A Save Predicts a Stream

Streamers using StreamGist make one decision on every game we show them: save it or skip it. We wanted to know whether that decision means anything, so we matched it against what those same streamers actually broadcast on Twitch. Between late February and mid May, about one saved game in ten went live on the saver's channel within 30 days. Skipped games almost never did, fewer than one in two hundred. Counted the strictest way we know, a saved game was 26x as likely to get streamed as a skipped one.

It happens quickly, too. Half of the saves that turned into broadcasts were live within three days, and three quarters within ten.

There's an obvious objection: popular games get saved and streamed for the same reason. They're popular. So we compared streamers who looked at the same game and chose differently. The one who saved it went on to stream it and the one who skipped it didn't, by 22 games to 1. The same split shows up channel by channel: whenever one streamer's saves and skips converted at different rates, the saves won. And the gap was there for niche games as well as heavily streamed ones.

Two caveats. This is prediction, not cause. StreamGist keeps saved games visible and rests skipped ones for a while, so part of the raw gap comes from our own product mechanics. That's why we quote the cautious figure and lean on the same-game comparison. And the result says nothing yet about whether demand before a launch predicts the launch itself. That answer has to wait until enough releases have passed through the panel.

The part nobody else can see is the skip. Platforms record what creators play and what audiences watch. No one records the games a creator looked at and declined. And not every pass is equal: skips made after some deliberation went on to get streamed several times more often than instant ones. That sample is still small, so we treat it as a direction rather than a measurement. The idea goes back to stream fit rejection.

Before publishing, we tried to break our own result nine different ways: recounting streamers who changed their minds, widening the name matching between store catalogs and Twitch categories, assuming the worst about streamers who left the panel, even counting streamers we never saw go live. The harshest rerun thinned the gap to roughly 19x. None of them erased it. Exact denominators, confidence intervals, and the full set of checks are documented and available to partners.

If creator adoption is part of a launch plan, this decision stream is the earliest read on it that exists. Our demand reads are built on it, always aggregated, never about an individual streamer. How we use AI and behavioral data is on our AI transparency page.