FIRST-PARTY DATA

YouTube Transcript Speed & Caption Source Report

A live aggregate view of how TubeScript processes YouTube transcripts: caption source mix, cache coverage, AI fallback usage, and request speed.

Updated June 16, 2026

Cached transcripts
30,247

Durable transcript rows available for fast reuse and shareable transcript pages.

30-day requests
56,912

Transcript API and web requests logged in the last 30 days.

30-day cache hit rate
0%

Share of recent requests served from existing transcript cache.

Cached median
30 ms

Median cached response time from the latest sampled transcript requests.

SOURCE MIX

Captions first, AI when needed

TubeScript tries to avoid expensive AI transcription whenever YouTube already has a usable caption track. The cache records whether a transcript came from manual captions, auto captions, or AI fallback.

Manual YouTube captions
718
2%
Auto-generated YouTube captions
27,769
92%
AI fallback transcripts
1,760
6%
Legacy rows without source metadata
0
0%

What this means for searchers

The fastest YouTube transcript workflow is not pure AI transcription. It is source-aware: use YouTube's manual captions when available, use auto captions when they are the best available source, and reserve AI transcription for videos that have no usable captions.

This is why TubeScript now behaves differently from slower generic video transcribers. A fresh captioned video can finish in seconds, while a cached transcript can return almost immediately. AI fallback remains important, but it should be the exception rather than the first move.

Recent uncached requests have a sampled median of 6.1 sec. Cached requests have a sampled median of 30 ms. The exact timing varies by video length, caption availability, and provider response time.