Frequently asked questions

Most question apps hand you a static list and call it a day. Aperi actually pays attention to how you react. Every rating you give — like, dislike, neutral, love, skip — feeds a scoring model that reshuffles what comes next. After about 15 ratings the recommendations stop feeling generic and start feeling oddly specific to you.

Anyone who wants better conversations. Couples use it the most, but it works just as well for friends, roommates, or solo self-reflection. The algorithm doesn't care about your relationship status — it just learns what topics and depths resonate with you personally.

Right now it's a web app that works on any device with a browser — phone, tablet, laptop, whatever you have. Native iOS and Android apps are in the works and will be available after the beta period.

Soon. If you join the waitlist, you'll be among the first to get access. Beta users get to shape the product directly — we actually read the feedback and ship changes within days, not quarters.

Most people spend around 5 minutes — rating a question, maybe reflecting on it, sometimes discussing it with someone. Some questions spark a 30-minute conversation. Others you just rate and move on. There's no timer and no pressure to do more than feels right.

Nothing bad. There are no streak penalties or guilt notifications. Your taste profile and progress sit exactly where you left them. Aperi is designed around a daily habit, but life happens and the app doesn't punish you for it.

Your account tracks one personal taste profile. You can use Aperi solo or pair up with one partner at a time. When paired, you get the double-blind reveal experience — both of you answer before seeing each other's response. Your personal preferences carry over if you unpair and start fresh with someone else.

Yes, because the whole point is personalization over time. Without an account, the algorithm can't remember your preferences between sessions. During onboarding you pick the topics and depth levels you're drawn to — that's what kickstarts the algorithm before you've rated anything.

Not yet, but it's planned. Right now all questions go through a multi-step pipeline — generation, human review, deduplication checks, and quality scoring. An evaluation algorithm makes sure questions aren't repeated and meet quality standards. We want user-submitted questions to go through the same process so quality stays consistent.

It combines five scoring signals: tag preferences (35%) learn which topics you like, Wilson Score (25%) ranks questions by how well they've been received globally, depth fit (20%) matches your comfort level, freshness (10%) bumps newer content, and exploration (10%) occasionally throws in something unexpected so you don't get stuck in a filter bubble.

Every time you rate a question, the algorithm updates a Bayesian preference model for each topic tag on that question. Rate a vulnerability question as "love" and the model increases your affinity for that tag. Over time it builds a detailed map of what resonates with you, then uses Thompson Sampling to balance your known favorites with unexplored territory.

The algorithm doesn't know you yet, so it leans on global popularity — questions that most people rate highly. As you rate more, a blending factor (alpha) gradually shifts weight from global scores to your personal model. By rating 15, personalization takes over almost entirely.

Your comfort level is tracked with an Elo-like rating system, similar to chess rankings. Skip or dislike deep questions and your comfort score dips — the algorithm eases back to lighter topics. Love them and it leans in. There's built-in hysteresis so a single skip doesn't suddenly drop you two levels.

Two things. First, 10% of the composite score goes to exploration — Thompson Sampling occasionally picks a question from a category you haven't tried, just to check. Second, temperature annealing means new users get more variety while established users get more of what they love. The balance shifts gradually, never all at once.

The depth progression model draws from Arthur Aron's interpersonal closeness studies (the "36 Questions" experiment) and John Gottman's work on relationship dynamics. The scoring system borrows from information retrieval — Wilson Score lower bounds, Bayesian inference, and multi-armed bandit theory. It's academic machinery applied to a simple daily question.

Wilson Score is a statistical method for ranking items with few ratings — the same approach Reddit uses for comment sorting. A question with 3 likes out of 3 ratings shouldn't outrank one with 95 likes out of 100. Wilson Score accounts for sample size so newer questions don't unfairly dominate or get buried.

1600+ questions across 8 categories and 4 depth levels — and counting. Every question goes through a generation-and-review pipeline before it's published. New questions are added regularly, and the scoring system surfaces them alongside proven favorites.

Level 1 (Icebreaker) is fun and low-stakes — hypotheticals, preferences, silly "what ifs." Level 2 (Personal) gets into opinions and experiences. Level 3 (Vulnerable) touches on fears, regrets, and insecurities. Level 4 (Deep) is existential — identity, mortality, core beliefs. The algorithm decides when you're ready for each level based on your rating patterns.

Eight categories that cover a wide range: from everyday life and fun hypotheticals to intimacy, values, and future planning. Each category has questions at all four depth levels. The algorithm tracks your category preferences independently, so loving "Fun & Playful" questions won't flood you with only those.

Questions that go past surface-level small talk. Instead of "What's your favorite movie?" you get things like "What belief have you changed your mind about in the last five years?" Aperi organizes these by depth so you build up to the harder ones gradually rather than jumping straight into vulnerability with no warmup.

They're completely separate. Daily questions are chosen by the algorithm based on your taste profile and depth level — you don't pick them and they don't come from packs. Question packs are themed collections ("Before Marriage," "Money & Values," "New Parents") that you browse and work through at your own pace. Pack questions never appear in your daily feed. Think of daily questions as the algorithm's picks and packs as your picks.

Every question runs through a multi-gate validation pipeline: toxicity screening, readability checks, semantic deduplication (so you don't get the same question rephrased), an LLM judge for tone and depth accuracy, and tag verification. Questions that don't pass all five gates go to human review or get rejected.

Yes — when the algorithm presents a question, you see the full text, its category, and its depth level before deciding what to do. You can rate it (like, dislike, neutral, love) or skip it entirely. Skipping still gives the algorithm a signal, but a weaker one than a direct rating.

Unlikely. With 1600+ questions and new ones added regularly, the algorithm has plenty to work with. It also won't repeat questions you've already rated. If you somehow exhaust a category, it'll explore others — which is usually where the most interesting discoveries happen anyway.

One algorithmically chosen question per day, access to 3 starter categories, full personalization, and all four depth levels. The free plan isn't a crippled trial — it's a real daily experience. Premium gives you more volume, all categories, voice memos, and custom delivery times.

There's no free trial. Subscribe for a month, try everything, and cancel if it's not for you. The free plan already gives you the core daily experience with full personalization — premium adds more categories, voice memos, and flexibility.

Yes. If you're not satisfied with a premium subscription, contact us within 14 days for a full refund. Question pack purchases are final since you get immediate access to all the content, but we're reasonable — if something went wrong, reach out.

We don't sell your data, period. Your ratings feed the algorithm but only your personal model sees the specifics. Global question quality scores use aggregated, anonymized signals — no individual behavior is exposed. We don't run ads and we don't share data with third parties.

Yes. You can delete your account from the settings page and all your data goes with it — ratings, preferences, the personalization model, everything. Deletion is permanent and happens immediately, not after a 30-day "cooling off" period.