A 95% match is not more compatible than a 70% match. Dating app percentages mostly measure how similarly two people answered a set of preference questions — not attraction, not conflict skill, not whether either person is actually ready for a relationship. The number is real math on fake inputs, and treating it as a verdict on two strangers gets the causality backwards.
What the percentage is actually computing
Match algorithms take the answers two users gave to a set of prompts — values, lifestyle preferences, sometimes personality-style questions — and score how closely they align. That's it. It's a similarity score on self-reported answers.
What it isn't is a prediction. No major dating app has published outcome data showing that higher match percentages correlate with longer relationships, because the platforms mostly aren't optimizing for that in the first place — engagement and continued swiping are the metrics that keep the business running.
It's worth sitting with that gap for a moment: a number presented with the precision of a percentage, generated from a process that has never been checked against the outcome it implies it predicts. That's not a criticism of the math. The math is fine. The claim riding on top of it is the part that doesn't hold up.
This isn't a conspiracy so much as a mismatch of incentives. A metric that reliably filtered people into lasting relationships would, definitionally, shrink the user base fastest for the people it worked best on.
None of that requires bad intent from any single company. It's a structural feature of a free, engagement-funded product category, and it's worth keeping in mind every time a match percentage is presented with more confidence than the underlying method can actually support.
Why similarity-based scoring misses attraction
The research on relationship similarity, discussed in more depth in the science of compatibility research, finds that similarity predicts liking reasonably well and predicts long-run satisfaction much less reliably — and predicts almost nothing about the initial spark that gets two people to message each other at all.
Two introverts who both prefer quiet nights in can match at a very high percentage and generate zero chemistry in person. An introvert and an extrovert who match lower on paper can create a dynamic neither would have predicted from their profiles — one pulling the other out, one grounding the other down.
Compatibility on paper and chemistry in a room are measuring two different things, and only one of them shows up in a percentage.
Attraction has behavioral and even physiological components — voice, timing, humor delivery, how someone listens — that a questionnaire simply has no access to. The algorithm isn't bad at measuring these. It was never asked to.
What the algorithm structurally cannot see
Even a well-designed matching system is working with a narrow slice of what actually determines whether two people build something together.
| What apps measure | What they can't measure |
|---|---|
| Stated preferences (music, politics, lifestyle) | Emotional maturity and how someone handles being told "no" |
| Demographic and lifestyle overlap | Conflict style — whether someone stonewalls, escalates, or repairs |
| Self-reported values | Whether stated values match actual behavior under pressure |
| Answers to a fixed set of prompts | Readiness for commitment at this specific point in someone's life |
| Text-based communication style | In-person chemistry, timing, and humor |
The right column is where relationships actually succeed or fail, and none of it is collectable from a profile and a questionnaire before two people have spent real time together.
From preference forms to behavioral scoring
Not every app computes its percentage the same way, and the shift matters for how much to trust the number.
Older matching systems, and the ones that still show an explicit percentage, mostly run on stated preferences: what you said you want, compared against what someone else said they want. This is the version discussed above — transparent, and transparently limited to self-report.
Many current apps have moved toward behavioral matching instead — ranking who you're shown based on who you've swiped on, who's swiped on you, and how similar users behaved. This drops the explicit percentage but doesn't fix the underlying problem, it just hides it. A system trained on your past swipes reinforces your existing patterns rather than correcting for the blind spots in them.
Behavioral systems also inherit a well-documented issue from recommendation engines generally: they optimize for what keeps you engaged and swiping, which is not the same objective as what gets you into a relationship and off the app. Neither generation of algorithm was built to solve for long-term outcomes, they've just gotten quieter about it.
What's actually known about relationships that start on apps
It's worth separating "the matching algorithm is a weak predictor" from "meeting on an app is a bad way to meet someone," because those are different claims and only the first one is well supported.
Meeting through an app has become one of the most common ways couples now meet, alongside or ahead of meeting through friends, work, or school. How a couple met has a much weaker relationship to long-term outcomes than what happens between them afterward — the predictors covered earlier in this piece apply regardless of whether the first message came through an app or a mutual friend.
What's genuinely unresolved, and worth being honest about, is whether the volume and speed of app-based dating changes behavior in ways that affect outcomes — more options can produce more selectivity, but it can also produce more comparison and slower commitment. The evidence on this is mixed rather than settled, and treating either direction as proven overstates what's actually known.
What actually predicts outcomes, and why it's invisible upfront
The predictors with the strongest research support — attachment security, conflict repair speed, mutual respect — are all things that only become observable once two people interact under some kind of pressure.
Attachment style shows up in how someone responds to a cancelled plan or an unanswered text, not in how they describe themselves in a bio. Conflict style shows up in an actual disagreement, not in a preference question about "how do you handle conflict," which everyone answers aspirationally rather than accurately.
This is the structural limit no algorithm redesign fixes: the things that matter most require behavior over time to observe, and a matching system operates entirely before any of that behavior has happened.
Why more matches can make the decision worse, not better
Beyond what the algorithm can and can't measure, there's a separate effect worth naming: a large pool of options changes how people evaluate any one of them, independent of the quality of the matching.
Decision research distinguishes between "satisficers," who commit once an option clears a reasonable bar, and "maximizers," who keep comparing against what else might be available. Dating apps, by design, push almost everyone toward maximizing — there's always another profile, another slightly higher match percentage, another possible upgrade one swipe away.
The effect isn't that choice is bad. It's that an interface optimized for continuous comparison works against the kind of sustained attention a new relationship actually needs in its early, fragile weeks. A high match score doesn't protect against this — if anything, it can make someone more likely to keep shopping, on the theory that an even better match must be close behind.
Recognizing this in yourself is a more useful check than recalculating anyone's match percentage: if you notice you're mentally comparing a person you're actually enjoying talking to against a hypothetical better option who hasn't appeared yet, that's the paradox of choice operating, not new information about the person in front of you.
A better way to filter, before you meet
None of this means match percentages are useless — it means they're a weak first filter that needs supplementing with things that are actually visible in a profile and in messaging.
- Read the profile for specificity, not polish. Vague, generic profiles reveal less than specific, slightly awkward ones. Detail is a better signal than a well-edited photo set.
- Watch how they respond to a normal disagreement in chat. Even a small friction point — a scheduling mismatch, a joke that lands wrong — is more informative than ten matched preferences.
- Notice whether they ask questions back. One-way messaging patterns in the first exchanges tend to continue, not improve, once you meet.
- Weight stated intent over match score. Someone actively looking for partnership at a 70% match is a better bet than someone ambivalent about commitment at a 98% match.
This isn't about ignoring the algorithm entirely — high alignment on genuinely load-bearing preferences, like wanting kids or not, is still useful information. It's about not letting a single number override everything the conversation itself is telling you.
When a high match number is worth trusting
The percentage becomes more meaningful when the underlying questions are actually load-bearing ones — timeline for children, monogamy versus non-monogamy, religious practice, willingness to relocate. Alignment there removes real disqualifiers before a first date.
It becomes far less meaningful when the underlying questions are about music taste, favorite foods, or personality-style prompts modeled loosely on frameworks like MBTI. Those questions produce a satisfying-looking number without touching anything that predicts whether the relationship survives contact with reality.
Knowing which category a given app's questions fall into is worth ten minutes of actually reading what's being asked, rather than trusting the summary percentage at face value.
What this means for how you actually swipe
Treat the match percentage as a rough pre-filter, not a ranking. A profile with lower stated alignment but specific, genuine detail is generally a better use of your attention than a high match with a generic, copy-pasted-feeling bio.
The things that end up mattering — attachment security, how someone handles friction, whether their stated values match their actual behavior — only reveal themselves once you're actually talking to and eventually meeting someone. No amount of algorithm refinement changes that; it's a property of what actually predicts relationship success, not a flaw in any specific app's engineering.
If you want a clearer read on your own patterns going into that process, the attachment style assessment is a more useful place to start than another round of profile questions, and the MBTI assessment can help you name what you're actually looking for in how someone communicates, rather than relying on a stranger's algorithm to have already worked that out for you.