How Tinder Actually Works: The Algorithm Behind the Swipe

The Swipe Is a Binary Classifier. That's It.

Let's get this out of the way: the swipe is not innovative. It is the simplest possible user input, a binary classifier operated by a human. You see a thing, you approve or reject it. Left or right. Yes or no. One bit of information.

Instagram does this with the double-tap. TikTok does it with scroll-past vs. watch-again. Bumble and Hinge copied the swipe directly. Every content-driven app reduces user feedback to the same primitive: present content, capture a binary (or near-binary) signal, feed it back into the ranking system. The interaction pattern is trivially simple, and there is zero engineering novelty in it.

The real engineering is behind the swipe. Who gets shown to whom, in what order, and why. That is where Tinder's actual system lives, and it is far more interesting (and more cynical) than most people realize.

The Original Elo System

Tinder's first ranking system was borrowed directly from chess. Specifically, from the Elo rating system designed by Arpad Elo in the 1960s for the United States Chess Federation.

Here is how Elo works in chess. Every player has a numerical rating. When two players compete, the expected outcome is calculated using the rating difference:

E_A = 1 / (1 + 10^((R_B - R_A) / 400))

Where E_A is the expected score for player A, and R_A and R_B are their respective ratings. After the game, ratings are updated:

R_A' = R_A + K * (S_A - E_A)

Where S_A is the actual score (1 for a win, 0 for a loss, 0.5 for a draw), and K is the K-factor, a constant that controls how much a single game can shift your rating. New players typically have a higher K-factor so their rating converges faster. The system is zero-sum: rating points gained by the winner are lost by the loser.

Tinder adapted this for dating. Every swipe was treated as a "match" between two players. When someone swiped right on you, that was a "win" for you. Swipe left, a "loss." Critically, the rating of the person swiping mattered. A right swipe from a user with a high Elo boosted your score significantly more than a right swipe from a low-Elo user. Conversely, getting left-swiped by a low-Elo user barely affected you, but getting rejected by a high-Elo user pulled your score down harder.

This created a single scalar ranking of "desirability." Users with similar Elo scores were shown to each other. If your Elo was 1800, you mostly saw other 1800s. If it was 1200, you saw other 1200s. The system was elegant in the way chess Elo is elegant: simple, mathematically grounded, and self-correcting.

It also had serious problems.

Why Elo Was Replaced

Tinder publicly announced they moved away from the Elo system around 2019, in a blog post that was part damage control and part PR spin. But the actual reasons were structural.

The feedback loop problem. Elo created a rigid caste system. Attractive users saw only other attractive users, reinforcing their high ratings. Average users got trapped in a dead zone: shown mostly to other average users, receiving average engagement, never able to break out. The rich got richer. The system converged to a stable hierarchy very quickly and then ossified.

The cold-start problem. New users had no data. Elo systems handle this with a default starting rating and a high K-factor, but in practice, new users would experience wild rating swings in their first few days. Tinder partially masked this by giving new profiles a "newbie boost" (showing them to more people initially), but this created its own distortion: users would get a burst of matches in their first week, then see a sharp dropoff as their Elo settled. This felt terrible and drove churn.

The single-axis problem. Reducing a person to one number is lossy even in chess, where there is a clear performance metric. In dating, "desirability" is not a scalar. Someone can be physically attractive but have a terrible profile. Someone can be average-looking but write messages that actually get responses. Elo captured none of this nuance.

The PR problem. Once journalists and researchers started writing about Tinder's "secret attractiveness score," the optics were bad. Nobody wants to be told they are a 1200 in the dating market. Tinder needed to distance itself from the concept, even if the replacement was functionally similar.

The Noob Boost and Decay Curve

That "newbie boost" mentioned above deserves its own section, because it is one of the most deliberately engineered psychological mechanisms in the app.

When you create a new Tinder account, your profile gets artificially elevated visibility for the first 24 to 72 hours. During this window, you are shown to a significantly larger pool of users than your eventual ranking would justify. Your profile also gets tagged with a "new user" badge, a small visual indicator that signals freshness to other swipers. That badge is not just cosmetic. It functions as a social proof signal: new profiles are novel, and novelty drives engagement. People are more likely to swipe right on someone they have never seen before, especially when the app itself is flagging that person as recently arrived.

This boost serves two purposes, and neither of them is altruistic.

The first purpose is data collection. The algorithm needs swipe data on you before it can rank you accurately. By showing your profile to hundreds or thousands of users in a short window, the system rapidly accumulates the right-swipe and left-swipe signals it needs to calibrate where you belong in the hierarchy. Think of it as a placement test. The system is not being generous. It is gathering information as efficiently as possible.

The second purpose is dopamine hooking. During the boost period, you experience a flood of matches. Your phone buzzes. Notifications stack up. The app feels like it is working. You form a strong positive association with Tinder in those first few days. This is classical conditioning: the app becomes linked to the dopamine hit of social validation.

Then the boost decays. Your profile settles into its "true" ranking, which is almost always dramatically lower than where the boost placed you. The match rate drops. The notifications slow to a trickle or stop entirely. The cliff is steep, and it is entirely by design.

Here is where the monetization kicks in. You experienced the high. You know what Tinder feels like when it is "working." Now it has stopped working, and you want that feeling back. Tinder Boost ($5-8) promises to recreate the visibility you had for free during your first 72 hours. Super Likes promise to cut through the noise the way your new-user badge once did. You are not paying for a new experience. You are paying to recreate an experience the app gave you for free, precisely so you would pay to get it back.

The entire new user funnel, from boost to decay to monetization prompt, is a single designed sequence. The generous onboarding is the first half of the trap.

The Current System: Multi-Signal Desirability

What replaced Elo is not a single score. It is a recommendation system, closer to what Netflix or Spotify uses than what chess federations use. Tinder has not published the full details (obviously), but based on patents, blog posts, and reverse-engineering efforts, the current system incorporates multiple signals:

Profile completeness. Filled-out bios, linked Instagram/Spotify, multiple photos. The system rewards profiles that give it more data to work with. An empty profile with one photo gets deprioritized.

Photo quality. There is strong evidence Tinder uses ML models to score photos. Not just "is this person attractive" but "is this photo well-lit, is the face clearly visible, is this a group photo." Smart Photos, a feature Tinder launched in 2016, already used A/B testing to determine which of your photos performed best. The underlying scoring system almost certainly feeds into ranking.

Message response rate. If you match with people but never message, or if you message but never get replies, that is a signal. High match-to-conversation conversion rates likely boost your ranking.

App usage frequency. Active users get prioritized over dormant ones. This is both practical (no point showing someone who hasn't opened the app in two weeks) and strategic (it rewards engagement).

Selectivity. Your right-swipe ratio matters, a lot. More on this below.

Recency. Recent activity is weighted more heavily than historical activity. This helps with the cold-start problem and means your ranking is not permanently locked by early data.

Geographic demand. Your ranking is partially local. Being in a dense market with high competition is different from being in a small town. The system adjusts.

The result is not a single number but a multi-dimensional embedding that the recommendation engine uses to decide who appears in your stack. "Desirability" still exists as a concept, but it is computed from many inputs rather than just swipe-in/swipe-out ratios.

The Photo Scoring Pipeline

The "photo quality" signal mentioned above is worth unpacking, because it is far more sophisticated than most users realize.

Start with Smart Photos, the feature Tinder launched in 2016. The concept is straightforward: if you upload multiple photos, the system rotates which photo is shown first to different users, then measures which lead photo generates the most right swipes. Over time, it converges on your highest-performing photo and leads with it. This is a simple A/B test, and Tinder was transparent about it. What they were less transparent about is everything happening underneath.

There is strong evidence that Tinder runs computer vision models on every photo you upload. The pipeline likely includes several layers of analysis. Face detection determines whether a human face is present and clearly visible. If your photo does not contain a detectable face (landscapes, memes, pets without their owner), it gets scored lower. Face count matters too: group photos where the system cannot easily identify which person is the profile owner get penalized. The algorithm does not know which one is you, and that ambiguity is a negative signal.

Image quality scoring evaluates resolution, lighting, and composition. A well-lit, high-resolution photo taken in natural light scores higher than a grainy, dimly lit bathroom selfie. Sunglasses and hat detection flags photos where the face is partially obscured. These are not hard bans, but they reduce the photo's contribution to your overall profile score. The system wants clear faces because clear faces generate more decisive swipe behavior (both left and right), and decisive swipes produce better training data.

There is also likely a selfie versus non-selfie classifier. Photos taken by another person, implying social context, tend to outperform selfies in engagement metrics. The system can detect this from camera angle, arm position, and framing.

Match Group has filed patents related to ML-based profile scoring and image analysis. While patents do not always reflect shipping products, they indicate the direction of investment. The patent filings describe systems for scoring profile attractiveness, photo quality, and even predicting match likelihood from visual features alone.

The existence of third-party services like Photofeeler is itself evidence of this dynamic. Photofeeler lets users get their photos rated by strangers before uploading them to dating apps. The entire business model is built on the premise that photos are being algorithmically scored and that small differences in photo quality have outsized effects on visibility. That premise is correct. Your first photo is not just making an impression on other humans. It is making an impression on a computer vision model that decides whether those humans ever see you at all.

The Message Quality Signal

Tinder tracks not just whether you message your matches, but how you message them. This behavioral data feeds directly back into your ranking.

Response time is a signal. Faster replies indicate higher engagement, and the system treats engaged users as higher-value participants in the ecosystem. This does not mean you need to reply instantly, but consistently ignoring matches for days before responding signals low investment.

Message length carries weight. Very short messages ("hey", "what's up", a single emoji) may be weighted lower than substantive openers. The system is not reading your messages for content (at least not for ranking purposes), but message length serves as a rough proxy for effort. A three-word opener and a three-sentence opener produce measurably different response rates, and the system can observe that difference at scale.

Conversation duration is a high-value signal. Matches that lead to five or more message exchanges represent successful outcomes from the algorithm's perspective. If your matches consistently produce multi-message conversations, the system has evidence that showing you to people generates real engagement, not just idle swipes. That evidence improves your ranking.

The most punishing signal is the unmatch-after-first-message rate. If a significant percentage of your matches unmatch you after receiving your opening message, that is a strong negative indicator. It tells the system that whatever you are sending is actively repelling people. The match itself was fine; the follow-through was not. Users with high unmatch-after-message rates likely get deprioritized, because the system learns that matching with you leads to negative experiences for the other party.

All of this data is aggregated and fed into the ranking model. You are not just being scored on your photos and your swipe patterns. You are being scored on your ability to hold a conversation, measured at every step from first message to sustained exchange.

How Men Get Penalized

This is where the system gets adversarial, and it is backed by data. A 2016 study from Queen Mary University of London and other researchers found that men swipe right on approximately 46% of profiles, while women swipe right on approximately 14%. Other studies put male right-swipe rates even higher, around 60%. The behavioral asymmetry is enormous.

Here is the problem: the algorithm interprets a high right-swipe rate as a signal of low selectivity. If you swipe right on everything, you are providing zero useful signal. The system cannot learn your preferences if your preference is "yes." More importantly, the system treats indiscriminate swiping as a negative quality signal. You are, in the algorithm's view, desperate.

The penalty is concrete. Users with high right-swipe ratios get shown to fewer people. Their profiles are deprioritized in other users' stacks. The logic, from Tinder's perspective, is rational: if you swipe right on everyone, a match with you means nothing, so showing you to others provides low-value matches.

This creates a vicious cycle. A man swipes right frequently because he gets few matches. Because he swipes right frequently, his ranking drops. Because his ranking drops, he gets even fewer matches. Because he gets even fewer matches, he swipes right even more desperately. The system converges on a state where a large population of male users have functionally zero visibility.

This is not a bug. It is the business model. A frustrated user is a monetizable user. Tinder Boost ($5-8 per use) temporarily puts you at the top of the stack. Super Likes ($1-3 each) signal intent that a normal right swipe cannot. Tinder Plus/Gold/Platinum ($10-30/month) unlock unlimited swipes, rewinds, and the ability to see who already liked you. Every one of these features "solves" a problem that the algorithm created.

How Women Get Penalized

The system is not kind to women either, just in different ways.

The primary penalty is engagement-based. Women receive a disproportionate number of right swipes. A 2014 analysis found that the average woman on Tinder matches with about 10% of the profiles she likes, while receiving likes from a much larger pool. The result is an inbox full of matches she never responds to.

Low response rate is a negative signal. If you match with 50 people and message 2 of them, the system sees 48 dead-end matches. Your "engagement quality" score drops. The algorithm learns that matches with you are unlikely to produce conversations, so it may deprioritize showing you to high-value profiles, the ones most likely to generate actual engagement.

Being extremely selective creates a different problem. If you swipe right on only 2-3% of profiles, the algorithm has very little positive signal to learn from. It cannot effectively determine your preferences, which means its recommendations get worse, which means you see even less relevant profiles, which reinforces the selective behavior.

There is also the overwhelm problem. When every session produces dozens of new matches, the marginal value of each match approaches zero. Women disengage not because the system fails to match them but because it matches them too successfully, with too many low-quality connections. The result is the same as the male experience, just from the opposite direction: low conversation rates, app fatigue, churn.

The Geography and Timing Game

Where and when you swipe matters more than most users realize. The algorithm does not operate in a vacuum. It operates within geographic and temporal constraints that fundamentally shape your experience.

Peak usage hours, typically Sunday evenings between 8 and 10 PM, concentrate the largest number of active users into the same window. More active users means a larger pool, but it also means more competition for visibility. Your profile is competing with every other active profile in your area for limited screen real estate. During peak hours, only the highest-ranked profiles get consistent exposure. Everyone else gets buried deeper in the stack.

Off-peak swiping (Tuesday mornings, weekday afternoons) presents the inverse tradeoff. The pool is smaller, so your options are more limited. But there are fewer profiles competing for visibility, which means your profile may surface more frequently to the users who are active. For users stuck in a mid-tier ranking, off-peak swiping can produce better results than fighting for scraps during the Sunday evening rush.

Population density is an enormous variable. In Manhattan, your potential match pool contains hundreds of thousands of active users within a few miles. The competition is brutal, but the pool is deep enough that the algorithm can be highly selective about who it shows you. In a small town with a few thousand residents, you might exhaust the entire available pool within days. The algorithm behaves differently in sparse markets: it widens age ranges, relaxes distance constraints, and resurfaces profiles you have already swiped on more aggressively, because there is simply nobody new to show you.

Tinder's Passport feature (available on paid tiers) lets you set your location to a different city. The use case is obvious for travelers, but the algorithm likely deprioritizes non-local profiles. A user sitting in Chicago who sets their location to Miami is a lower-quality match for Miami residents, because the probability of an actual meeting is low. The system knows this and adjusts accordingly. Passport users may get shown, but likely in a lower-priority position than genuinely local profiles.

Travel and tourism create interesting edge cases. Cities with high tourist volume (Las Vegas, Barcelona, Bangkok) have large transient populations. The algorithm has to handle users who appear in a location for a few days and then vanish. Transient users likely get different treatment than residents: the system may prioritize showing them to other transient users or to locals who have historically engaged with short-term visitors. The signal the algorithm extracts from a tourist's swipe behavior is fundamentally different from a resident's, because the tourist's preferences are shaped by vacation psychology rather than long-term compatibility.

None of this is visible to the user. You open the app, you see a stack of profiles, and you swipe. But the composition of that stack, and your position in other people's stacks, is being shaped by geography and timing in ways the interface never reveals.

The Monetization Trap

Tinder's parent company, Match Group, generated $3.2 billion in revenue in 2023. The vast majority comes from subscriptions and a la carte purchases. The business model requires a specific emotional state from its users: hopeful enough to keep swiping, frustrated enough to pay for advantages.

Consider the product design:

Boosts exist because the default algorithm buries most profiles. You are paying to undo the ranking system for 30 minutes.

Super Likes exist because a normal right swipe has been devalued by the system's own dynamics. You are paying for a signal that a right swipe should already provide.

"See Who Likes You" (Tinder Gold) exists because the app deliberately hides information it already has. Somewhere, someone has swiped right on you. The app knows this. It will not tell you unless you pay.

Tinder Platinum lets your messages appear before a match. This is literally paying to bypass the matching system entirely.

The algorithm and the monetization are not separate systems. They are the same system. The ranking engine creates scarcity. The paid features sell relief from that scarcity. If the algorithm showed you your best matches immediately and efficiently, there would be nothing to sell.

The Shadow Ban

There is a tier below low ranking, and it is worse than being explicitly banned.

Tinder shadow bans users who violate community guidelines or exhibit patterns the system flags as abusive. The key word is "shadow." You do not receive a notification. Your account is not suspended. You can still open the app, browse profiles, swipe, send messages, and purchase paid features. Everything looks normal from your side. The difference is that nobody sees you. Your profile has been deprioritized to near-zero visibility.

Tinder has never officially acknowledged shadow banning. The company's public position is that accounts are either active or banned, with no middle ground. But the evidence from users is overwhelming and consistent. The pattern is always the same: a sudden, complete drop to zero matches despite no change in profile content, location, or swiping behavior. Not a gradual decline. A cliff. Users report swiping for weeks without a single match, even with Boost active, even with a paid subscription.

The behaviors that trigger a shadow ban include mass-reporting by other users, spam-like messaging patterns (copy-pasting the same opener to every match), bot-like swiping behavior (right-swiping at machine speed), and violations flagged by automated content moderation. The threshold is opaque, and Tinder has no appeals process for something it does not admit exists.

The only known "fix" is account deletion followed by recreation. But Tinder has gotten increasingly aggressive about detecting this. The system now tracks device fingerprints (hardware identifiers, screen resolution, OS version), phone numbers (even if you use a new one, the old number is flagged), photo hashing (perceptual hashing algorithms that recognize reused photos even with minor crops or filters), and Facebook or Apple ID associations. If the system detects that a new account belongs to a previously shadow-banned user, the new account starts in the penalty pool rather than receiving the new user boost.

There is a perverse incentive at play. Shadow-banned users who create new accounts inflate Tinder's user registration metrics. From Match Group's perspective, these are "new users" in quarterly reports, even though they are the same frustrated person creating their third account. The company reports total registered users and active users to investors. Shadow-banned users who keep trying, who keep paying for subscriptions on accounts that will never produce matches, are revenue that costs Tinder nothing in algorithm resources.

The shadow ban is the logical endpoint of an opaque ranking system. When users cannot see their score, cannot understand why their experience changed, and have no mechanism for appeal, the platform has total power and zero accountability.

The Reset Meta

The shadow ban problem gave rise to a meta-game among power users: the account reset.

The strategy was simple. Delete your account entirely. Wait a few days or weeks. Recreate it with the same photos and bio. Because Tinder treated you as a new user, you triggered the noob boost again. Fresh visibility, a flood of matches, the full onboarding dopamine cycle. Some users reported doing this monthly, treating the account lifecycle as disposable.

For a while, it worked. Tinder's detection was limited to basic checks: same phone number, same Facebook login. Users learned to use new phone numbers (prepaid SIMs, Google Voice), create fresh social media accounts, and stagger their recreation by a few weeks. Reddit threads and YouTube guides documented the optimal reset procedure in granular detail.

Tinder caught on. The detection system now layers multiple identification methods. Phone number memory means your number is flagged even after account deletion, and using it on a new account links back to your history. Device fingerprinting goes deeper than cookies or app storage: hardware IDs, screen resolution, installed fonts, and other device characteristics create a composite fingerprint that persists across app reinstalls. Photo hashing uses perceptual hashing algorithms (like pHash or dHash) that generate a fingerprint of image content. These algorithms detect reused photos even if you crop them, adjust brightness, or apply filters. The hash is resilient to minor modifications. Facebook and Apple ID linking ties your Tinder history to platform accounts that are difficult to replace. IP tracking adds another layer, though this is the easiest to circumvent with a VPN.

Users who are detected resetting no longer get the new user boost. Instead, they are placed in a lower-priority pool. Some reports suggest detected resets result in immediate shadow ban status, effectively making the reset strategy worse than doing nothing.

The cat-and-mouse game continues. Users experiment with factory-resetting phones, using entirely new photo sets, purchasing new SIM cards, and waiting longer between deletions. Tinder continues to refine its fingerprinting. The arms race has no endpoint, because the incentive on both sides is structural: users want free visibility, and Tinder wants to prevent users from getting for free what it charges money for.

Other Apps Are Not Better

If you think the competition has solved this, think again.

Bumble uses the same swipe mechanic with one modification: women must message first within 24 hours, or the match expires. The timer creates artificial urgency, a classic engagement trick. The "women message first" rule was marketed as empowerment, but in practice it just shifts the cold-start problem. Instead of men sending low-effort openers, women send low-effort openers ("hey"), and the dynamic is largely the same. Bumble uses similar ranking signals (selectivity, activity, profile completeness) and has the same paid feature stack (Spotlight, SuperSwipe, Premium).

Hinge markets itself as "designed to be deleted," which is a bold claim for an app owned by Match Group, a company whose entire business depends on you not deleting dating apps. Hinge uses a different UI (prompts and comments instead of pure swipes), but the underlying system is the same: a recommendation engine that ranks profiles by predicted engagement. The "Most Compatible" feature is just the recommendation engine surfacing its highest-confidence predictions. Hinge's algorithm uses the Gale-Shapley algorithm (also called deferred acceptance, originally designed for stable matching problems) as one component, but it is still fundamentally an engagement-optimized ranking system.

Instagram and TikTok are not dating apps, but they use the exact same pattern. The feed is the algorithm. You scroll (or swipe), the system captures your dwell time, likes, comments, shares, and follows, then updates its model of your preferences and re-ranks content. The input mechanism (tap, swipe, scroll) is irrelevant. What matters is the ranking engine behind it. TikTok's recommendation system is just a much more sophisticated version of what Tinder does: predict engagement, rank by predicted engagement, optimize for time-on-platform.

Every app that shows you content in a ranked order is doing the same thing. The swipe is a UI pattern, not an innovation.

The Data Tinder Has On You

Before talking about better systems, it is worth understanding the scale of data the current system collects. The list is longer than most users expect.

Tinder collects your GPS location history, not just when you are actively swiping but whenever the app is running in the background. It logs your device information (model, OS version, carrier, screen resolution), your IP address, and every swipe you make, both left and right, with millisecond timestamps. Every message you send and receive is stored. Every match, every unmatch, every report.

Then there is the behavioral data that users rarely think about. Tinder tracks dwell time: how long you spend looking at each profile before swiping. It knows which photos you lingered on and which you scrolled past. It records your typing patterns in the message interface, including messages you started typing but deleted before sending. Your purchase history and spending patterns are logged in detail, including how you respond to promotional offers and price sensitivity tests.

If you linked your Instagram, Tinder has access to your public photos and engagement data. If you connected Spotify, it has your listening history. These integrations are marketed as "show more of your personality," but they also feed data into the recommendation engine and Match Group's broader data infrastructure.

Tinder's privacy policy is remarkably broad in scope. It grants the company the right to collect, process, and share user data for purposes that extend well beyond matching. When GDPR took effect in Europe and users began filing data access requests, some received hundreds of pages of personal data. The journalist Judith Duportail famously received 800 pages from Tinder in 2017, including every message she had ever sent, her login timestamps, her location at every login, and her internal ranking information.

This data is not just used for the matching algorithm. Match Group operates multiple dating platforms (Tinder, Hinge, OkCupid, Match.com, PlentyOfFish, and others), and data can flow between them. Beyond Match Group's own properties, the data enters the broader advertising and data broker ecosystem. A 2019 report by the Norwegian Consumer Council (Forbrukerradet) found that Tinder was sharing user data with at least 45 third-party companies, including advertising networks and data brokers. The data included age, gender, IP address, device IDs, and in some cases GPS coordinates precise enough to identify a user's home address.

Every swipe, every message, every second spent looking at a profile is not just an input to your ranking. It is a data point in a commercial surveillance system that extends far beyond the app itself.

What a Better System Could Look Like

If matching were optimized for outcomes rather than engagement, the system would look very different.

Compatibility scoring based on actual preferences. Not "you both swiped right," but analysis of what conversational styles, interests, and values actually predict successful connections. OkCupid tried this with their question system, and their data blog (before Match Group gutted it) published genuinely interesting findings about what predicts compatibility. The approach was abandoned in favor of swiping because swiping drives more engagement.

Communication style analysis. Some people write long, thoughtful messages. Some prefer quick banter. Matching people whose communication styles are compatible would reduce the "match but never talk" problem. This data exists in every messaging platform. None of them use it for matching.

De-emphasizing photos. Research consistently shows that photo-first interfaces produce worse outcomes than interest-first or text-first interfaces for long-term compatibility. But photo-first drives more swipes, more engagement, and more revenue.

Transparent scoring. If users understood how the algorithm ranked them and what behaviors helped or hurt, they could make informed decisions. But transparency is antithetical to the current business model. Mystery drives engagement. Frustration drives purchases.

The technology to build a genuinely better matching system exists today. The incentives to build it do not, at least not within a publicly traded company whose shareholders expect quarterly revenue growth from users who are, by design, kept slightly dissatisfied.

The swipe was never the innovation. The ranking engine was. And the ranking engine was never optimized for you. It was optimized for the quarterly earnings call.