Adobe's own transcription has been the bottleneck for auto-editing forever — not a you problem, it's just not competitive. Two workflows that actually fix it: 1. Descript first — their transcription is Whisper-based, much cleaner. Edit word-by-word in Descript, then export the EDL back into Premiere. Keeps your existing setup, just removes the bad transcription layer. 2. Run Whisper separately — OpenAI Whisper API on your raw audio gives you a near-perfect transcript. Feed that into whatever auto-editor you're using instead of letting Adobe generate it. Both approaches basically cut Adobe out of the transcription step entirely. That alone fixes most of the junk cuts.