Build a reviewable YouTube-description workflow with n8n

Keep the video and transcript linked

Luke’s June 2025 tutorial uses Telegram to start an automation, Airtable to manage records and an AI agent such as Gemini to create descriptions and timestamps. The workflow also connects YouTube and short.io. A useful metadata draft must describe the exact video rather than a similar topic or a generic channel template.

1. Choose one test video and record its exact identifier with the workflow input. Follow the Telegram trigger at 0:27 and preserve the associated transcript and record as the task moves between services.
2. Inspect transcript extraction and the regex step shown at 5:00. Compare a few passages with the source and keep any missing or malformed text visible before sending it into description generation.

Generate the draft and test a specific correction

The agent generates metadata at 1:26 and handles feedback at 2:43. A feedback loop is useful when it changes the intended draft while preserving approved content and the correct video association. A new plausible description is not automatically a correction of the original problem.

1. Review the description, title and chapter markers against the transcript. Check names, links, topic order and whether timestamps point to the content they label instead of assuming an SEO label proves accuracy or ranking.
2. Give one concrete correction and inspect the revised draft in its stored record. Compare the changed passage and confirm that unrelated approved wording and the target video remain intact.

Check links and the actual YouTube result

YouTube API connection appears at 3:46, titles and short links at 7:55 and the automated update at 9:55. Draft storage, URL creation and a live metadata change are separate stages. Preserve the full approved version so the final external action can be checked against it.

1. Verify each short link’s destination and retain its intended URL parameters. Confirm the account, exact video and complete final draft before authorising an external update.
2. After the approved update, reopen the video’s metadata and compare it with the stored draft. Check chapter formatting and links, keeping any failed or partial update distinct from a successful workflow execution message.

The takeaway

Make transcript fidelity and feedback visible before the publishing step. The workflow is complete when the intended video contains the approved metadata, not merely when the agent returns text.