While building a small transcript exporter for YouTube, I hit a silent failure: the panel-based transcript path returned zero rows, with no error anywhere.
The old selector ytd-transcript-segment-renderer matches nothing on current YouTube. The transcript panel now renders <transcript-segment-view-model> elements, with the timestamp in .ytwTranscriptSegmentViewModelTimestamp and the text in span[role="text"].
What I verified in a real browser (headless Chrome, extension loaded, panel opened on a real video): the new selectors returned 24 cues / 2,066 characters, and the TXT / MD / SRT / VTT exports built from those cues came out at 2,335 / 2,668 / 3,142 / 3,162 bytes, spot-checked correct. The old selector returned 0 hits on the same page.
The fix is boring, and that is the point: try the new DOM first, fall back to the legacy renderer rows, and treat "zero rows" as a signal to re-check the DOM rather than as an empty transcript.
function panelRows() {
const rows = [...document.querySelectorAll('transcript-segment-view-model')];
if (!rows.length) return null; // fall back to the legacy renderer rows
return rows.map(row => ({
time: row.querySelector('.ytwTranscriptSegmentViewModelTimestamp')?.textContent?.trim() ?? '',
text: row.querySelector('span[role="text"]')?.textContent?.trim() ?? '',
}));
}
Two caveats from the same testing session. First, the panel only gives the rows currently rendered (24 on the first screen here), not the full transcript; the timedtext endpoint is the only full-length path. Second, from a datacenter IP, timedtext returned empty bodies even inside a real browser — real users on residential IPs should be fine, but I could not verify that from my environment.
The exporter itself is in review on Edge. Its privacy page is at https://alapha888.github.io/yt-transcript-export-privacy/
Top comments (0)