Dating apps are deleting the swipe and letting AI pick
Bumble, Tinder and Ditto are replacing swiping with AI matchmaking. What each app announced, and what its privacy policy says it collects.
Bumble’s chief executive said on 7 May 2026 that the swipe was going away, and repeated it to analysts on 6 August. Tinder has been testing an alternative since at least February 2026. Ditto, which started at Berkeley and texts students a date every Wednesday, never had a swipe to remove. What replaces the gesture is a model that narrows the field for you, raising a question about what the app needs to know before it chooses for you.
That second question is answerable without speculation. Each company publishes a privacy policy, and two publish help pages describing what gets scanned, what leaves your phone and how long it is kept. We read those pages on 12 September 2026 and quote them as they stood. This piece publishes in October, so check the dates on anything you act on.
What each company has actually announced
Bumble founder and chief executive Whitney Wolfe Herd confirmed the end of swiping in an Axios interview on 7 May 2026, quoted by TechCrunch that day: “We are going to be saying goodbye to the swipe and hello to something that I believe is revolutionary for the category.” On the second-quarter call on 6 August she described the replacement without naming it, saying the core idea was “a shift away from optimizing for swipe speed and velocity towards something more intentional, fewer, better, more considered signals”. She added that Bumble would use more AI tools under the hood while cautioning it was not becoming an “AI-driven experience”. In the same reports, paying users fell about 21% to 3.2 million in the first quarter, and second-quarter revenue was down 15.2% year on year to $210.5 million.
Five days later, on 11 August 2026, Bumble published a release ending the rule the brand was built on. Anyone in a match can now send the first message, and recipients have 72 hours to reply instead of 24. The release also sets a constraint: only one initial message is allowed, the chat opens only when the match replies, and a one-word opener triggers a prompt to write something better. Bumble says 66% of women surveyed on Bumble prefer men to send the first message, a figure its own footnote sources to 2,267 US women surveyed in August 2024. Wolfe Herd's line is that “being women-first was never about prescribing just one way to connect”.
Tinder's version arrived earlier and under a different name. On Match Group's fourth-quarter call, reported by TechCrunch on 4 February 2026, chief executive Spencer Rascoff said the Chemistry feature was designed to combat swipe fatigue and let people “get just a single drop or two, rather than swiping through many, many profiles”, while it was still in testing in Australia. The same report gives the pressure behind it: registrations down 5% year on year and monthly active users down 9%. Tinder's help centre documents what has rolled out in select markets: a diamond icon on the Discovery screen, a one-time set of questions, an optional camera roll permission, a Daily Drop you reveal by tapping a button, and an Insights Hub where you can read and delete the insights generated about you.
Ditto has no legacy to unwind. Co-founders Allen Wang and Eric Liu dropped out of UC Berkeley to build it, and students sign up by texting an iMessage number rather than installing an app, as TechCrunch described on 6 August 2026. Its privacy policy, effective 28 July 2026, describes the company plainly: Popcorn AI Tech Inc., trading as Ditto, “provides an automated dating service powered by AI”. The homepage lists a four-step flow: preferences by Tuesday night, one match by text at 7pm Wednesday, a scheduler, then the date.
How the matching is supposed to work, in the founders' framing
Wang's explanation to TechCrunch is the clearest public statement of the premise the category rests on. Ditto reads what hobbies imply rather than pairing people who share them. His example: a man into rock climbing, skydiving and outdoor sports and a woman into hip hop, streetwear and skateboarding look nothing alike on paper. “I can tell you deeper down, both are very adventurous; maybe they're very individualistic”. From that he draws the claim the product is sold on, that “chemistry is actually predictable with the right signals”. What follows the match is deliberately blunt. The weekly message arrives with a person, a time and a location, and, in Wang's words, “the users simply just need to show up and date each other and come back”, after which Ditto collects feedback.
Signals, in practice, means whatever the app can read, and Tinder's Photo Insights page is unusually specific. If you opt in, a scan of your camera roll runs on the device to judge photo quality and content, then a selected set of photos is “temporarily uploaded to our server” and shared with an AI technology provider to generate insights and suggestions, after which, the page says, those photos are deleted unless you add one to your profile. Bumble's use of AI is narrower, and its help page for Bumble For Friends icebreakers, last updated a year before we read it, draws the boundary: the feature sends interest badges, age, gender, name and city to OpenAI Ireland Limited, no special category data, and that provider “cannot use any data we share for training its AI models” and deletes what it receives within 30 days.
Stage | The swipe model | AI matchmaking as the apps describe it | What that version asks you to hand over |
|---|---|---|---|
Finding candidates | An open stack of profiles you rate one at a time | A short list the app chooses: Tinder's Daily Drop, Ditto's single Wednesday text | Answers to a one-time questionnaire, plus your profile data and in-app activity |
Deciding | You judge a photo in a second | The app narrows first, what Wolfe Herd called “fewer, better, more considered signals” | Whatever the model reads as a signal, including what your interests imply about you |
Photos | You choose and upload each one yourself | Photo Insights scans the camera roll on device, then uploads a selected set | Optional camera roll access, and optional on-device biometric data to find images of you |
The first message | Someone writes it, and on Bumble that was the woman until 11 August 2026 | AI-drafted icebreakers, and on Bumble either side may open | Profile badges, age, gender, name and city, sent to OpenAI Ireland for icebreakers |
Meeting up | You arrange it yourselves in the chat | Ditto supplies the person, time and place, then asks how it went | Phone number, university email, your availability, and feedback afterwards |
Sensitive detail | Optional fields you can leave blank | Fuel for the prediction, since the premise is that taste predicts chemistry | Ditto lists racial origin, religious and political beliefs and sexual orientation as sensitive data collected with consent; Tinder says profile data may include sexual orientation, health or political beliefs |
The numbers, and why the industry is moving now
Ditto is the only one of the three publishing growth rather than decline. TechCrunch reported 150,000 signups on 6 August 2026 and, from Wang, that about 20% of matches made through Ditto end in a date. It has $9.2 million in seed funding from investors including Gradient and Peak XV. Its homepage counter has moved since, reading more than 185,000 students across more than 100 schools on 12 September 2026, with a graphic advertising more than 80,000 dates arranged. Treat the 20% as what it is, a founder's figure for his own product; TechCrunch's aside was that the rate sounds low only to people who have never used Tinder.
On the incumbent side the direction of travel explains the urgency. Bumble's paying users and revenue both fell through the first half of 2026 on its own reported figures. Match Group's first-quarter revenue of $864 million, up 4%, came with lower guidance for the second quarter. In that context AI is both the product pitch and a cost line. On the first-quarter call, reported by TechCrunch on 6 May 2026, Match Group chief financial officer Steven Bailey said the company was giving “every employee in the company access to all the cutting-edge tools” and wanted to “become an AI-native company”, then added: “these tools cost a lot of money, as I'm sure you know, and so the way we're helping to pay for that is by slowing our hiring plans for the rest of the year”. The quarter in which a dating app decides the machine should pick your matches is the quarter it decides the machine is worth a hiring freeze.
What these apps learn about you, and what happens to that profile
Ditto's policy is the broadest of the three in what it says it takes. Alongside contact details, university affiliation, date of birth and photographs, it lists sensitive data including racial and ethnic origin, religious and philosophical beliefs, political views and sexual orientation, collected with consent or as law otherwise permits. It covers user-generated content and “associated metadata”, defined to include how, when, where and by whom a piece of content was collected, and says session replay software from PostHog records clicks, touches and scrolls so the team can watch replays of sessions. On sharing it names generative AI providers directly: information from your interactions may go to “AI Service Providers” such as Anthropic, OpenAI and Google. Its retention section sets no number, describing instead the factors it weighs before deleting, anonymising or isolating data. What it does give you is a section in your account called “What Ditto knows about you”.
Bumble, whose joint GDPR controllers are Badoo Trading Limited and Bumble Trading LLC, publishes retention periods line by line. Almost all profile data is deleted 28 days after you delete your account, and an account can be restored inside that window. Core registration fields such as email, phone number, name and geocoordinates are kept six years from deletion. Biometric scans from photo verification are kept no longer than three years after your last interaction and a government ID for 90 days. Moderation results from message scanning are held 90 days after deletion, investigation outcomes for a year, six years where a member was blocked, and up to 15 years for certain blocked members depending on the seriousness of the harm. Prompts and inputs from the AI icebreaker feature are kept six years after deletion. One line in the marketing section is worth reading twice: the behavioural factors used to build lookalike advertising audiences include photo uploads, purchases, verification, prompt responses and swipe speed.
Tinder's policy, which applies from 4 April 2026, is where the machine learning language is explicit. Under content it lists photos, videos, audio, text, your search queries and your chats with other members, and under profile data it warns that what you share may be sensitive in some countries, naming sexual orientation, sexual life, health and political beliefs. One stated purpose is “developing and improving new features and services, including through machine learning and other technologies”, and a later clause says that where legally permitted it may keep and use data that cannot by itself be attributed to you to improve the service. Retention is layered: a three-month safety window after you close an account and a year after a ban, transaction data for ten years, profile data for a year against potential legal claims, face data deleted within 30 days, and automatic closure after two years of inactivity. Photo Insights you do not keep are deleted within 90 days, and in some markets, the help page says, insights only suggest photos, bios and interests and are then promptly deleted, with no Insights Hub shown at all.
Before you grant the camera roll permission
The practical reading of all this is narrow and checkable. The swipe made you the filter and asked for photos and a bio. A model that predicts chemistry needs material, and what these companies describe is your camera roll, your messages, your inferred interests and, if you offer it, the category of information European law treats as special. Each transfer is disclosed on a page you can read in a few minutes, and two of the three give you a screen for inspecting or deleting what has been inferred: Tinder's Insights Hub and Ditto's “What Ditto knows about you”. Open it in your first week rather than your first year, and decide the camera roll question deliberately, because on Tinder's own account that permission does the most work.
None of this is legal advice. We are a Belgian company reading published policies on the day they were served, and any of them can change the week after we quote it. We are a chat app with no dating features of any kind, so the only policy we can usefully put beside those three is our own: we do not train on your conversations, what the provider behind a given model does with them varies by model, and the models where we know of a difference carry a notice on their info card in the picker. Ours is at multichats.ai/privacy, worth holding up against the three above.