Hinge is more open about its algorithm than most dating apps, partly because its brand depends on the idea that the app is trying to get you off it. Between its help center, newsroom and a well-reported 2018 launch, you can piece together a fairly clear picture of how it decides who you see. All Hinge pages cited here were checked on 2026-11-06.
The short version
Hinge shows you people who fit your preferences and whose preferences you fit. Within that pool, a machine-learning system predicts who you’re both likely to be interested in, based on what you and people like you have liked before. It also learns from what happens after a match: whether you swap phone numbers, and whether you tell it a date went well.
What goes in
Hinge’s automated decision-making page lists the inputs to what it calls its proprietary matching algorithm:
- age, gender and location
- your preferences and filters
- who you like, skip and match with
- who you exchange phone numbers with
It adds that the same process is used to recommend you to others. That’s the two-way part.
Two more inputs come from features you can see:
We Met. A few days after you exchange numbers with a match, Hinge asks whether you went on a date and whether you’d like to see that person again. Its help page says a “no” is taken into account when recommending new people, that the feedback is private, and that it’s used to improve the algorithm. This is the closest any big app comes to learning from offline outcomes.
Your profile. Prompts and photos are what other people react to. Hinge’s tips page says its algorithm learns from your activity and that you should like profiles you think would be a good match, rather than liking as many as possible.
Where Gale-Shapley comes in
The Gale-Shapley algorithm, published by David Gale and Lloyd Shapley in 1962, solves the “stable marriage problem”: pair up two groups so that no two people would both rather be with each other than with their assigned partners. Each person ranks the other side. One side proposes, the other tentatively accepts its best offer so far, and the process repeats until nobody wants to switch.
When Hinge launched its daily Most Compatible pick in July 2018, TechCrunch reported that it used a Gale-Shapley approach, with preferences learned from users’ likes and passes instead of written rankings. CEO Justin McLeod’s description at the time was that the person you’re seeing is also seeing you. Early tests, as reported, found users eight times more likely to go on dates with Most Compatible picks than with other recommendations. That’s Hinge’s figure.
Hinge’s mission page still describes “a Nobel-Prize-winning algorithm” and says people are eight times more likely to have a great date with Most Compatible suggestions (checked 2026-11-06). A note on that label: the 2012 Nobel prize in economics went to Lloyd Shapley and Alvin Roth for work on stable matching and market design, which built on the 1962 algorithm. The algorithm itself isn’t a prize.
Today’s Most Compatible help page doesn’t name Gale-Shapley. It says Hinge tries to send one recommendation a day, based on mutual preferences, recent activity and shared patterns in who you and others like, and that it expires in 24 hours.
The 2025 rebuild
In November 2025, Hinge said it had rolled out global changes to its core recommendation system, using deep learning to better predict mutual compatibility, and credited it with a double-digit increase in matches. It also started nudging daters to widen some filters, such as distance, and explaining how dealbreakers and “Prefer Not to Say” answers affect who can see them.
That last point is easy to miss. Hard dealbreakers shrink not only who you see but who sees you.
Features that shape the feed
| Feature | How it affects who you see | Source |
|---|---|---|
| Discover | Mutual interest: you fit their preferences and they fit yours | Help center |
| Most Compatible | One daily pick at the top of Discover | Help center |
| Standouts | Daily feed of profiles getting a lot of attention; Roses only | Help center |
| Your Turn Limits | Too many unanswered matches blocks new Likes until you reply or end chats; applies to subscribers too | Help center |
| Boost / HingeX | Paid extra visibility, including HingeX’s “Get seen by more daters” | Help center |
| Signals badge | Subscribers can soft-sort by daters with consistent, thoughtful activity | Help center |
Your Turn Limits is a quiet algorithmic choice: Hinge stops serving you new people if you’re sitting on unanswered conversations.
Myths and what the sources say
“Hinge ranks you by attractiveness.” Not stated anywhere official. Hinge describes mutual preferences and patterns in likes. Standouts does surface profiles getting a lot of attention, which is a popularity signal, but that’s a separate feed.
“Liking everyone gets you more matches.” Hinge’s own advice points the other way: its algorithm learns from your likes, so it suggests sending quality likes over quantity.
“Most Compatible is just the most popular person nearby.” No. Hinge says it’s based on mutual preferences and shared patterns, and the 2018 reporting describes a mutual-match design.
“Hinge still uses Gale-Shapley.” Partly supported. Hinge’s mission page still credits a Nobel-linked algorithm, but its help pages don’t name one, and the 2025 deep-learning rebuild means the current system is more than a textbook stable-matching routine. How the two fit together today isn’t public.
“Paying is the only way to be seen.” Partly true for extra visibility, false for basic matching. Free users get Discover and Most Compatible; subscribers can buy reach.
Practical takeaways
- Use We Met honestly. It’s the only signal that tells Hinge what happened in real life.
- Keep dealbreakers to the few that truly matter, and mark others flexible.
- Reply or close conversations. Your Turn Limits will force it anyway.
- Like with intent, ideally with a comment on a specific prompt or photo.
For Hinge’s AI coaching tools, which sit on top of this algorithm, see Hinge’s AI features explained. For how Tinder and Bumble compare, see how dating app algorithms work.