Build
02concept system

Content Pipeline

Analytics → research → script → thumbnail test

Channel analytics in, tested thumbnails out — one pipeline.

A YouTube growth pipeline that reads what the channel's own retention data is telling it, mines the winning topic space, shapes a script around the drop-off points, and ships thumbnail variants into an A/B test — so every upload is informed by the last one.

4 stages
collapsed into one run
3–4
thumbnails tested, not guessed
closed
loop — each upload trains the next
Concept system — fictional channel data, real architecture. Shown to demonstrate how I'd wire the pipeline.
— the old way

Most channels guess. Topic picked on vibes, script written blind to retention, one thumbnail shipped on gut. The data that would close the loop sits unused in YouTube Studio.

— the workflow
01Signal
YouTube Analytics API
02Research
search · competitor scrape
03Angle
Claude · hook rubric
04Script
Claude · retention map
05Thumbnail
muapi · Higgsfield
06A/B test
thumbnail test · CTR read
— the system, running

Retention curve drives the brief; the thumbnail A/B board picks the winner before publish.

Content Pipelinerunning
audience retentiondrop-off · 0:31
7.9%
Avg CTR
4:12
Avg view dur.
182K
Views / 48h

→ Re-hook scripted at 0:31 to recover the drop-off.

research · ranked topics
“The system underneath” teardown94
Why your AI output sounds generic88
I automated reporting in a weekend81
Multi-agent vs one big prompt76
thumbnail A/B test · CTRlocked to channel visual system
A winner
var A9.4% CTR

face + bold type

B
var B7.1% CTR

product UI crop

C
var C6.3% CTR

contrarian text

— stage by stage
01Signal

Read the retention data

Pulls channel analytics — audience-retention curves, CTR, average view duration, top-performing videos — and locates the exact second viewers drop off.

YouTube Analytics API
02Research

Mine the winning space

Maps competitor outliers, rising search demand, and comment-section questions into a ranked topic board. Surfaces the gap between what's working and what hasn't been made yet.

search · competitor scrape
03Angle

Hooks, scored

Generates concept + hook variations and scores each on curiosity, specificity, and fit to the channel's proven patterns. Weak hooks die here.

Claude · hook rubric
04Script

Retention-shaped script

Drafts the script around the retention curve — re-hooks placed at the historical drop-off points, cold open matched to the winning angle.

Claude · retention map
05Thumbnail

Generate the variants

Produces 3–4 thumbnail directions from the title and concept, locked to the channel's visual system — same type, same colour grade, same face crop.

muapi · Higgsfield
06A/B test

Test, then publish

Runs the variants through a CTR test, reads the winner, and only then greenlights publish. The pipeline's output feeds the next run's Signal stage.

thumbnail test · CTR read
— the system underneath

The edge isn't any single step — it's that the loop is closed. Most teams treat analytics, research, scripting and thumbnails as four disconnected jobs done by four people on four timelines. Wired into one pipeline, last week's retention curve literally writes next week's cold open, and no thumbnail ships without earning it. The system compounds; manual workflows reset to zero every upload.

— stack
YouTube Analyticsretention + CTR signal
Research agenttopic + competitor mining
muapi / Higgsfieldthumbnail generation
Claudeangle scoring + scripting
pipelineYouTubethumbnailsA/B