Create AI newsletters from YouTube with LangChain, Gemini, Apify & Gmail

Created by

Last update

Last update 6 hours ago

Video → Newsletter AI Agent

This n8n workflow converts a YouTube video into a polished, email-ready newsletter. It scrapes the transcript, extracts a thumbnail/logo and brand color theme, uses multiple AI agents to (1) clean & summarize the transcript into three newsletter sections, (2) convert that content into a styled HTML newsletter (color-aware), then saves the draft to Google Sheets and sends the email to subscribers via Gmail. The flow is optimized for batch sending and brand-consistent HTML output.

How it works (step-by-step)

Trigger — On form submission accepts Brand Name, Brand Website, and YouTube video link.
Site scrape & colour study — HTTP requests + Information Extractor → AI agent derives brand color theme (primary/secondary/accent/background).
Transcript retrieval — Two YouTube transcript scrapers (Apify acts) fetch the video transcript and thumbnail; a small Code node merges transcript chunks.
Summarization & journalism — AI Agent2 (LangChain/Gemini) cleans the transcript, extracts thesis + key points, and writes 3 newsletter sections in a journalistic tone.
HTML conversion — Convert Newsletter to HTML (AI) agent applies the fixed layout and injects only text color variables (keeps layout intact) and outputs Subject + HTML body (≤1000 words).
Aggregate & merge — Merge + Aggregate assemble files, assets, and parsed outputs.
Save & send — Save the email draft to Google Sheets (Save Newsletter Draft in Google Sheet) and loop through subscribers from a subscribers sheet; Sending Emails to all the Subscribers (Gmail node) sends the HTML to each address in batches.
Batching & looping — Split In Batches handles large subscriber lists; Loop Over Items triggers the HTML-conversion per recipient batch.

Quick Setup Guide

👉 Demo & Setup Video
👉 Sheet Template
👉 Course

Nodes of interest

On form submission (formTrigger) — entry point for video + brand inputs.
You Tube Transcript Scraper, You Tube Transcript Scraper1 (HTTP Request → Apify) — transcript + thumbnail fetching.
Information Extractor & AI Agent1 — website color/theme extraction.
Code in JavaScript — merges transcript pieces into a single text payload.
AI Agent2 (LangChain agent + Gemini Chat Model) — transcript → journalist-style newsletter sections.
Convert Newsletter to HTML (AI) (LangChain agent + Structured Output Parser) — builds constrained, brand-aware HTML email and subject.
Structured Output Parser1/2 — enforce schemas for color theme / structured outputs.
Get row(s) in sheet & Save Newsletter Draft in Google Sheet (Google Sheets) — subscriber list + draft storage.
Loop Over Items / Split In Batches — batch processing for sends.
Sending Emails to all the Subscribers (Gmail) — SMTP/OAuth send.
OpenRouter Chat Model — LM compute provider configured in the workflow.

What you’ll need (credentials & resources)

Google Sheets OAuth2 (for reading subscribers & saving drafts).
Gmail OAuth2 (for sending HTML emails).
Gemini / LLM provider credentials (Gemini API key or equivalent) for the LangChain agents.
Apify API key (for the YouTube transcript scrapers).
ConvertAPI (or similar) key if you convert logos (SVG→PNG) server-side.
Host storage / publicly accessible URLs for images (thumbnails, logos) or a file-store (S3).
Optional: SendGrid / Mailgun credentials if you swap Gmail for a transactional email provider.
Security note: do NOT hardcode credentials in node parameters; use n8n credentials manager or environment variables.

Recommended settings & best practices

Batch size & rate-limits: set Split In Batches to a conservative batch size (e.g., 50–200) and add delays between batches to avoid provider rate limits and Gmail throttling.
Retries & timeouts: enable retries for HTTP Request nodes and set sensible timeouts (e.g., 30–60s). Use exponential backoff.
LM controls: set token/response length limits and max_output_tokens (or equivalent) to avoid runaway costs; enforce the 1000-word HTML hard limit in the prompt.
Validation: validate the YouTube URL and that transcript content exists before invoking AI summarization (fail fast with a clear error).
Schema enforcement: use Structured Output Parser nodes with strict JSON schemas to prevent malformed outputs.
Testing: run with a small subscriber test sheet and use a safety test Gmail account before sending to production lists.
Logging & monitoring: log each run (video URL, subject, send count, errors) to a monitoring sheet or external logging service.
Privacy & compliance: ensure recipients have consent to receive emails (store opt-ins); include unsubscribe handling if you move beyond one-off sends. Comply with CAN-SPAM / local laws.
Credential rotation: rotate API keys periodically and revoke compromised tokens.
Content safety: instruct the LM agents to avoid hallucinated citations — only include links you can verify.

Customization ideas

Multi-language support: auto-detect video language and run summarizer in that language.
A/B subject testing: generate 2–3 subject lines and send variations to subsets.
Scheduling: add a scheduler node to delay sends or publish at optimal send-times per recipient timezone.
Integrate with SendGrid/Mailgun for higher throughput and analytics (opens/clicks).
Add personalization tokens (first name, company) from subscribers sheet to the HTML (merge fields).
Auto-attach transcript as plain-text footer or include “Read more” link to a hosted full article.
Add analytics: record opens, clicks, and engagement back into Google Sheets or a database.
Support other platforms: ingest videos from Vimeo, Loom, or uploaded MP4s.
Use a templating engine to allow multiple newsletter layouts and style variants.
Auto-generate social posts (Twitter/X, LinkedIn) from the newsletter summary.