You have one specific question, and the answer is somewhere inside a forty-minute video. You know the drill: open it, scrub the timeline, hover the chapter markers (if any), skip around at 2×, land close, overshoot, back up — four minutes of manual archaeology for ninety seconds of answer. Multiply across every how-to, review and lecture you consult in a week, and scrubbing is a real tax on knowing things.

Titles won't help — they describe the video, not its minutes. Chapters help when creators bother. What you actually need is the search operating inside the videos: not "which video discusses X" but "which minute of it."

deep_search answers at the minute level because it reads transcripts, not titles. Every result card carries an evidence pointer: a timestamp with a snippet showing exactly where in the video your query is addressed — sometimes a tight segment ("your answer runs from 12:40"), sometimes flagged as discussed throughout, so you know before clicking whether it's a jump or a watch.

The habit change is small and compounding: ask the specific question instead of searching the general topic, and you land inside videos at the right minute — across tutorials ("the part where they configure X"), reviews ("what do they say about battery"), lectures ("the segment on Y"). The scrubbing tax just stops being charged.