What Tinder's algorithm actually scores
Tinder's ranking model is proprietary and the company has not published the full feature list. Any tool that claims to know exactly how the algorithm scores a profile is guessing. What follows is the publicly defensible part, and the part the audit can actually measure.
Tinder has stated, in product blog posts and engineering talks, that the system optimises for engagement quality. Profiles that pull a high right-swipe rate from the population segment Tinder is currently showing them to tend to be served to more eyeballs over time. That is a feedback loop: small early gains on photo quality compound into outsized exposure later, and small early losses compound into a stalled queue. Tinder has also confirmed that profile photos, especially the lead photo, are the single largest input the user controls.
Beyond that the model is opaque. We do not know the exact weights, the exact features, or the geographic and demographic adjustments the platform applies. What we do know is that fixing the inputs the model can see — eye visibility, smile authenticity, lighting, framing, slate composition — is the part the user controls. The audit measures those inputs against benchmarks computed from a corpus of dating-profile uploads, so the lead-photo decision the report returns is grounded in the same first-impression signals the apps surface to the model.
One more piece of the puzzle worth flagging: Tinder publicly states that profile activity (regular logins, recent likes, finished conversations) is part of the surfacing logic. A profile that goes inactive for a month is harder to recover from than a profile that is actively iterating. The practical implication is that the time to fix a slate is now, not after another quarter of slow signal. The audit is designed to compress what would be three months of slow A/B learning into a single 5-page deliverable, so the lineup the user runs next week is already the version the audit recommended rather than the version Tinder spent weeks teaching them was weak.
Common Tinder photo mistakes
Five structural problems account for most of the score gap between a strong Tinder slate and a weak one. The audit flags each automatically, but the patterns are common enough to spell out.
- ●
Lead photo is a group shot
The viewer has a fraction of a second to identify the subject. Even when you are the obvious focal point to you, you are not to a stranger seeing the photo for the first time. A cropped group shot with a residual hand or shoulder reads as off in the same way. Move the group photo to slot five or six, never the lead.
- ●
Sunglasses or hat hides the eyes
The eye region is the highest-weighted area in the scoring stack. Hiding the eyes on the only photo most viewers see is the single most expensive unforced error in the dataset. One hat or sunglasses photo later in the slate is fine; two or three becomes a pattern.
- ●
Heavily filtered or beauty-mode photos
Smoothing skin into a porcelain finish flattens the texture cues that the model uses to read three-dimensionality. Viewers consistently rate over-filtered photos as less trustworthy than slightly imperfect natural shots. Light colour correction is fine; full beauty mode is not.
- ●
Slate of near-duplicates
Six photos taken in the same outfit, in the same room, at the same angle, inside a 90-second window. Variety of context, lighting, and outfit is itself a signal of an active life. The audit calls out any cluster of three or more near-duplicate shots.
- ●
Distance shots where the face is small
Rule of thumb: on the lead photo, the head should fill at least one third of the frame. A scenic shot where the subject is a speck against a mountain backdrop belongs in slot four or later, not in the slot that decides the swipe.
Lead-photo composition, by the numbers
Three framing rules account for most of the score gap on a lead photo and you can check them with a ruler before uploading. The diagram below is the geometry the audit measures on every shot. A vote-based tool tells you the photo "feels" weak; the audit tells you the head fills 18 percent of the frame when it should fill at least 33, which is something you can actually fix. Want a Photofeeler-style breakdown on top of the geometry? The $49 audit includes an AI Voter Panel that scores every photo on Smart, Trustworthy, and Attractive.
Head-fill ratio
Head should fill at least one third of the frame
Measure crown to chin against frame height. Below 25 percent the viewer cannot read the face on a phone-sized first impression. Below 15 percent the photo belongs in slot four or later, never the lead.
Eye-line height
Eyes sit roughly one third from the top
Standard rule-of-thirds placement. Eyes higher than the upper fifth crops the forehead awkwardly; eyes below the centre line buys headroom you do not need and shrinks the face.
Headroom
A small gap above the crown, never a deep void
Roughly five to ten percent of frame height between the top of the head and the top of the frame. A larger gap reads as accidental; a zero-gap crop reads as cramped. The audit flags both ends.
RealSmile vs Tinder rating in-app
It is worth being precise about the difference between what RealSmile scores and what Tinder scores internally. Tinder rates a profile after it has been served to other users; the score moves based on real swipes, real likes, and real conversations. That is a strong signal, but it is also a slow signal. By the time a Tinder slate has accumulated enough impressions for the in-app ranking to settle, the user has already paid the cost of every weak photo in their lineup.
RealSmile rates a profile before it is uploaded. The audit runs the photos through a 17-axis scoring pass against benchmarks computed from a corpus of dating-profile uploads, then writes a 5-page PDF with the lead-photo decision, the delete list, and the 30-day improvement plan. The signal is faster and cheaper, but it is not the same thing as the in-app outcome. We measure first-impression inputs; Tinder measures realised engagement.
The two work in sequence rather than in competition. Use the audit to fix the inputs the model can see — eye visibility, smile authenticity, lighting, framing, slate composition — before paying the cost of running a weak slate through the live queue. Then let Tinder do its own thing. The full pricing ladder (free ranker, $20 quick rankings, $49 audit) is on the tools pricing page, and the per-axis breakdown for a given face is visible inside your face report.
How to A/B test your Tinder photos
Yes, A/B testing the lead photo is worth doing, but the way most users approach it is wasteful. The classic mistake is to swap the lead every two days and watch the like count, then declare a winner before enough impressions have stacked up to mean anything. Dating-app metrics carry heavy noise; you need a meaningful sample before any change is signal rather than randomness.
A workable workflow looks like this. Generate three lead candidates that score within five points of each other on the audit. Run candidate A for one full week without changing anything else on the profile. Note the like count and conversation rate. Swap to candidate B for a week. Then C. Compare the three weeks. If one candidate is clearly winning at the end, keep it as your lead. If they look the same, pick the one you like best and stop optimising.
Beyond the lead, photos in slots two through nine matter less individually but matter as a set. Test them as a lineup, not one at a time. The $49 audit returns a recommended lineup with each photo placed in the slot where its strengths do the most work, plus a delete list of shots that are dragging the slate down. That is more useful than running 30 individual swap tests over a quarter and learning nothing because the noise floor swallowed every result. If your goal is the same exercise across Hinge and Bumble, the full dating profile audit covers all three apps in a single deliverable.
Tinder swipe-stage pass-through · Where photos gate the swipe
Three swipe stages, three different reads.
Tinder is single-lead skim mechanics — unlike Hinge, viewers commit or reject before they see your second photo. But the lead has to clear three sequential attention stages, each with a different scan and a different failure mode. Most profiles pass stage 1 and break at stage 2.
Lead photo loads as a thumbnail in the swipe stack. Fast pre-screen.
What gets scanned
Face recognizability at thumbnail size. Whether the photo is a single person or a group. Whether the face is centered or cropped at the edge.
Common failure
Group lead, sunglasses lead, or face-too-small lead. Viewer cannot identify who you are at thumbnail and pre-rejects without expanding the card.
Intervention
Lead must be solo, face-fills-50%, eyes visible, in even daylight. Test at thumbnail size before committing — squint at your screen.
Card opens to full-frame. The lead photo is now read for trust + warmth signals.
What gets scanned
Microexpression read. Whether the smile crinkles the eyes. Whether the photo looks recent. Whether the outfit and background read as deliberate.
Common failure
Forced smile reads as cold. Bathroom mirror selfie reads as low-effort. 2019 photo where you look noticeably different reads as bait. This is the swipe-left moment for most profiles.
Intervention
Replace forced smiles with candid laughs. Replace bathroom mirrors with friend-shot portraits. Re-shoot annually so the photo matches who shows up to the date.
Viewer taps to expand the profile. Now photos 2-5 and the bio are read in sequence.
What gets scanned
Body proof in photo 2. Lifestyle signal in photos 3-4. Bio read for red flags. Final like/no-like decision happens here.
Common failure
Five near-identical face shots — viewer cannot place a body to the face. Or all group shots after the lead — viewer cannot tell which person is you. Or no bio.
Intervention
Photo 2: full-body. Photo 3-4: lifestyle in different settings. Photo 5: a strong second portrait at different lighting. Bio: one sentence specific enough to start a conversation.
The audit scores all three stages, not just lead-photo attractiveness. Most users discover their break-point is stage 2 — microexpression — which competitor scoring tools that rate facial geometry alone do not measure.
Swipe-stage funnel — where your set actually loses matches
Tinder is not one decision. It is four sequential filters that compress in roughly two seconds. A "low match rate" almost never fails at all four — it fails at one stage and the others never get the chance to compensate. The map below tells you which stage to fix first based on what you observe in the app, because the photo intervention for each stage is different.
Signal: Profile views are flat or low even with location/age set wide. People are not opening you.
Fix: Lead-photo problem. Shoulders-up framing, eyes visible, daylight, clean background. No group shots. No sunglasses. No hat-shadow on eyes.
Signal: Profile views climb but match rate stays under 5%. People open you and bounce immediately.
Fix: Microexpression problem. Lead photo reads as flat, anxious, or contemptuous. Fix: relaxed half-smile with eye crinkle, taken talking to a friend behind the camera, not posed.
Signal: Match rate decent on weekends and zero on weekdays — people swipe through your full set then bail.
Fix: Set-monotone problem. 6 photos that all show same angle, same outfit, same backdrop. Mix: 1 portrait, 2 lifestyle, 1 social proof, 1 hobby, 1 full-body.
Signal: Matches happen but conversations die before second message. Or "you look different in person" comments.
Fix: Photo-bio mismatch. Bio promises (job, height, vibe) do not match what the photos signal. Fix: align photo content to the persona you describe in the bio.
The failure mode this map prevents: rebuilding all 6 photos because match rate is low. Fix the failing stage first, run a 2-week test, then move on. Stacking 4 changes simultaneously means you cannot tell which one moved the needle.
Tinder photo questions, answered
How many photos should a Tinder profile actually have?+
Tinder allows up to nine photos. Most strong profiles use six to eight. The marginal value of the seventh and eighth photo is real but small; the marginal cost of a weak ninth photo dragging the average can be larger than the lift from one more decent shot. The audit returns a per-photo score so you can see exactly where the floor sits and whether your last two photos belong in the lineup.
Is a selfie or a portrait better for the lead photo on Tinder?+
Either can work, but a portrait taken by another person tends to score better on framing and depth-of-field metrics than a held-out-arm selfie. Selfies are not banned from a Tinder slate, but using one as the lead is a common avoidable mistake when a portrait alternative exists. The audit flags whichever is stronger on the specific metrics that drive first-tap rate.
Do filters and beauty mode hurt or help on Tinder?+
Heavy filters typically hurt. Smoothing skin into a porcelain finish flattens the texture cues viewers use to read three-dimensionality, and over-filtered shots are consistently rated as less trustworthy than slightly imperfect natural shots. Light color correction and exposure adjustments are fine. The audit calls out any photo where the filter is doing visible work against the score.
Should I keep hat or sunglasses photos in my Tinder lineup?+
A hat or sunglasses photo can sit at slot four or later, but it should never be the lead. The eye region is the highest-weighted area in the scoring stack, and hiding the eyes on the photo most viewers see first is the single most expensive unforced error in the dataset we benchmark against. One hat or sunglasses photo deeper in the slate is fine; two or three becomes a pattern that signals the user is hiding something.
Are group photos okay, and where should they go?+
One group photo is useful as a social-proof signal in slot five or six. Zero group photos is fine. Two or more is a problem because the viewer has to figure out which face is yours twice, and many will not bother. A group photo as the lead is the worst configuration in the dataset; the viewer cannot identify the subject before they swipe past. The audit flags every group shot and recommends placement.