Ask a room what they want from a night together and nobody says "small talk". They want the conversation that happens at 1am when the music is off and someone says something true. Everyone wants that. Almost nobody starts it.
People want deeper conversation than they think they do
This is not a vibe, it is one of the better-replicated findings in social psychology. Across twelve experiments with more than 1,800 participants, researchers at Chicago Booth had people predict how a deep conversation with a stranger would feel, then had them actually have one. Participants consistently expected deep conversations to be awkward, and consistently rated them as more enjoyable and more connecting than the small-talk control. The gap between predicted and actual awkwardness was one of the largest effects in the literature.
The reason given is specific and worth sitting with: people underestimate how much others care about what they have to say. Everyone in the room is quietly assuming they are the only one who would find this interesting. So the room defaults to weather and work, which nobody wanted, because each person is protecting the others from a conversation all of them would have preferred.
What this means for a deck: the job is not to persuade people to open up. They already want to. The job is to remove the moment where someone has to decide to go first.
What actually builds connection: escalation and reciprocity
The best-known experiment here is Arthur Aron's, from 1997. Pairs of strangers spent about 45 minutes working through 36 questions that escalate in intimacy, in three sets of twelve. A control group made small talk for the same length of time. The question pairs reported closeness comparable to close relationships they already had. One pair married six months later.
The mechanism has two parts, and neither is optional.
Escalation. The questions start light and get gradually more personal. Set one is easy. Set three is not. Nobody is asked for anything real until the cost of answering has already come down.
Reciprocity. Both people answer everything, in turn, at matched depth. One person opens, the other meets them there. Disclosure that only travels one way is not connection, it is an interview.
That is the engine. Everything else in a question game is either serving it or getting in its way.
What we do about it
Four things, each of which is just the research made mechanical.
Decks are staged, not shuffled flat. New Friends opens on "Most likely to remember everyone's name after one introduction" and only later arrives at "Most likely to stay much later than they originally planned". You are not being asked for anything real in the first five cards, which is exactly the point: by card fifteen the cost of an honest answer has already fallen. That is Aron's escalation, applied to a room rather than a pair.
The whole room answers, not one person at a time. Formats like "Most likely to" and "Would you rather" get every person committing to a position on the same card. Reciprocity is built into the format rather than left to the host to enforce.
Anyone can pass, no reason needed, and we say so before the first card. This looks like a safety feature and is actually a connection feature. Disclosure only builds closeness when it is voluntary; a room where passing is allowed answers more honestly, not less, because the honest answer stops being a trap.
The formats rotate. Point at someone. Take a side and defend it. Choose between two things. Tell us a story. Those are four different social muscles, and a round that only uses one turns into a survey by card ten, regardless of how good the individual questions are.
What we deliberately do not do
Each of these is a rule we hold ourselves to in writing, and each exists because it protects the engine above rather than because it sounds responsible.
No accounts, no names, no everyone-installs-the-app. The first answer has to be cheap, and a signup screen is the most expensive thing you can put in front of a room that has already sat down. Nothing in Chinwag requires anyone but the phone's owner to have it.
No scores, no streaks, no profiles. The moment a question game keeps score, people start answering for the score. Connection is not a performance and it does not have a leaderboard.
Nothing that forces a disclosure. No card demands sex, trauma, money, identity, or a crush in the room. Not because those topics are forbidden between friends, but because a card cannot know whether this room is one where they are welcome, and coerced disclosure does not produce closeness. It produces silence for the rest of the night.
Nothing that makes one person the joke. Humiliation, dares and targeted cruelty all get a laugh, once. Then the person it landed on stops answering, and so does everyone who noticed what happened to them. It is the fastest way to end the thing you started the game for.
No drinking required. A prompt might mention an optional sip. Making a drink the price of passing turns the pass rule into a penalty, and you have lost the thing that made the room honest.
Nothing interrupts a live round. No advert, no paywall, no rate-us prompt, no notification. A round that gets interrupted at card twelve does not resume, because the room has already picked up their phones.
These are edits, not opinions
Every rule above has been used to delete prompts we had already written and liked. Questions that would have put a spotlight on one person's history in front of a room that could not unhear it. Questions where a truthful answer had a real-world cost the card had no way to know about. Questions that were funny to write and produced nothing once read aloud.
The largest category we cut was not unsafe at all. It was flat: whimsical hypotheticals with nothing at stake, which produce an answer and no conversation. That is the most common way a card is wasted, and the easiest one to miss, because it never causes a problem. It just quietly does nothing.
Which is why a deck beats a list
Every rule here depends on knowing who is holding the phone. The same card can be flat in one room and the best question of the night in another, and the register that works for a group of old friends is not the register that works for people who met an hour ago. A list of five hundred questions has no idea which of those it has landed in. A deck is written for one of them on purpose, which is the only reason it can escalate at the right speed and stop at the right place.
The New Friends deck is free and does that escalation for you across three stages, from questions that cost nothing to answer through to the ones worth staying up for. If you would rather read the design side of the same argument, see what makes a question game good.
Read nextHow to host a question gameNine things good hosts do differently →