The 30-second personalization problem
Most reading apps take days to learn what you like. Here's why txtfeed calibrates in three votes — and what that means for first-session retention.
Every reading app makes the same promise: tell us what you like and we'll personalize your feed. The problem is what "tell us" means in practice. Most apps ask you to subscribe to topics, follow accounts, or rate a starter pack of articles before they'll show you anything useful. By the time the algorithm has enough signal to be helpful, you've already left.
This is the 30-second personalization problem. Reading is a low-commitment activity. Users land, scroll for thirty seconds, and decide whether the app is worth coming back to. If the feed feels generic in those thirty seconds, they're gone. The window for proving you're worth their time is brutally short.
txtfeed's answer is to start the algorithm before the user does anything. The first feed you see isn't random — it's trending content filtered by your detected language, your local time of day, and your geographic region. Morning in São Paulo gets different content than evening in Berlin. This isn't personalization yet, but it's already better than "top stories worldwide" and it costs the user nothing.
Then the real signal kicks in. The first vote you cast tells the algorithm two things: what you like (the topic, source, and length of the piece you upvoted) and what you skipped to get to it (everything you scrolled past without voting). Three votes later, the algorithm has a working profile. By the fifth, the feed visibly shifts.
The reason this works isn't algorithmic sophistication — it's signal density. A single upvote on a 600-word essay about distributed systems is worth more than ten passive views of the same article. Reddit figured this out twenty years ago. txtfeed is just applying the same principle to the discovery layer instead of the community layer.
The deeper lesson is that personalization speed is a UX problem, not an ML problem. The hard part isn't predicting what users will like — modern recommender systems do that well. The hard part is collecting enough signal in the first thirty seconds to make any prediction at all. Solve that, and retention follows.
See it for yourself. No signup required.
Open txtfeed