What Counts as a Match
A match begins the moment a person enters the matching flow — technically, when the matching screen loads. We record that as a match requested event. A match is requested; it is not yet a conversation.
OpenTalk Research
Every number OpenTalk.tv publishes comes from the same pipeline: anonymous events, strict cleaning, fixed thresholds, and a human review before anything goes public. This page defines every term we use, in plain language, so a published result can be checked against the rules that produced it.
One thing before the definitions: our results describe only the sessions OpenTalk.tv itself observed. They do not describe any other platform, and they do not describe the video chat industry as a whole.
The Definitions
A match begins the moment a person enters the matching flow — technically, when the matching screen loads. We record that as a match requested event. A match is requested; it is not yet a conversation.
We call a conversation video ready when the live call screen actually opens. Two systems can report that moment: our own page tracking, which fires when the call screen loads, and the video platform's first-frame signal, which fires when real video is confirmed on screen. When both exist for the same conversation, the platform's first-frame signal wins — it is the stricter measure — and the page event is set aside.
Before any counting starts, every event passes a cleaning stage. Events are deduplicated by their unique ID, and a session is marked invalid or excluded when its data cannot be trusted: missing or malformed fields, impossible timestamps, zero or absurdly long durations, or a conversation with no coherent start and end. Borderline cases are held for review instead of being silently counted. Only valid sessions ever appear in a published metric — the raw and excluded counts are published alongside, so the cleaning is visible too.
Every conversation is tagged with how it started. A random match means the person entered the matching flow and let the pairing decide. A selected profile means they chose a specific person from a profile page before the call. We also distinguish arrivals from the who's-online gallery and direct entries into matching. When we cannot tell, the conversation is honestly tagged unknown rather than guessed.
Within one anonymous session, video conversations are numbered in order: the first conversation is position 1, the next is position 2, and so on. This lets us compare the very first match of a session against the matches that follow it.
Duration is the time spent on the live call screen, from the moment it opens to the moment it is left. When a conversation ends we record how long it lasted in milliseconds; published results round that into human units like seconds.
A conversation is counted as a skip when the person leaves the call within ten seconds and asks for a new match within the following minute. A short call that isn't followed by a new match is just a short call — leaving the site is not skipping.
While a conversation is live we note whether it reaches four fixed milestones: 10, 30, 60 and 180 seconds. Milestones show where conversations actually end — a format that loses most people before 10 seconds is a very different product from one that loses them after a minute.
Sessions are tied together by a random identifier generated in the visitor's own browser. It cannot be reversed into an identity, and we never attach an account ID, email address, phone number or name to research events. We never collect what people say, show, type or stream — research events describe the shape of a session (when it started, how long it lasted, how it began), never its content.
Each event carries a coarse device class — mobile, desktop or tablet — and a browser family such as Chrome or Safari, both derived from standard browser signals. We use this only to describe the mix of devices in a sample, never to fingerprint anyone.
Country or region comes from a network-level header and is approximate. More importantly, the countries represented in a sample simply reflect who visited OpenTalk.tv during the study window. Our traffic mix is not a cross-section of any country's population, and results should never be read as statements about a country or its people.
Our samples consist of people who chose to visit OpenTalk.tv and start conversations — a self-selected group, measured on one platform, during a specific window. That makes the results honest descriptions of what happened here, and nothing more. They are not estimates for other platforms, other audiences, or "people in general".
Research is published only as aggregates — rates, shares and averages across hundreds of sessions. We never publish session-level records, and minimum sample thresholds exist partly so that no published figure could ever describe a small, identifiable group. The Privacy Policy covers data handling beyond research.
Collected data is aggregated into a snapshot: a fixed study period with a raw count, a valid count and the excluded count. A snapshot can only be proposed for publication once the study clears its thresholds — a minimum number of valid sessions in every compared group, a minimum number of collection days, and a valid-data rate of at least 80%. A human then reviews the snapshot, including data-quality warnings, and decides: publish, keep collecting, or reject. Every number on a published study page traces back to exactly one reviewed snapshot with a fixed period — nothing is estimated, previewed or extrapolated before that moment.
Questions about these definitions, or about a published result that references them? The research hub lists every study, published and in progress.