Built a free transcription tool for stupidly long recordings. Here's what actually converts to paying customers (and what doesn't).

Longscribe exists because every "free" transcription tool I tried had the same catch: a 30-minute cap, a 5-minute preview, or a per-minute meter that quietly adds up. I wanted something that didn't punish you for uploading a 3-hour recording, so I built one.

How it works, roughly: captions-first when the source already has them (YouTube mostly), Whisper as a fallback when it doesn't — same pipeline whether it's a YouTube link or you're uploading an audio file directly. Long files get chunked so a 3-hour podcast doesn't just time out or blow up memory on a small server. 600 free minutes a month, $4.99/mo for more plus batch upload.

811 users so far, and 6 of them are paying — modest, but real, and none of the 6 have churned yet. The part worth sharing is what actually drove those 6 conversions, because it wasn't what I expected.

I assumed the "you've used your free minutes" wall would be the main trigger to pay. It isn't. Of the users who hit that wall in the last 30 days, only 4.4% went on to buy anything. The people who actually paid did it fast — registered and checked out within minutes, often before running a single job — and came in disproportionately from one referral source (ChatGPT recommending the tool in response to someone's question). They arrived already intending to pay; the paywall didn't create that intent, it just had to not get in the way of it.

Honest gap: no speaker diarization yet. It's the most common feature request and I haven't shipped it.

Curious if anyone else running a freemium tool has found a way to convert the free users who aren't already primed to pay before they land on the page — so far our paywall placement seems to matter a lot less than where the traffic came from in the first place.