Bottom Line: Beli replaces the tyranny of the 4.3-star average with a ranking engine built from your own comparisons, and it works better than it has any right to — provided you bring friends, because the app gates real value behind an invite treadmill that reviewers rightly resent.
The Ranking Loop
The comparison flow is the whole product, and it's tuned well. Logging a restaurant takes seconds: you pick a broad sentiment bucket, then the app serves a handful of A-or-B matchups drawn from your existing list. Binary search does the rest. The cognitive load is near zero — you're not asked to quantify anything, only to answer a question your gut already knows the answer to. Anyone who has stared at a five-star widget wondering whether a very good bowl of pho is a 4 or a 5 will recognize immediately what Beli has solved. You are much better at comparing two things than at scoring one thing.
The payoff compounds. Fifty entries in, the list stops being a log and starts being an argument — a ranked, defensible statement of what you actually like, assembled from decisions you made without agonizing over any of them. The derived 10-point scores feel earned in a way that self-reported ratings never do, because you never inflated one to be nice.
But the mechanic has a real structural flaw, and users have already named it: it lacks category awareness. The most common request in reviews is for comparisons scoped by context, so a $14 lunch counter isn't shoved into a matchup against a tasting menu. The complaint is correct. A single global ordering forces a question with no honest answer — is the best banh mi in the city "better" than a competent steakhouse? — and every time you're made to answer it, the list absorbs a little noise. The fix is well understood (bucket by occasion or price, rank within bucket), and its absence is the clearest thing standing between Beli and a genuinely rigorous preference model. Adjacent to this: users want a richer tag vocabulary — brunch, rooftop, the situational descriptors that actually drive a Tuesday-night decision. Both requests point the same direction, toward context.
The Social Layer, and Its Price
Here's where Beli gets genuinely interesting and genuinely frustrating, often in the same session.
The Match Score is the best idea in the app. It converts the vague social proof of "my friend liked it" into something with a confidence interval attached. A 92% match recommending a place is worth more than a hundred anonymous four-stars, and Beli is right to weight its suggestions accordingly. Seeing friends' reservations adds a live, ambient texture that no static review site achieves — you're watching your circle eat in something close to real time.
The problem is that all of this is conditional on network density. Beli without friends isn't a diminished version of Beli. It's a worse version of Google Maps with a longer setup process. And the company knows it, which is why the app leans on a friend-invite system to unlock features.
This is where the goodwill starts leaking. Android reviewers in particular describe what they call a "social paywall" — core functionality, reportedly including the ability to view restaurant ratings, gated behind repeatedly recruiting friends. Some report that promised premium unlocks simply never materialized after they did the recruiting. That last part is the damning bit. A growth mechanic that asks users to spend social capital is aggressive but defensible; a growth mechanic that takes the social capital and doesn't pay out is something else. It corrodes trust in the exact users who were enthusiastic enough to evangelize.
There's a legitimate argument that Beli's value genuinely requires a network, and that invites are the honest price of admission. Fine. But there's a wide gulf between "this is better with friends" and "you cannot see a rating until you've conscripted three people," and Beli currently sits closer to the second than it should.
Utility Beyond the Gimmick
Strip away the social layer and a durable tool remains. The filtering — by price, cuisine, neighborhood, custom tags — is the mundane feature that earns the most daily use, because it answers the actual question ("cheap Thai, walkable, open now") rather than the abstract one. The want-to-try list as a map is quietly excellent: every saved intention rendered geographically, so the list surfaces itself when you're standing near one of its entries. That's good product design — information appearing at the moment of relevance rather than waiting to be retrieved.
City leaderboards extend this to travel, and this is where Beli's model beats the incumbents most decisively. A crowd-averaged top-10 for Lisbon tells you where tourists go. A leaderboard weighted toward people whose match scores with you are high tells you where you'd go. That's a materially different product, and it's the strongest argument for Beli's existence.