Video ideas from YouTube comments with web research

What it is

This n8n workflow reads a YouTube video’s comments for ideas you could film next. You paste the video’s link in n8n’s chat. It fetches up to 100 of the newest top-level comments, and OpenAI’s gpt-4.1-mini answers Yes or No for each one. The question is whether a viewer is asking for, or hinting at, a video worth making. Every Yes is written to a Google Sheet, together with the commenter’s name, the comment’s text and its like and reply counts, so you can sort the requests later. A second gpt-4.1-mini step then takes those rows one at a time, searches the web through Tavily and writes a topic into each. YouTube itself is only read.

1. You paste a YouTube link in n8n’s chat
2. YouTube returns up to 100 top-level comments
3. OpenAI keeps the ones that suggest a video
4. Each one becomes a row in Google Sheets
5. An agent with Tavily search adds a topic to each row

Who it suits

Creators whose viewers ask for videos in the comments, and the teams that plan content for them. It reads one video per run and leaves a sheet of requests you can sort. Skip it if you want the research write-up as well. The model writes one for each idea, but only the topic reaches the sheet.

Where to use it

Any n8n that can reach Google, OpenAI and Tavily will run it. The video can be any public one.

• n8n Cloud or your own server: the same file and steps
• n8n’s chat: where you paste each link
• Google Sheets: one row per idea, updated in place when the same comment comes round again

Setup

Get it saves a zip with workflow.json and Mindket’s setup notes in your downloads folder. Its author puts setup at about 18 minutes. Make the sheet first: a tab named Sheet1 with these eight headers in its first row.

id | text | likeCount | replyCount | publishedTime | author | contentIdea | topic

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 Comments.
3. Add an OpenAI credential and select it in Message a model and Message a model1.
4. Open Tavily and replace <YOUR_TAVILY_API_KEY> in its Authorization header with your key, keeping Bearer and the space after it.
5. Add a Google Sheets OAuth2 credential and select it in Append or update row in sheet and Update row in sheet. In both, replace YOUR_SHEET_ID with the ID in your sheet’s address.
6. Save, open the chat, paste a video link and watch the rows fill in.

How to use it

Paste a full YouTube link, a youtube.com/watch?v= address or a short youtu.be one. All the comments are judged first, then the saved rows go through research one at a time, so the topics arrive after the rows. A comment already in the sheet is matched on id and updated, never added twice.

Example

Under a video about running a small online shop, a viewer writes: “Could you show how to send the day’s orders to a spreadsheet every night?” The first model answers Yes. A row lands with the viewer’s name, the text and contentIdea set to Yes. The research step searches the web and answers in this shape, and only the topic lands in the row:

{
  "topic": "<a video topic built from the request>",
  "research_overview": "<300 to 500 words from the search, not saved>"
}

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 12. Tavily is a search service made for AI models, and here the model decides what to search for. Loop Over Items sends the saved rows through research one at a time, so a long list takes a while. Start with a video that has a few dozen comments.

For experts. Both model calls use gpt-4.1-mini through the OpenAI node’s message operation, with JSON output on and simplify off. That is why later nodes read choices[0].message.content.

Filter keeps an exact Yes in output. The judge’s prompt asks for a bare word, so if a run saves nothing, open Message a model and check which key it answered with. Append or update row in sheet matches on id, and Update row in sheet matches the same column and writes topic alone.

Good to know

Commenters’ names land in your sheet, and only the topic of each research run is kept.

• Personal data: the sheet stores each commenter’s display name beside the comment, so keep it private
• Research dropped: research_overview is written and then thrown away, since no column takes it
• One page: the comments call reads a single page of up to 100, newest first
• Billing: one OpenAI call for every comment read, and another, with Tavily searches, for every idea saved
• The Tavily key: it sits in the node’s header, not in a credential, so anyone who can open the workflow can read it
• The original copy: its Get Video ID code had lost its backslashes, so its fallback pattern could not compile and the node failed on every run. This one has them back
• Credentials: the JSON holds none, and nothing from its author’s accounts