?), and industry context that keyword scoring simply can't read. On searches that are already well-formed and precise, it leaves the results alone. Of 34 searches tested, 15 had zero changes at all.

The bottom line: finding the right worker shouldn't require knowing exactly how they described themselves. Semantic Search closes that gap and based on what we've seen so far, it's doing it without sacrificing the literal relevance that makes search trustworthy in the first place.

Try out Semantic Search and see the improvements in results for yourself.