New AI Tool Finds Apps That Actually Do the Same Thing

Search for a budgeting app today and you'll get a flood of results — but many of them just mention the word "budget" somewhere. They don't actually do what you need. The same problem hits researchers studying app trends, developers checking out competitors, and anyone hunting for a free alternative to a paid app. The usual fix is called "embeddings" (turning text into number patterns that capture meaning). It's good at spotting apps that talk about similar things, but bad at spotting apps that *do* similar things.

ReFilter, from researchers Buthayna AlMulla, Maram Assi and Safwat Hassan, fixes that with a two-step approach. Step one is the cheap, fast pass: use embeddings to pull a shortlist of maybe-similar apps. Step two is the careful pass: hand that shortlist to a large language model (an AI that reads and reasons about text) and ask it to judge whether each candidate truly works the same way. Think of a librarian who first grabs 50 books on your topic, then reads the back covers to hand you the three that actually answer your question.

The team reports that ReFilter identifies functionally similar apps with about 90% accuracy — roughly nine correct matches out of ten. The work was accepted at ASIS&T 2026, a major information-science conference. The payoff sounds modest but adds up: better app recommendations, faster competitor research for small developers, and a clearer picture of how app stores are actually organised.

The catch: this is a research paper, not an app you can download today. Having an AI read through hundreds of app pairs is expensive and slow, which is exactly why ReFilter does the cheap filtering first. Until a store or a startup builds this in, you'll still be squinting at screenshots and star ratings like everyone else.