No Bannerbear templates needed.
Thumbnails generated via Shotstack HTML rendering — brand colors and headlines injected automatically per video.

One-time 2-minute setup.
Sign in as compoundclarityai@gmail.com and create the YouTube channel. OAuth credentials are already set.

When topics drop below 10, Claude auto-generates 30 new topics. Runs 30 min after daily production at 9:30AM ET.

Every Monday Claude searches the web for trending topics in each niche. Viral angles, recent news, high-CPM questions injected with priority 8-10.

Word overlap similarity check at 65% threshold. Runs before production daily. Stops the channel publishing near-identical videos.

Every Monday best-performing content pillars get their queued topics priority boosted automatically. Queue always leads with what works.

Every morning: total videos produced, revenue estimate, API costs, per-channel status, and failure alerts — all in one clean email.

If any channel's daily run fails at 9AM, it's automatically retried at 11AM ET. Zero missed production days. Failed runs are logged with full error details.

Estimates monthly revenue per channel using niche CPM rates ($12-50 range) multiplied by a view ramp model based on channel age. Updated monthly and logged to Supabase.

Calculates API costs vs AdSense revenue per channel. Shows exactly when each channel becomes profitable based on current CPM and video output rate.

Every Monday Claude reads the top 5 comments on each recent video. For positive/neutral comments it writes a genuine, specific reply that adds value and asks a follow-up question. Never generic.

Every Monday Claude writes a niche-specific community post — poll, insight, or question. Rotates through formats to keep the audience engaged between video uploads.

Checks subscriber count every day at 10AM ET. When a channel hits 100, 500, 1K, 5K, 10K, 50K, or 100K — it automatically posts a celebration to the Community tab and sends you an email.

Every Monday Claude reads up to 300 comments across recent videos. Returns % positive/negative, top recurring themes, audience questions, and content improvement suggestions — logged to Supabase.

Before every video: Claude scores the script on 4 dimensions — hook strength, specificity, pacing, CTA (0-10 each). If total < 6.0, the script is automatically regenerated with specific feedback. Up to 2 regeneration attempts, best version wins.

Every Monday Claude searches the web for top channels in each niche, identifies what they haven't covered well, and injects those gap topics into the queue with priority 8-9. Your content always hits angles competitors miss.

Every video generates two thumbnail variants:
A: Brand dark background, bold statement format
B: High-contrast light background, question framing
After 48 hours YouTube Analytics CTR determines the winner automatically.

Every Monday: checks where each video ranks in YouTube search for its target keyword. Rank history logged to Supabase. Low-ranking videos get their topic re-added to the queue with a better angle.

Sequential (old): 20 channels × 5 min = 100 minutes
Concurrent (Phase 8): 20 channels × same work = ~5-8 minutes

All channels run as parallel async tasks. One failure doesn't affect others.

Five dimensions scored weekly: production consistency (20) · topic queue depth (20) · pipeline reliability (20) · audience sentiment (20) · content volume (20). Claude writes one-line insight per channel.

Tracks every channel's path to YouTube Partner Program: 1,000 subscribers + 4,000 public watch hours. Shows estimated date to eligibility based on current growth rate. Auto-alerts when any channel qualifies.

1st of every month: Claude generates fresh trending tags for each niche, then updates all uploaded videos automatically. Old content stays discoverable as search trends shift. Runs on 1st of month at 12PM ET.

The gap being closed: Analytics data has been collected since Phase 4 but never fed back into the topic queue. A video hitting 2x average CTR is a signal the angle works — the factory now doubles down automatically, generating 5 similar topics and injecting them at the front of the queue.

How it works: Weekly after analytics runs → Claude identifies CTR winners → analyzes what made the angle work → generates follow-up topics with the same structural DNA → injects at priority 10.

The gap being closed: Every A/B thumbnail test logs a winner but the results are never analyzed as a pattern. By video 100, the factory has 100 data points on what works — but was ignoring all of them.

How it works: Monthly → Claude analyzes ALL resolved A/B tests per niche → identifies patterns (dark vs light bg, statement vs question, number prominence) → builds a winner profile → future thumbnails generated from proven templates, not fresh guesses.

The gap being closed: Every week scripts score 9+ and get used once. The structural DNA — the hook type, narrative arc, pacing rules — is never preserved.

How it works: Weekly → Claude extracts the structural blueprint from every 9+ script → saves hook pattern, narrative arc, CTA style → future ScriptEngine calls check for a matching template → new scripts START from a proven framework instead of from scratch.