Gemini CLI: YouTube Shorts Viral Idea Scorer with Reddit/RSS/DeepSeek AI v1

This workflow uses n8n 1.82+ with Gemini CLI AI to turn a daily flood of 500+ trending stories into 3 prioritized Shorts ideas. It harvests headlines from 30+ RSS feeds (including TechCrunch, The Verge, Ars Technica, PC Gamer, Hacker News, and niche-specific blogs) plus hot posts from 10 subreddits in your niche. The agentic reasoning step uses Gemini CLI's deepseek-chat model to score each trend against a Supabase vector store containing your channel's historical performance data — past 90 days of Shorts with their view counts, swipe-away rates, and engagement metrics. Gemini CLI evaluates each trend on a 5-axis matrix: niche alignment (0-10), hook potential (0-10), competition density (0-10), evergreen score (0-10), and production feasibility (0-10). Trends scoring below 35/50 are filtered. The remaining trends are ranked and the top 3 are written to a Google Sheet and a Notion database. A human review step in Notion lets you mark ideas as selected, in-progress, or rejected. The workflow runs once daily on the Schedule Trigger and outputs 3 ready-to-shoot briefs with hook suggestions, hashtag sets, and reference video links.

BUSINESS PROBLEM

A YouTube Shorts creator or small content team spends 8-12 hours per week manually hunting for content ideas: scrolling Reddit, reading RSS feeds, checking competitor channels, and trying to guess what might go viral. The process is subjective — one creator's good idea is another's dud. Without data-backed scoring, you are guessing. YouTube Shorts drives 200 billion daily views as of 2026, and channels that post daily grow 41% faster. (Source: Loopex Digital, 2026). But the limiting factor is not production speed — it is idea quality. A single viral Short can bring 500,000+ views and 10,000+ subscribers, while a bad idea wastes a full day of production and gets 200 views. Without a systematic scoring approach, creators leave viral potential on the table. They pick ideas based on gut feel or what is trending in their bubble, not what their specific audience historically rewards. This workflow replaces gut-feel decisions with a repeatable data-driven scoring process that surfaces the ideas most likely to perform for your specific channel and niche.

WHO BENEFITS

YouTube Shorts creators with 1,000-100,000 subscribers who are stuck in the 500-2,000 view range and need data-backed idea selection to break through to viral territory. The scoring system identifies high-potential niches that your channel is already partially ranking for. Content agencies managing 5+ YouTube Shorts channels who need a standardized ideation process across clients. Instead of each channel manager running their own manual research, the workflow produces consistent scored briefs for all channels from a single daily run. Faceless Shorts channel operators who publish 14+ shorts per week across multiple niches. Manual ideation at this volume leads to burnout and repetitive concepts. The AI scoring prevents topic fatigue by surfacing diverse ideas from different RSS sources each day.

HOW IT WORKS

1. RSS Feed Harvest: n8n RSS Feed Read node fetches the latest 10 items from each of 30+ configured RSS feeds. Output: flat JSON array of 300+ items with title, URL, published date, and source. 2. Reddit Hot Posts: HTTP Request node to Pushshift API fetches top 25 hot posts from 10 subreddits in your niche (configurable). Output: 250 posts with title, subreddit, score, comment count, and URL. 3. Dedup and Normalize: Code node merges RSS + Reddit arrays, removes duplicates by normalized title hash, filters items older than 48 hours, and strips HTML entities. Output: deduplicated array of 400-600 unique items. 4. Channel Intelligence Load: HTTP Request to Supabase REST API fetches your last 90 days of Shorts performance data: title, thumbnail description, view count, swipe-away rate, average view duration, and engagement rate. This data is embedded into the AI prompt as context. 5. AI Scoring (Gemini CLI): HTTP Request node sends a batch of 25 trends per call to Gemini CLI API (deepseek-chat, temperature 0.3). The system prompt includes the channel intelligence data and scoring rubric. Gemini CLI returns scored JSON: trend_id, scores (5 axes), total, recommended hook angle, 5 hashtags, and a reference video suggestion. 6. Batch Processing: The workflow loops through all 400-600 items in batches of 25, collecting scored results into a single array. 7. Top 3 Selection: Code node sorts by total score descending and selects the top 3. For each, it generates a brief containing: the original source URL, the recommended hook sentence, the AI-classified content archetype, 5 hashtags, and the reference video URL. 8. Output: n8n Google Sheets node appends the 3 briefs as new rows. Notion API node creates a database entry for each with select status: Available. 9. Daily Reset: Workflow execution logs write to Supabase for monitoring, and the schedule trigger waits 24 hours for the next run.

TOOL INTEGRATION

n8n 1.82+: The workflow orchestration layer. Gotcha: Processing 400-600 items through Gemini CLI API in batches of 25 requires a loop with a Wait node (1 second between batches) to avoid rate limits. Without this, the Gemini CLI API returns 429 errors after batch 3. Gemini CLI API (deepseek-chat): AI scoring engine at platform.deepseek.com. Pricing: ~$0.28/M input tokens, ~$1.10/M output tokens. Each batch of 25 trends costs roughly $0.04-0.08 in tokens. Gotcha: Gemini CLI does not support JSON mode natively — you must include explicit JSON schema instructions in the system prompt and set temperature to 0.3 or lower to get consistent structured output. Supabase (Vector Store): Stores channel performance data. Use pgvector extension to store embedding vectors of past Shorts titles. Gotcha: The Supabase REST API has a 30-second timeout on free tier queries. If your performance dataset exceeds 500 rows, paginate with range headers or use the Supabase node with pagination enabled. RSS Feed Parser: n8n's built-in RSS node. Gotcha: Several major RSS feeds (notably TechCrunch and Ars Technica) have switched to partial-content feeds — the node fetches the truncated summary, not full article text. For idea scoring, titles alone suffice, but do not expect full article bodies from the RSS node. Notion API: For human review workflow. Integration token must have insert, update, and read capabilities. Gotcha: Notion API rate limits at 3 requests per second per integration. Batch your writes or add a Wait node set to 350ms.

ROI METRICS

1. Weekly ideation time: 8-12 hrs manual research -> 30 minutes reviewing AI-scored briefs. 2. Idea-to-viral hit rate: 1 in 20 manual picks hit 50K+ views -> 1 in 5-7 AI-scored picks (based on channel history matching). 3. Daily output capacity: 1-2 ideas/day manually -> 3 high-confidence ideas/day from AI pipeline. 4. Cost per idea at $50/hr labor: $25-50 per idea manual -> $0.15-0.30 in Gemini CLI API costs. 5. Metric measurable in week 1: number of scored ideas generated — expected 15-21 per week vs 5-10 manual.

CAVEATS

1. Gemini CLI score bias: The model may overweight recent trends (novelty bias) over evergreen content with proven long-tail performance. Manually review the evergreen score axis weekly and adjust weightings. 2. RSS feed breakage: Feeds change URLs or stop updating without notice. Monitor the feed failure rate in n8n execution logs and set up Slack alerts when >20% of feeds return errors. 3. Channel intelligence staleness: If you have not published a Short in 14+ days, the Supabase vector store has no recent signal data. In this case, the AI falls back to generic trend scoring without personalization. 4. This workflow does NOT write your scripts or produce your videos. It is an ideation and prioritization tool — creative execution remains manual.