How AroundMe's matching actually works

Ask someone what they think happens inside a matching system and you'll hear one of two fears. Either it's dark magic (an all-seeing algorithm that has quietly decided things about you) or it's a dice roll with good marketing. Both fears are earned; plenty of products are one or the other. AroundMe is neither, and the only convincing way to show that is to open the box.

This is the technical companion to What happens after you join the waitlist. That post walks the whole path from signup to meetup; this one stops at the single step that decides the most, how you and four or five strangers end up in the same group, and explains exactly who decides what.

How a group gets made

1. 1
   The pool formsEveryone who opted in this week for a hotspot's slot. Nobody else is considered.
2. 2
   Hard rules draw the linesSize 4-6, your windows, no-repeat, blocks, gender comfort, age bands, all enforced in code, not by AI.
3. 3
   An anchor interest is foundEvery group gets one shared interest everyone can open with.
4. 4
   The AI composes the groupComplementary intents and energies, then it writes your why-this-group note and icebreakers.
5. 5
   One invite goes outFirst names, a concrete plan, and the why. Confirming is always yours.

Does AroundMe use AI to match people?

Yes. A large language model composes the groups, and we'd rather say that plainly than hide it behind "our proprietary algorithm." Given a week's opted-in pool for a hotspot (a theme, a verified public meeting spot, and a recurring time slot), the model partitions it into groups of 4-6, finds each group one shared anchor interest, balances the mix of intents, writes the short why-this-group note that appears on your invite, and drafts three icebreakers tied to what the group actually has in common.

Two details most products would leave out. First, the model works inside hard rules it cannot change. They're enforced in code around it, and its output is checked: every opted-in person must be accounted for exactly once, either placed in a group or explicitly rolled to next week, and any output that breaks the rules is discarded, never shipped. Second, there's a fallback. When the AI is unavailable, or its output fails those checks, a simple deterministic method composes groups by interest overlap instead. The why-this-group notes get less lyrical; the rules and guarantees stay identical.

What rules can the AI never break?

The load-bearing ones. Who can even appear in the same pool is decided by ordinary, testable code, not by a model: group size, availability, no-repeat, blocks, the gender comfort rules, age bands, and one-invite-at-a-time are all enforced this way. The AI arranges people within those lines; it cannot redraw them.

Availability alone runs three layers deep: the recurring windows you gave at signup decide what you're eligible for, your weekly opt-in decides whether this week counts, and the invite itself is the final yes or no. No calendar syncing, no inference about your schedule: hotspots own the time, and you're either matched into a slot you said works or you're not matched at all.

What matching will never do

• Put you in a group smaller than 4 or larger than 6
• Match you outside your windows, or on a week you didn't opt in
• Seat you with someone you've blocked
• Rebuild last time's group and call it new, since no-repeat keeps faces fresh
• Target anyone by gender (the comfort question is the only gender setting that exists)
• Leave exactly one woman in a mixed group
• Show your age band to your groupmates
• Hand you a second invite while one is active

How does AroundMe handle gender?

With one question, asked of you, about you. Onboarding asks your gender (with non-binary and prefer-not-to-say as real options) and then a single comfort question: would you feel more comfortable in a same-gender group? That is the entire gender feature.

Notice what it isn't: targeting. There is no way to request the opposite gender, no ratio slider, no ordering up a group by who's in it. That's a hard rule, and it's the same principle that runs through the whole product: express your own intent, never order up other people. One balancing default sits on top: a mixed group never contains exactly one woman. Nobody should look around a group of new people and realize they're the only one.

Does AroundMe match by age?

By band, never birthdate. You pick an age band (18-24, 25-34, 35-44, 45-54, 55+) and social groups span at most two adjacent bands, so a 23-year-old isn't dropped into a group where everyone else could be their parent. Professional hotspots run looser on purpose; a build night is better with a 24-year-old and a 51-year-old at the same meeting spot. The platform is for adults (18 and over), and your band is never shown to your groupmates. When a hotspot is age-framed (a "20s & 30s brunch"), that's self-selection through programming, not a people filter: you choose the hotspot; you never filter humans.

How does the AI actually compose a group?

Around one shared anchor. Inside the rails, the model's first job is finding each group a single anchor interest every member genuinely shares, one guaranteed conversation anyone can open. Not five people who match on everything (that group is an echo), but one thing everyone can start with.

Then it balances what shouldn't be identical: complementary intents (people looking for advice seated with people glad to give it, quieter energy alongside more outgoing) because a group of five identical intentions has nothing to trade. Two more thumbs rest on the scale: freshness, so new faces outrank reruns, and first-timer priority, because a first-timer's first eligible cycle must never end unmatched. And when a good group can't be formed, the system doesn't force a weak one: it rolls people to the next cycle with priority instead. A skipped week stings less than a bad one.

Who decides what?

The pattern above is the product's whole philosophy: the code holds the guarantees, the AI does the composing, and every yes belongs to you.

Can I pick who's in my group?

No, nobody can. There's no browsing, no filters, no requesting people, for you or about you. What you control is everything on your side of the line: what you want to do, which windows work, whether you opt in this week, whether you confirm the invite, who you block, and (after the meetup) who you'd want to see again. That last one is the only person-picking on AroundMe, and it works only when the other person picks you back.

What does the AI never see or decide?

It never sees a photo, a surname, or contact details. Its entire view of you is a first name, your interests, and your intent, with the comfort and age rules applied by the code around it. And it never scores you. There is no desirability ranking, no A-list, no hidden number deciding whether you deserve a good group. The model decides who fits together this week, never what anyone is worth. It also decides nothing once the invite exists: it can't confirm for you, can't unlock a profile, can't override a block, and can't hand you a second meetup while one is active.

What will I see about my group, and when?

Deliberately little at first, more as trust becomes mutual. Your invite shows first names and the why-this-group note (no photos, no surnames, no links), enough to judge the plan without judging the people. After you confirm, the icebreakers written for your specific group land. Full profiles (photo, LinkedIn, contact) unlock only after the meetup, and only between two people who both said they'd like to see each other again. Meanwhile the platform has verified everyone from the start (real, accountable accounts, never anonymous burners) and people stay private to each other until they choose otherwise. Verified but private.

Does matching get smarter over time?

Yes, through feedback, not surveillance. The intake gives matching its floor; meetups give it its ceiling. Every cycle adds ground truth: what you confirmed and declined, how the meetup felt, and the question that quietly powers everything: who would you see again? A mutual yes creates a connection; every answer, mutual or not, teaches the next composition something a form could never ask. No-repeat compounds this: because faces stay fresh, feedback keeps arriving about people you haven't met yet, not reruns of the same group.

What if my group falls flat?

Sometimes it will, and we won't pretend otherwise. Matching stacks the odds (a real anchor, a balanced mix, everyone verified, everyone confirmed), but chemistry isn't a deliverable, and any product that promises it is lying to you.

What we can promise is what happens next. Your feedback lands directly on the next composition. No-repeat means a flat group never becomes a recurring one. When someone declines or an invite expires, backfill refills the seat fast rather than letting the group wobble. And if a good group can't be formed at all, you roll to the next cycle with priority instead of being handed a bad evening. We hold this stance because we've paid for it before: our 2024 campaign ran 105 real meetups, and we retired that entire product because of what those nights taught us. Candor about limits is cheaper than letting you discover them.

The short version

Code enforces the guarantees: 4-6 people, your windows, no repeats, blocks honored, gender comfort respected, age bands instead of birthdates, one invite at a time. Inside those rails, an AI composes each group around a shared anchor interest and writes the note telling you why. It never sees your photo and never scores your worth. You see first names first, more only when it's mutual, and every yes, from opt-in to see-again, stays yours. Neither dark magic nor a dice roll. And when the match lands, none of that machinery is what you'll remember. What stays with you is the evening it makes: the others already at the meeting spot, the anchor interest giving you an easy first thing to say, an hour in when the icebreakers are long forgotten because the conversation found its own legs, and someone on the way out asking whether the group's on for next week.