AI Agent n8n: Auto-Publish to YouTube

The Manual Trigger node serves as the entry point for your entire automation. When you click the "Execute" button in n8n, this node kicks off the workflow and passes control to the subsequent nodes.

This is ideal for testing and on-demand execution. Once you've validated everything works, you can easily swap this trigger for a Schedule Trigger (to run daily at a specific time) or a webhook trigger (to respond to external events).

💡 Tip: For production use, consider replacing this with a Schedule Trigger set to run every morning. That way, any new audio files dropped into your folder overnight get processed automatically.

This Google Drive node searches for audio files in your designated folder. It's the first data-gathering step that identifies which files need to be processed through the transcription and metadata generation pipeline.

The node connects to your Google Drive, looks inside a specific folder, and returns all files found there. Each file becomes a separate item in n8n, allowing the workflow to process multiple audio files in sequence.

💡 Tip: Create a dedicated "to-process" folder in Google Drive. After the workflow runs successfully, you can add a final node to move processed files to an "archive" folder, preventing duplicate processing.

This node downloads the actual audio file content from Google Drive. The previous node gave us file metadata (ID, name, etc.)—this node retrieves the binary audio data needed for transcription.

The file ID is passed dynamically from the search results, meaning each audio file found in Step 2 gets downloaded automatically. The binary data is stored in n8n's memory and passed to the OpenAI transcription node.

💡 Tip: OpenAI's Whisper has a 25 MB file size limit. If your audio files are larger, consider adding a compression step or splitting longer recordings before upload.

This is where the magic begins. The OpenAI node uses the Whisper model to transcribe your audio file into text. This transcript becomes the foundation for all the AI-generated metadata that follows.

Whisper handles multiple languages and accents remarkably well, producing accurate transcriptions that capture what you actually said in your video. This raw text is then analyzed by the AI agent to extract key topics and generate relevant metadata.

💡 Tip: The transcription quality directly impacts your metadata quality. If your audio has significant background noise, consider running it through a noise reduction tool before uploading to Google Drive.

The AI Agent node is the creative engine of this workflow. It takes your transcript and transforms it into YouTube-ready metadata. This node defines how the AI should behave and what output format to produce.

The prompt configuration passes the transcript text to the AI, while the system message establishes the AI's role as a YouTube content strategist and SEO copywriter. This ensures the generated titles, descriptions, and tags are optimized for discoverability.

System Message (in Options):

You are an expert YouTube content strategist and SEO copywriter. Your task is to analyze the provided video transcript and generate: 1. A compelling, click-worthy title (under 60 characters) optimized for search 2. A structured description with hook paragraph, key takeaways, timestamps, call-to-action 3. 10 relevant, searchable tags. Focus on clarity, engagement, and SEO best practices.

💡 Tip: Customize the system message to match your channel's voice. If you have a casual, humorous style, instruct the AI accordingly. If you're more formal and educational, reflect that in the prompt.

This node configures the specific OpenAI model that powers the AI Agent. It's connected as a sub-node to the main AI Agent, providing the language model capabilities for content generation.

GPT-4.1-mini offers an excellent balance between quality and cost for metadata generation tasks. It's fast, accurate, and significantly cheaper than the full GPT-4 model while still producing professional-quality output.

💡 Tip: If you find the output quality lacking for complex topics, you can switch to gpt-4 or gpt-4-turbo. Just keep in mind the increased API costs and slightly longer processing times.

The JSON Schema node ensures the AI's output follows a consistent, predictable structure. By providing a JSON example, you tell the AI exactly what format to produce—making it reliable for downstream nodes like the YouTube upload.

This structured output parser is connected to the AI Agent and enforces that every response includes a properly formatted title and description. No more parsing inconsistent AI responses manually.

JSON Example:

{"title": "How I Automated My Entire Workflow with AI in 2026", "description": "🚀 In this video, I walk you through my complete automation setup...\n\n⏰ TIMESTAMPS:\n0:00 - Intro\n1:30 - Setup overview\n...\n\n👍 Like & Subscribe for more automation content!"}

💡 Tip: Expand the JSON example to include a tags array if you want structured tag output: "tags": ["automation", "n8n", "youtube"]. This makes tag handling cleaner in the upload node.

This secondary GPT agent node provides additional AI model connectivity for the workflow. Depending on your workflow structure, this may handle tag generation, description refinement, or serve as a backup processing path.

The configuration mirrors the primary AI model setup, ensuring consistent behavior across all AI-powered nodes in the workflow.

This Google Drive node searches for the video file that corresponds to your processed audio. It looks in a designated video folder to find the matching file that will be uploaded to YouTube with the generated metadata.

The node uses the folder ID to target the correct directory and returns all video files ready for upload. The search can be refined with a query to match specific naming conventions.

💡 Tip: Use consistent naming conventions between your audio and video files (e.g., episode-42.mp3 and episode-42.mp4). This makes it easy to add logic that matches audio transcripts to their corresponding video files.

This node downloads the actual video file from Google Drive, preparing it for upload to YouTube. The binary video data is stored in memory and passed to the YouTube upload node.

Like the audio download, the file ID is pulled dynamically from the search results, ensuring the correct video file is retrieved for each workflow execution.

💡 Tip: Video files can be large. Ensure your n8n instance has sufficient memory allocation. For self-hosted instances, consider increasing the N8N_DEFAULT_BINARY_DATA_MODE to filesystem for better handling of large files.