Dima Persiianov
We shipped three improvements to AI Company Search this month. Most of them came from watching how people actually use search - it has become our most-used feature, so we went deep on making it better.
Here's what changed.
Smarter Deep Search Filtering
Deep Search now runs a lightweight relevance filter before building your final list.
Some context: on average, we source around 1,500 candidates per query. They come from the web, our own database, maps, and LinkedIn. That's a lot of raw signal, and not all of it is useful. Turns out about 60% of those candidates can be discarded right away - they don't match the intent of the query closely enough to deserve a spot.
Before this update, some of those weaker results would slip into your final list and take up space. Now they get filtered early, and every slot goes to a stronger match. The result is a cleaner, more relevant list without you having to do the filtering yourself.
Lookalike Search Upgrade
Lookalike Search now blends product, business model, firmographics, and website traffic into a single ranking. Previously these signals were weighted separately, which meant results could skew toward one dimension. Now they're combined - so when you hit "find similar," the top results actually resemble the input company across multiple dimensions, not just one.
Better matches surface higher. Less scrolling.
If you want to go further, we put together a guide on using the Lookalike API with Claude Code to build prospect lists programmatically in minutes: Lookalike API + Claude Code Guide
10M+ Company Profiles in Our Index
Our index now contains over 10 million company profiles. Each one includes what the company does, firmographics, website traffic estimates, and socials - no extra enrichment step needed. This is what powers our embeddings, and it's what you get out-of-the-box with every search.
We're deliberate about what goes into the index. From what we see in our data processing pipelines, most of what inflates those "100M+ company" databases is dead entities, shell companies, and profiles with nothing useful attached. We'd rather give you 10M companies you can actually work with than 100M you have to filter yourself.
We grow the index based on what users actually need, and we focus on maintaining the highest quality we can - not just factual firmographic accuracy, but semantic accuracy too. When Extruct says a company does X, it should actually do X.
Everything Is Live
All three updates are available now in the web app, Chrome Extension, and API.
What We Published This Month
Two research pieces from the team:
- Data-Driven VC: How Funds Are Building Product Teams - we mapped 114 funds that now hire product managers across their investment platforms.
- Industrial Tech Thesis - our breakdown of AI companies targeting manufacturing, logistics, and the physical economy.
If anything looks off or you have questions, reach out - we're always happy to hear what you're building with Extruct.
