Setup Guide

Set Claude up to write
viral scripts for you

I went viral on my first reel, after 8 years off social media. This is the exact setup: one free skill, a 15-minute build, and Claude starts writing scripts modelled on what's already performing in your niche.

5 Steps
~15 min One-time setup
3+ Hooks per script
$0 Extra cost

Here's what most people do wrong: they open Claude, type "write me a 30-second script about X," and wonder why it sounds hollow. Claude doesn't know what's landing in your niche right now. It's drawing from training data, not from what got 400k views last Tuesday.

This gives it that context. One research skill, pointed at who's actually winning in your space, and it studies those patterns before it writes a word. My first reel hit 114k views, not because I'm good at content, but because Claude had something real to model from.

What you'll have after this
Inspiration Library
10 to 20 viral creators in your niche, analysed and saved permanently
3 Hook Variations
A/B testable hooks ranked by predicted engagement for each script
Production Brief
Full script with pauses, text overlay cues, filming notes, and CTA
Feedback Loop
System that gets sharper with every post based on your real data
Before you start: You need a paid Claude account (Pro, Max, Team, Enterprise, or API/Console) and Claude Code installed. The free plan doesn't include Claude Code. If you haven't set that up yet, open Terminal and run:

curl -fsSL https://claude.ai/install.sh | bash

That installs the command-line version of Claude that can run skills.
1

Install the research skill

/last30days is a free skill you add to Claude Code. Give it your niche, and it finds who's been getting views: their hook styles, what topics are working, how long their videos run. It pulls from Reddit, X, YouTube, TikTok, Instagram, Hacker News, Polymarket, Bluesky, and more, and only looks at the last 30 days, so you're not modelling content from two years ago.

Open Terminal and run this:

Terminal
git clone https://github.com/mvanhorn/last30days-skill.git ~/.claude/skills/last30days

That's it. Open Claude Code and type /last30days and it should autocomplete. If it does, you're ready.

2

Add API keys (optional, skill works without them)

X/Twitter is free. The skill uses your browser session, no API key needed. The optional keys unlock deeper data from TikTok and Instagram, and let you swap in a different search engine if you want more coverage.

If you only add one, make it ScrapeCreators. It gives you 100 free credits when you sign up (one-time, no expiry, pay-as-you-go after) and pulls structured data from TikTok and Instagram profiles (view counts, hook text, engagement rates) instead of surface-level web results.

What it costs long-term: a single /last30days run uses multiple API calls, so 100 credits won't last forever. Once you burn through them, a 25,000-credit pack is $47, enough for months of normal use. You're not on a subscription; you top up when you want to.

The easy way: let Claude set up the file for you. Paste this into Claude Code. It'll create the config file in the right location, then open it so you can paste your keys in and save. No terminal commands to copy.

Claude Code
Create the /last30days config file at ~/.config/last30days/.env
with empty placeholders for SCRAPECREATORS_API_KEY and BRAVE_API_KEY.
Set its permissions to 600. Then open it in my default editor
so I can paste my keys in.
  1. Grab your keys: ScrapeCreators at scrapecreators.com (100 free credits, one-time) and Brave Search at api.search.brave.com (2k free searches/month).
  2. Paste each key next to the right name in the file Claude just opened, replacing the placeholder text.
  3. Save the file (Cmd + S) and close it. Done.
Skip this for now if you want. The skill works without any keys. Come back and add them once you've run it once and seen what it can do. You'll know immediately if you want more data depth.
3

Build your inspiration library

Before you write anything, you need a reference pool: 10 to 20 creators in your niche who are actually getting views. Not follower counts. Views. Claude is going to study how they open their videos, how fast they move, when they ask for the follow, and use that as its baseline when it writes for you.

Open Claude Code and run your first research query. Be specific:

Claude Code: example prompt
/last30days top viral creators making short-form content about [insert topic] for [insert niche]

What comes back: a list of creators, their rough view averages, the hooks showing up in their top posts, and the topics that keep outperforming. Swap the placeholders for your own topic and niche and run it.

Make it personal. The more specific you are, the better the results:

→
Reference creators you already like
"I like how [insert creator handle] does rapid screen recordings. Find more creators doing this."
→
Filter for disproportionate engagement
"Focus on creators with under 50k followers who are getting outsized views. That's where the freshest patterns are."
→
Match your format
"I'm going to be on camera. I want talking-head creators, not faceless accounts."
Results look thin? Your niche might be too narrow for 30 days of data. Broaden it one step ("AI for dentists" → "AI for small business owners", "Pilates for postpartum" → "women's fitness"), then narrow the scripts later. The research pool needs volume; your content doesn't.

Once you have your list, save it permanently:

Claude Code
Save the research above as a skill at ~/.claude/skills/my-inspiration-library/SKILL.md.

Structure it as:
1. Top creators — handle, platform, avg views, why they work
2. Hook patterns — verbatim examples grouped by type (curiosity, result, mistake, contrarian)
3. Pacing signals — avg length, cut frequency, face-cam vs screen recording
4. Common CTAs

Include a description line that triggers auto-load when I ask for a script, so you reference it without me re-pointing you to it.
From here on, whenever you ask Claude for a script, it already knows your reference pool. You don't have to re-explain it. It's just there.
4

Write your first script

Give Claude a topic. It goes back to your reference pool, pulls the hooks and pacing that are working for those creators, and builds your script around that structure. You're not getting a generic 30-second script. You're getting one built on what's actually been performing.

The exact prompt is in the Ready-to-copy prompts section below. Use "Write your first script." It's tuned to pull from your inspiration library, give you three hook types (curiosity, result, mistake), include overlay timing and filming notes, and flag any claims it can't verify before writing.

What you get back: Claude doesn't just write a script. It gives you a full production brief:

1
3 hook variations
Ranked by predicted engagement, based on what's already performing in your niche
2
Full script with markers
Natural pauses, emphasis cues, and delivery notes built in
3
Text overlay suggestions
What to show on screen and when, timed to the script beats
4
Filming notes
Camera angle, pacing, where to cut
5
CTA
A closing line modelled on what's converting for similar creators right now
3 seconds
The first three seconds of your video decide whether Instagram keeps distributing it. If people scroll past, the algorithm stops pushing it. That's the whole game, which is why getting three hook options and actually testing them isn't optional. It's how you find out what your specific audience wants to watch.
5

Film, post, and feed the loop

Film Hook A exactly as written. Resist the urge to paraphrase. First few posts, stick to the script so you know what you're actually testing.

  1. After 24 hours, check your 3-second retention. Under 40% is a rule-of-thumb floor. It means most people aren't making it through the hook. Try Hook B on the next post.
  2. Give Claude your numbers: views, retention rate, and anything that felt awkward to say out loud. It uses that to adjust what it writes next.
  3. Re-run /last30days once a month. What's going viral in January isn't what's going viral in April. Let the research go stale and the scripts follow.
The first post probably won't land perfectly. That's expected. The setup gets better the more data you give it, so post regularly, check your numbers, and keep feeding Claude what you're seeing. It adjusts. That's the whole point.

Ready-to-copy prompts

Once you're set up, here are three prompts to run first. Copy them as-is and swap in your niche or topic.

Find your top competitors
Claude Code
/last30days find 10–15 short-form video creators making content
about [insert topic] for [insert niche] on Reels, TikTok, or Shorts.
Skip anyone over 500k followers — I want the hungry middle where
fresh patterns emerge, not dormant celebrity accounts.

For each creator: handle, platform, avg views on last 10 posts,
3 best hooks verbatim, and top recurring topic angles.

End with a "patterns that repeat across creators" section —
opening words, video length, face-cam vs screen recording, CTA style.
Write your first script
Claude Code
Using my inspiration library, write a 30-second Reel script
about [your topic] for [your niche].

- 3 hook variations: one curiosity, one result-first, one mistake/contrarian
- First-person voice, no corporate filler
- On-screen text overlays with timing
- Filming notes: camera angle, cuts, pacing
- CTA modeled on what's converting in my library right now

If you need to make a specific factual claim — a tool name,
a number, a source — flag it before writing so I can verify.
Don't invent details to make the script land.
Improve based on performance data
Claude Code
My last Reel:
- Views: [X]
- 3-second retention: [Y%]
- Topic: [topic]
- Spoken hook: [paste]
- On-screen text: [paste]
- Caption first line: [paste]

Find the 3 hooks in my inspiration library closest to this topic.
Tell me what specifically differs — word count, opening word,
visual pattern, CTA placement.

Then rewrite my hook 3 ways, each explicitly modeled on one of
those winning library hooks. Name which library hook each rewrite
is based on.
The honest version

I'm not a content creator. I'm a business operator who started posting about AI because I was using it every day and couldn't find anyone giving straight answers about what actually works. I hadn't posted to social media in eight years. No content strategy, no creator background, just this setup and a willingness to test.

The 114k views on my first reel wasn't luck. It was Claude studying what was already going viral in the niche and writing a script that followed those patterns, but on a topic I'd actually tested myself. The AI did the pattern work. I did the "is this actually true" check.

That's the filter I'd recommend applying to everything this setup produces: use the patterns, but only post things you've personally tested or can honestly stand behind.