Radical Tempo is free to use, no credit card required.Get started free →
Radical Tempo
Log inSign up
Case study · PrankU × Claude × Radical Tempo

One afternoon. A whole clip campaign, from a few chats with Claude.

PrankU connected Radical Tempo to Claude and asked it to plan the next batch of JoJo Siwa clips. Claude read every past post and retention curve, found what viewers actually watched, cut 7 clips from those moments, scheduled 35 posts, and set up a daily report. Nobody opened an editor.

Get the 5 prompts →PrankU

By Joe White, CEO of Radical Tempo8 min read

The afternoon, in Pacific time

  1. 3:22 PMFirst message. Analysis starts
  2. 3:34 PMStrategy agreed. 7 clips rendering
  3. 3:48 PMAll 7 clips ready
  4. 5:18 PM35 posts scheduled, daily report set
Claude

Let's create new clips for the JoJo Siwa prank. Start by analyzing all the JoJo clips we've ever published, plus the long-form video. Look at the retention graphs, the view counts, what makes a good opening, and where users drop off. Don't build them yet. Do the work first.

Claude
list_videos · 20 foundget_performance_reportget_retention_momentsget_clip_evidence × 8get_playbookget_transcript

I've been through all 20 JoJo videos, 37 posts, and the long-form retention curve. The short version: one scene carries this whole prank.

Illustration built from the real session
  • 6.2Mviews of past JoJo posts analyzed, across 3 platforms
  • 20videos and 37 posts read, with transcripts and retention
  • 7stitched 9:16 clips, each with a reason and a test
  • 35posts scheduled on Instagram, TikTok, YouTube Shorts, Facebook, and X
How it fits together

  1. Claude

    Where you talk. You ask in plain English. Claude plans, explains its reasoning, and asks before anything goes live.

  2. Radical Tempo

    Connected to Claude as an MCP connector. It holds the video library, retention data, and channel playbook, and it cuts, schedules, and measures.

  3. Your channels

    Posts go out on every connected account, and the results come back into Radical Tempo for the next round.

    • Instagram
    • TikTok
    • YouTube Shorts
    • Facebook

Step 0 · ConnectAbout 60 seconds

  1. 1
    Open Claude
    Customize › Connectors
  2. 2
    Add custom connector
  3. 3
    Paste the Radical Tempo link
    https://mcp.radicaltempo.com/mcp
  4. 4
    Connect and sign in
    Your brands and channels appear
Add custom connector
Customize › Connectors › Add custom connector

Add custom connector

Name
Radical Tempo
Remote MCP server URL
https://mcp.radicaltempo.com/mcp
CancelAdd
Illustration of the Claude screen. Labels may differ slightly.
  • list_videos
  • get_retention_moments
  • get_clip_evidence
  • get_playbook
  • get_transcript
  • generate_clips
  • update_post
  • execute_publish_post
  • lookup_post_performance
Some of the tools Claude gets once connected

Chat 1 · Analyze the catalog3:22 PM

Claude

Look at the retention graphs, the view counts, what makes a good opening, and where users drop off. Give me a summary of your findings. Don't build them yet.

Claude
get_performance_report · 37 postsget_clip_evidence · TikTok × 8get_retention_moments · long formget_playbook · PU 2026
Illustration built from the real session

What Claude did

  • Pulled views, likes, shares, and completion for all 37 JoJo posts on Instagram, TikTok, and YouTube.
  • Read the second-by-second retention curve of every TikTok clip, with the words spoken at each drop.
  • Mapped where viewers left and stayed across the 11-minute episode.
  • Checked the channel playbook: what Radical Tempo has learned works on PrankU.

What it found

From PrankU's own data
  • 74%of all views came from one scene: the fake JoJo face-off
  • 0:02Every clip lost the most viewers at 2 seconds. The winners kept the rest
  • 0clips had ever used the funniest line, or the reveal

Same drop. Different hold.

TikTok · share still watching
Share of TikTok viewers still watching each clip at ¼, ½, and ¾ of the way through. Both start at 100%.
Clip¼ through½ through¾ through
Face-off: one scene, one argument64%54%44%
Fraud charges: a montage of set-up23%12%7%

The full episode

YouTube retention
  • Viewers stayed: the traffic stop and charges (2:27–3:58), and the face-off, jail line, and helicopter (7:49–10:02)
  • Viewers left: control-room set-up (0:21–1:03, 5:50–6:32)

The line nobody had clipped

9:17 in the episode

“They like young, pretty girls in jail.”

“I have a girlfriend.”

“Doesn't matter. They don't care.”

Chat 2 · Clip strategy and ideasMinutes later

Claude

Suggest all non-contiguous clips. Stitch together at least two sections so it's punchier. Make them 30–50 seconds, as at 1.1x they get shorter. Strategy first. Radical Tempo will write the hook overlay and post text.

Claude

Seven stitched clips, each joining two to six moments. Each one starts on JoJo already under pressure, avoids the control room, and ends on a payoff, except one cliffhanger we can test against the rest.

Illustration built from the real session

What Claude did

  • Read the transcript of the whole episode to find exact in and out points.
  • Built each clip from moments the retention data said held viewers.
  • Gave every clip a hypothesis and paired clips into tests.
  • Left the hook text and post copy to Radical Tempo, which writes them from what has worked before.

The plan

Sections from the 11:33 episode
#ClipSectionsLengthWhy
1Face-off + the punchline8:23 · 8:39 · 9:1742sThe scene behind 74% of views, finally ending on its punchline
2The whole story in 45 seconds4:04 · 8:26 · 9:17 · 11:0245sAccusation to reveal. Tests full story vs. one scene
3Fake cops? Real fraud6 sections from 1:41 to 8:3036sEscalates every few seconds, ends on a cliffhanger
4Helicopter to LAX5:51 · 9:36 · 10:35 · 11:0245sFear, then relief. Tests payoff vs. cliffhanger
5“Joelle or JoJo?”5 sections, 2:57 to 9:5739sThe name joke 3 times, landed by the fake JoJo
6Mom's in on it4:59 · 5:07 · 5:36 · 11:0233sThe panicked call, when you know she knows
7“I can take her”From the reaction video33sJoJo's side of the story, a new angle

Built-in tests

One variable per pair
  • Punchline endingvs.No punchline
  • Whole storyvs.One scene
  • Cliffhangervs.Payoff
  • Question titlevs.Statement

Chat 3 · Create the clips3:34 – 3:48 PM

Claude

Do them all at 1.1x. Make the framing short. Create away!

Claude
generate_clips · 6 stitchedgenerate_clips · 1 stitchedget_transcript · check

All seven are built, 9:16 at 1.1x. I checked each one against the transcript: every section is there, in order.

Illustration built from the real session

What Radical Tempo did

  • Cut 26 sections from 2 source videos and stitched them into 7 clips.
  • Snapped each cut to the nearest sentence so no line is chopped.
  • Reframed 16:9 to 9:16 and sped up to 1.1x.
  • Drafted a post for every platform, with titles and hook text.
  • Still from PrankU clip 1: Face-off + the punchline142s
    Face-off + the punchline8:23 · 8:39 · 9:17The scene behind 74% of views, finally ending on its punchline
  • Still from PrankU clip 2: The whole story in 45 seconds245s
    The whole story in 45 seconds4:04 · 8:26 · 9:17 · 11:02Accusation to reveal. Tests full story vs. one scene
  • Still from PrankU clip 3: Fake cops? Real fraud336s
    Fake cops? Real fraud6 sections from 1:41 to 8:30Escalates every few seconds, ends on a cliffhanger
  • Still from PrankU clip 4: Helicopter to LAX445s
    Helicopter to LAX5:51 · 9:36 · 10:35 · 11:02Fear, then relief. Tests payoff vs. cliffhanger
  • Still from PrankU clip 5: “Joelle or JoJo?”539s
    “Joelle or JoJo?”5 sections, 2:57 to 9:57The name joke 3 times, landed by the fake JoJo
  • Still from PrankU clip 6: Mom's in on it633s
    Mom's in on it4:59 · 5:07 · 5:36 · 11:02The panicked call, when you know she knows
  • Still from PrankU clip 7: “I can take her”733s
    “I can take her”From the reaction videoJoJo's side of the story, a new angle
  • 7 clips, 33–45s each
    • 9:16
    • 1.1× speed
    • Stitched

Chat 4 · Posts and schedulingSame afternoon

Claude

Do two per day on IG, TT, YTS, FBR, and X. Start them on Saturday after the NLE clips are done. Just one on the last day.

Claude
list_posts · scheduledupdate_post · TikTok settings × 7execute_publish_post × 35

All 35 posts are scheduled. The NLE clips finish Friday, so this starts Saturday at noon.

Change the second set to 3pm PT. 6pm is too late for ET or Europe.

Claude

Done. Clips 4, 5, and 6 now go out at 3pm PT on all five platforms.

Illustration built from the real session

What Claude did

  • Checked the existing calendar so the new posts start after the last clip already queued.
  • Filled in the TikTok settings that were blocking the posts, then scheduled every platform.
  • Moved slots and cleared out-of-date posts when asked, then re-read the calendar to confirm.
  • Rewrote YouTube titles using what the channel playbook has proven works.

The week

Pacific time

  1. Sat, Nov 7

    • 12:00 PMFace-off + the punchline
    • 3:00 PMHelicopter to LAX

  2. Sun, Nov 8

    • 12:00 PMThe whole story in 45 seconds
    • 3:00 PM“Joelle or JoJo?”

  3. Mon, Nov 9

    • 12:00 PMFake cops? Real fraud
    • 3:00 PMMom's in on it

  4. Tue, Nov 10

    • 12:00 PM“I can take her”
    • 3:00 PMNo post
  • Instagram
  • TikTok
  • YouTube Shorts
  • Facebook Reels
  • X

Titles from the playbook

Learned from PrankU's own posts
  • ConfirmedPlain statements beat questionsOn YouTube click-through, measured on the channel
  • ConfirmedShow a face on the coverOn TikTok views, measured on the channel
  • ConfirmedEnd on the payoffOn TikTok watch time, measured on the channel
  • A Fake JoJo Siwa Accuses the Real One of Identity Theft
  • Two Women Both Say They're Joelle Siwa. Only One Is JoJo
  • JoJo Siwa Calls Her Mom for Help. Her Mom Is In on the Prank

Chat 5 · Daily reporting5:18 PM

Claude

Yes please, set up the daily report. Do it from Weds, when the NLE clips go out.

Claude

Set: a scheduled task, daily at 9:41 AM PT. From Wednesday, a fresh Claude session reads every post's numbers each morning and sends you the report, with a push to your phone. It only reads. It never publishes or changes anything.

Illustration built from the real session

What happens each morning

  • Claude checks what went out in the last 24 hours, and flags anything that failed.
  • It pulls views, completion, and retention for every clip on every platform.
  • It scores each JoJo clip against its own test, as a small sample, never as proof.
  • It suggests what to cut next, based only on the numbers.

PU daily clip performance report

Every day · 9:41 AM PT · from Wednesday, November 4
Scheduled
  • What went outEvery post, every platform, and anything that failed
  • What held viewersViews and completion, clip by clip, against its test
  • Where they leftThe steepest drop, and what was on screen or said
  • What to cut nextOne or two ideas for the next batch
  1. Post
  2. Measure
  3. Learn
  4. Cut

Report layout shown. The first report runs Wednesday, November 4. Results to follow.

And the launch post

In the same conversation, Claude turned the analysis into a 12-slide carousel and a 43-second reel for PrankU's own Instagram, in the Radical Tempo house style.

43 sec
  1. Slide 1: We gave Claude our data. 6.2M views. It planned our next 7 clips.
  2. Slide 2: One prompt. Every clip we'd ever posted. Claude analyzed 20 videos, 37 posts, 3 platforms, every retention curve, and every transcript.
  3. Slide 3: One scene got 74% of our views: the JoJo Siwa face-off, 4.6M of 6.2M views.
  4. Slide 4: Everyone leaves at 0:02. The winners keep the rest. TikTok retention chart: 64%, 54%, and 44% versus 23%, 12%, and 7%.
  5. Slide 5: Our funniest line was never clipped. Only 2% were still watching the old jail clip by the end.
  6. Slide 6: 7 clips, every one with a reason. The plan Claude proposed.
  7. Slide 7: Stitched from the moments people stayed for. 3 sections of the 11:33 episode joined into one 42-second clip.
  8. Slide 8: We're not guessing, we're testing. 4 tests: punchline ending, whole story, cliffhanger, and question title.
  9. Slide 9: 35 posts, 5 platforms, one yes. The posting calendar from Saturday, November 7, to Tuesday, November 10.
  10. Slide 10: Every morning, Claude tells us what worked. The daily report: what went out, what held viewers, where they left, and what to cut next.
  11. Slide 11: Connect Radical Tempo to Claude in 60 seconds, using https://mcp.radicaltempo.com/mcp.
  12. Slide 12: Comment CLIPS and we'll send you how we run it with Claude.

12 slides · 1080 × 1350 · scroll sideways

Run it yourself

These are the PrankU prompts with the specifics taken out. Send them in order, read what comes back, and steer it like you would a producer.

  1. 01 · Analyze
    Using Radical Tempo, analyze every clip we've ever published from [episode], plus the long form. Look at the retention graphs, the view counts, what makes a good opening, and where viewers drop off. Give me your findings before we build anything.
  2. 02 · Plan
    Suggest stitched clips that join at least two moments, 30–50 seconds each. Give every clip a reason it should work and say which test it belongs to. Strategy first, don't build yet.
  3. 03 · Build
    Build them all at 1.1x speed, 9:16.
  4. 04 · Schedule
    Schedule two a day on Instagram, TikTok, YouTube Shorts, Facebook Reels, and X, starting [day], at 12pm and 3pm PT. Just one on the last day.
  5. 05 · Report
    Set up a daily report each morning: what went out, what held viewers, where they left, and what we should cut next.
What it took
StepWhat you doWhat Claude and Radical Tempo doClock time
ConnectPaste one linkBring in every brand and channel~1 min
AnalyzeAsk one questionRead 37 posts, every retention curve, and the playbookMinutes
PlanSet the brief, pick the ideasPropose 7 clips with timestamps, reasons, and testsMinutes
CreateSay “go”Cut, stitch, reframe, speed up, and write the posts14 min render
ScheduleSay when and whereBook 35 posts and fix what blocks themMinutes
ReportRead it with coffeeMeasure, compare against each test, suggest next cutsDaily, automatic
Setup guide

Add the Radical Tempo connector to Claude, sign in, and set the tools to Always allow. Then paste the ready-made prompts: check your channels, import your recent videos, analyze, plan the week, create the clips, schedule the posts, and set up a 7am daily review. Then save it as a skill that runs every week.

Read the setup guide →

Frequently asked questions

What did PrankU use Claude for?

PrankU used Claude, connected to Radical Tempo, to plan and launch a batch of JoJo Siwa prank clips. Claude analyzed 20 past videos and 37 posts with 6.2M views, found that one scene drove 74% of views, proposed 7 stitched clips with a reason and a test for each, had Radical Tempo build them, scheduled 35 posts, and set up a daily performance report.

How long did it take?

About 2 hours of one afternoon. The first message went in at 3:22 PM Pacific, the strategy was agreed by 3:34 PM, all 7 clips had rendered by 3:48 PM, and 35 posts were scheduled with a daily report set up by 5:18 PM. Nobody opened a video editor.

What does Radical Tempo add that Claude can't do on its own?

Claude can plan and write, but on its own it can't see your videos, your retention curves, or your results, and it can't post. Radical Tempo gives Claude the video library, second-by-second retention, the channel playbook, and the tools to cut, reframe, schedule, and measure clips.

Does Claude publish anything without approval?

No. Claude proposes the clips and the posting schedule and waits for a yes before building or scheduling. The daily report is read-only: it never publishes or changes anything.

Which platforms were the clips posted to?

Instagram Reels, TikTok, YouTube Shorts, Facebook Reels, and X: 7 clips, 2 a day at 12 PM and 3 PM Pacific, 35 posts in total.

Can I run the same workflow on my channel?

Yes. Add Radical Tempo to Claude as a custom connector with the address https://mcp.radicaltempo.com/mcp, connect your accounts, import your recent videos, and send the 5 prompts on this page in order. The step-by-step setup guide covers each stage.

About Radical Tempo

Radical Tempo is an AI video platform that cuts short clips from long videos, publishes them to YouTube, Instagram, TikTok, Facebook, LinkedIn, and X, measures how each one performs, and uses the results to make the next clips better. It connects to Claude and ChatGPT as an MCP connector.

Notes
  • Channel data: PrankU's Radical Tempo brand, used with permission. Past JoJo Siwa totals cover Instagram, TikTok, and YouTube. TikTok and Instagram metrics as of September 2, 2026, YouTube as of November 2, 2026.
  • Times are Pacific, November 2, 2026. Conversation panels are illustrations built from the real session.
  • Results from the new clips will be added once they have run (first report Wednesday, November 4, 2026).
  • Anthropic: Get started with custom connectors using remote MCP

PrankUPrankU is our own channel. We run it on Radical Tempo every day.

Made with Radical Tempo