Auto-reply to YouTube comments by type, logged in Supabase

What it is

This n8n workflow replies to the newest comments on one of your YouTube videos. You paste the video’s link in n8n’s chat. It reads the five latest top-level comments and, in a second branch, fetches the video’s transcript from a service you name. Each comment is looked up in a Supabase table and skipped if it is there, while a new one is saved and then labelled by OpenAI as criticize, compliment or specificdoubt. Praise gets a thank-you. A question gets an answer drawn from the transcript, and criticism gets no reply. Every reply is posted under its comment and written back to that comment’s row.

1. You paste a YouTube link in n8n’s chat
2. YouTube returns five comments, your service the transcript
3. Supabase skips handled comments and saves new ones
4. OpenAI labels each one and writes a thank-you or an answer
5. The reply goes up on YouTube and into Supabase

Who it suits

Creators who get more questions under their videos than they can answer, and who accept replies that a model writes and posts under their channel’s name. Skip it if a person must read each reply before it goes up. Nothing in the file waits for approval.

Where to use it

Any n8n signed in to the channel that owns the videos. Replies go out as that channel.

• n8n Cloud or your own server: the same file and steps
• n8n’s chat: where you paste each video link
• Your YouTube channel: replies appear under the comments, posted by the account the YouTube credential signs in with
• Supabase: the table holds every comment seen and, where one was posted, its reply

Setup

Get it saves a zip with workflow.json and Mindket’s setup notes in your downloads folder. Its author puts setup at about 13 minutes. The transcript call had no address in the original copy, so this one holds https://YOUR_TRANSCRIPT_SERVICE_ADDRESS until you name a service.

1. In n8n, create a workflow, open its menu, choose Import from File and pick workflow.json.
2. Add a YouTube OAuth2 credential and select it in Get Video Snippet, Comment Response and Comment Response1.
3. Add an OpenAI credential and select it in gpt-4o-mini, gpt-4o-mini1 and gpt-4o-mini2.
4. Create the YT_Comment table in Supabase, add a Supabase credential with your project URL and service role key, and select it in the three Supabase nodes.
5. In Get YouTube Transcript, replace https://YOUR_TRANSCRIPT_SERVICE_ADDRESS with your service’s address, and change the body if your service expects another shape.
6. Save, open the chat, paste the link of a video of yours with a few comments, and read the replies on YouTube before you rely on it.

The transcript call sends this body, with the video’s ID filled in. Cleanup transcript then reads transcript from the answer.

{
  "startUrls": ["https://www.youtube.com/watch?v=<video ID>"],
  "language": "Default",
  "includeTimestamps": "No"
}

How to use it

Paste the link of one of your videos, a youtube.com/watch?v= address or a short youtu.be one. Each run handles the five newest top-level comments. A comment already in the table is skipped, so a later run picks up only what is new among the latest five.

Example

A viewer writes under a tutorial: “Which plan do I need for the API step?” The comment is new to the table. It is saved and sent to the classifier, which answers:

{ "out": "specificdoubt" }

Specific Doubt Reply reads the comment with the cleaned transcript and writes an answer from what the video says. Comment Response1 posts it under the comment, and Update responce to Database writes it to the row’s reply column.

For newcomers and experts

If you are new. n8n draws a workflow as boxes joined by lines, and each box, a node, does one job. This one has 26 nodes, among them a loop that takes one comment at a time. Supabase is a hosted database, used here as the list of comments already seen. A credential is a key n8n keeps for you. The YouTube one signs in as your channel, which is why the replies carry its name. Try it first on a video with only a few comments.

For experts. The classifier and both reply chains run chatgpt-4o-latest, though their model nodes are named gpt-4o-mini. A structured parser holds the classifier to {"out": "..."}.

Cleanup transcript decodes HTML entities, drops filler words such as So, Now and Okay, and strips patterns for one channel’s usual greeting, sign-off and subscribe line.

Good to know

Replies go out under your channel’s name, and no one checks them first.

• No approval step: a wrong answer is posted as written, so read the first runs’ replies on YouTube
• Saved first: a comment is written to the table before it is labelled, so if its reply fails it is never tried again
• Criticism: those comments stay in the table with no reply
• The newest five: older comments outside that window are never read
• Other labels: Switch has no fallback, so a label outside the three leaves that comment unanswered and ends the loop there
• Several questions: the transcript joins the question chain through one Merge, so on your first run check that a second question also gets its answer
• Billing: up to two OpenAI calls per new comment, plus whatever your transcript service charges
• Privacy: comments and the transcript go to OpenAI, and comments and replies to your Supabase project
• The original copy: its Get Video ID and Cleanup transcript code had lost their backslashes, which stopped the first from compiling and made the second turn every letter s into a space. This one has them back
• Credentials: the JSON holds none, and nothing from its author’s accounts