Splitting conversations
Prompts 2–4 are follow-ups. In a new chat they can return confident answers built on no dataset.
Faceless YouTube · Operator Playbook
The 4-Prompt Faceless YouTube Workflow.
One conversation, four prompts. Tear down a channel’s real public data, find what is working now, turn it into ideas built for a channel with no audience, and finish with briefs your writer and editor can execute.
Free resource · your setup stays in your browser
The measurement underneath everything
The outlier multiplier.
A video’s views divided by the rolling median of the ten uploads before it and the ten after it. That removes much of the channel-growth curve so you compare the video against its moment—not a year-one audience against a year-five audience.
VIDEO VIEWS÷21-VIDEO CENTERED MEDIAN=OUTLIER MULTIPLIER
Set it up once
Fill the blanks. Copy clean prompts.
Everything in square brackets is a variable. Enter your choices here and the page updates every prompt automatically. The method constants—28-day maturity, five-video sample floor, 20–30 second hook, and three reference videos—stay intact.
Workflow setup
Your analysis run
Base URL only—no trailing slash and no /videos.
30–90 is usually useful.
Three to five.
Only titles claiming “I bought” or “I tested.”
Nothing is submitted. Values are stored only in this browser so you can return without rebuilding your setup.
The complete workflow
Run these in order.
Prompt 1 does the heavy lifting. The next three are follow-ups that re-slice and extend the same evidence. Customize above, copy below, and mark each step complete as you run it.
Running the full workflow
One view of every handoff.
The source PDF’s operating table is restored here so each step’s inputs, outputs, and human decision are visible before you start.
| Step | Fill in | Produces | Your job |
|---|---|---|---|
| PROMPT 1Teardown | [CHANNEL_URL] | REPORT.md, videos_enriched.csv, clusters.csv, thumbs/ | Pick the right model channel. Review the era split and cluster boundaries at the two pauses. |
| PROMPT 2Recent clusters | [DAYS], [N_CLUSTERS] | A cluster ranking scoped to the recent window. | Size the window to the channel’s cadence. Ask for the n. Compare it against the all-time table. |
| PROMPT 3Ideas | [N_IDEAS] | An idea slate per top cluster. | Edit the audience constraint if it does not fit. Cut to the three to six you would actually upload. |
| PROMPT 4Briefs | [N_TITLES], [TITLE_1…], [TITLE_NUMBERS] | script_brief.md and editor_brief.md per title. | Be honest about first-person claims. Act on the verified-on dates. Hand the two briefs to your team. |
Understand the score
Test an outlier multiplier.
This simple calculator is for intuition. Prompt 1 calculates the centered rolling median from the full ordered dataset and excludes immature videos from ranking calculations.
≥ 3.0×Breakout
1.5–3.0×Overperformer
0.7–1.5×Baseline
< 0.7×Underperformer
Outlier multiplier3.33×Breakout
This video earned 3.33 times the views of its local baseline.
Run discipline
The handoff checklist.
The prompts are only useful when the pauses, evidence labels, sample-size warnings, title cut, and production handoff are respected.
Protect the output
Four mistakes that break this.
Fluent output is not the same as valid analysis. These failures can produce a convincing answer even when the underlying workflow never ran correctly.
Using plain chat
No terminal, no files, no real data—only a description of an analysis that did not execute.
Skipping the pauses
A wrong era split or cluster boundary silently corrupts every decision that follows.
Ignoring the flags
Low-confidence and era-confounded findings are not proven just because the prose sounds certain.
You have the workflow
Now let’s run it on your channel.
We’ll choose the right model channel, read the cluster analysis together, and map the first 30 days of uploads—so you leave with titles you can hand to a writer and editor this week.
Free strategy call · opens Calendly