The AI keyboard for dating apps.

Online dating is a mass behavior, so the missing product is an operating system, not another opener pack

Pew Research Center's 2023 reports found that 30% of U.S. adults have used a dating site or app, including 53% of adults ages 18–29. That is not a niche hobby. It is a default adult workflow with three different UIs fighting for the same evening. I built KOPY as an iOS AI reply keyboard because the workflow is the same even when the apps are not: copy context, draft a reply, send, measure whether the thread lived. The apps refuse to share a composer. A custom keyboard is the only composer that can follow you from Tinder to Hinge to Bumble without each company building you an AI.

This hub is not trying to rank for "AI keyboard for Tinder." The Tinder page already owns first-message conversion and chat decay. This page owns the cross-app system: which app to use for which job, which KPIs to carry between them, and how Dating / Social / Work / Insight modes map onto a week of messaging. If you only use one app, leave this page and read that app's guide. If you run two or three, you need a policy, not three different personalities.

Pew's platform split is the allocation model

Among 18–29-year-olds, Pew found 42% have used Tinder (and 14% of all U.S. adults), 27% have used Bumble, and 21% have used Hinge. Among online daters under 30, 79% have used Tinder. Read that as coverage, not as a quality ranking. Tinder is where volume and decay live. Bumble is where a 24-hour first-move rule lives. Hinge is where prompt comments live. Using all three with the same "hey" is how you work hard and convert nothing.

Under 30, 35% say apps make finding a long-term partner easier versus 33% harder. The operating system I want is the one that moves you into the 35% with process, not with more swiping. Process means: pick a primary app for the next 14 days, track three KPIs, and only add a second app if the first is converting and you still have unused evenings. Collecting matches on three apps with zero date-asks is how the 33% feel right.

Cross-app KPI table: one scoreboard, three rules

Pew measured adoption (30% of adults, 53% of 18–29). Pew did not measure your time-to-first-reply. I do not publish fake KOPY user totals. I do not invent App Store ratings. I do publish the scoreboard I used while dogfooding the keyboard across apps. Copy this table into Notes. Fill it weekly. The definitions stay identical so you can compare surfaces.

KPI	Definition that stays stable across apps	How the app changes the rule
Time-to-first-reply	Minutes or hours from the relevant start event to your first send	Tinder: start = match. Hinge: start = you deciding to comment, or match accepted. Bumble: start = match, measured against the 24-hour fuse
Continuation past 5 messages	Threads with five or more total bubbles involving both people	Tinder: this is the quality filter after cheap matches. Hinge: should be easier if the comment was specific. Bumble: should be easier if the opener asked a real question
Date-ask rate	Explicit plan asks (daypart + area + activity) divided by threads past 5	Hinge can ask earlier when prompts leak activities. Bumble can ask earlier because the product already punished delay. Tinder should not ask in message two unless they already offered
Primary-app focus hours	Minutes you spent in each app this week	If Tinder ate 80% of time and 0% of dates, you are feeding the volume machine. Reallocate to the app where continuation actually happens

A weekly review question: which app produced the last real date, not the last dopamine match? Put 70% of keyboard generations there next week.

When to use which app (the policy I actually follow)

Pew's usage ranking is Tinder, then Bumble, then Hinge in the 18–29 band. My allocation is the reverse of panic and the same as conversion. I use Hinge when I have the energy to write five comments. I use Bumble when I can clear the timer the same evening. I use Tinder when I have a short batch window and I am willing to accept decay. I do not open all three at midnight. Midnight is how Work-mode sentences leak into Dating mode and how you send a paragraph you will hate in the morning.

• Choose Hinge as primary if your photos are fine and you can answer prompts. The unit of work is the comment. Read the Hinge page for prompt KPIs.
• Choose Bumble as primary if you can stand a countdown and you will actually open or reply the same day. The unit of work is the first move. Read the Bumble page for expire rate.
• Choose Tinder as primary if you need volume to learn opener skill, or if your city is thin on the other two. The unit of work is the first bubble and the war on decay. Read the Tinder page for opener conversion.
• Run two apps only after one app has a non-zero date-ask rate. A second app is extra surface area, not a personality.
• Drop an app if expire rate or zero-message matches dominate and you refuse to change time-to-first-reply. The keyboard cannot save a surface you will not attend.

Geography matters more than Pew's national percentages. If your city is Hinge-heavy, Pew's 21% still includes you, but local equilibrium can invert the ranking. Believe your match quality, not a national league table.

Why the OS is a keyboard: Apple's constraints as product features

Apple's custom keyboard documentation is the architecture document. Extensions get a small RAM envelope, about 70MB, which forbids shipping a giant on-device model in the keyboard process. They get Full Access as the gate for network and a useful pasteboard. They get ephemeral URLSession instead of background downloads. I treated those constraints as the product. Dating apps are foreground, bursty, and hostile to extra chrome. A two-second generate that inserts into the field they already have is the only UX that survives Tinder, Hinge, and Bumble without three separate partnerships.

The clipboard is the cross-app API. Tinder will not give me the last message. Hinge will not give me the prompt. Bumble will not give me the timer. You copy. We read. We POST mode + tone + context. We insert. You edit. That loop is identical on purpose. What changes is the clipboard payload: bio fragment on Tinder, prompt plus answer on Hinge, badge plus photo cue on Bumble. People who copy "hey" on all three apps then blame the model. The model answered "hey."

Auth and premium flags live in an App Group so the keyboard can call generate without a login screen inside Tinder. Dating mode is premium because it is the high-stakes generator. Social, Work, and Insight still matter on a dating week. Social is the iMessage-after-the-app tone. Work is the "I'm on call this weekend" boundary. Insight is the rare sincere question. If you leave Dating mode on for a logistics text, you will rizz your own arrival time. I have done this. It is not cute.

Mode map for a dating week, not a single opener

Pew's 53% of 18–29-year-olds who have used a dating site or app includes people who also have jobs, group chats, and family threads. KOPY is a general keyboard. Dating is one mode. Treating every message as Dating mode is how you become exhausting. The map I use:

1. Inside the dating app, first five bubbles: Dating mode. Tones: Rizz, Funny, Disagree, Agree, Spicy with a high bar.
2. Inside the dating app, logistics: Social mode or Neutral-leaning Dating. Confirm time and place. Stop performing.
3. After you move to iMessage: Social mode. The performance is over. Be a person who can pick a restaurant.
4. If they ask a real question about your work or a move: Insight or Work, then back. Do not stay in Work and start sounding like a performance review.
5. If you are done: a short, kind close. Do not generate a speech. Do not generate cruelty. The keyboard will draft either if you ask. Do not ask.

Continuation past five messages is the shared quality metric because it is the first moment a thread is a conversation rather than a performance. Date-ask rate is the shared outcome metric because matches are not dates. Time-to-first-reply is the shared process metric because Pew's 33% "harder" camp is largely a delay-and-generic-language camp in my experience, even though Pew did not measure keyboard latency.

A two-week operating cadence

Pew's 35/33 easier-versus-harder split under 30 is a two-week experiment, not a personality test. Week one: pick one primary app using the policy above. Enable Full Access so the pasteboard and network work, which Apple requires for a custom keyboard that calls a network model. Generate only from real clipboard context. Log the three KPIs every night in one line: time, continuation, asks. Do not install a new dating app this week. Do not rewrite your entire profile this week unless photos are actually broken.

Week two: keep the primary app. Add a second app only if date-ask rate is above zero and you have leftover evening time. Put Bumble first in the session if any matches are near expiry. Put Hinge second if you have five prompts in you. Put Tinder last as a batch, not as an infinite scroll. Reallocate next week toward whichever app produced continuation, not toward whichever app produced match badges.

The keyboard stays constant. Ephemeral session, sub-two-second draft, modes you switch on purpose. The apps stay different. Tinder decays softly. Bumble expires hard. Hinge judges the comment. Your job is to stop using one personality for three mechanics.

Internal map, so the pages do not cannibalize each other

Pew's platform split (Tinder 42%, Bumble 27%, Hinge 21% among 18–29) is why these URLs are separate jobs, not duplicate "AI keyboard for Tinder" pages. Use KOPY for Tinder when you need opener conversion and anti-decay tactics. Use KOPY for Hinge when the unit is a prompt comment and an early thoughtful thread. Use KOPY for Bumble when the unit is a women-move-first send against a 24-hour clock. Use this hub when you need the allocation policy, the shared KPI definitions, and an AI keyboard for dating apps as a category — one composer, three surfaces, Pew's usage map as the demand backdrop, Apple's keyboard sandbox as the reason the composer can exist at all.

I shipped KOPY because I was tired of dating apps owning the match graph while leaving the sentence to a tired thumb. The graph is still theirs. The sentence can be yours, drafted fast, edited honestly, measured weekly. That is the operating system.