AI Skill Report Card
Reverse Engineering Viral Posts
Reverse-Engineering Viral Posts
Quick Start14 / 15
Run these searches on X (adjust keywords/thresholds to the niche):
"AI automation" (affiliate OR agency OR "book a call") min_faves:1000 -filter:replies
"SaaS" ("made $") OR ("MRR") min_faves:500 -filter:replies
"AI agency" ("high ticket" OR "clients") min_retweets:100 -filter:replies
Pick 3-5 accounts that show repeated viral hits (not a one-off fluke), pull their top 10-15 posts, and break each one into: Hook → Emotion → Body format → CTA. Then rewrite each template using your own product's specifics.
Recommendation▾
Add a brief note on handling accounts that delete/edit viral posts or platforms where search operators differ, to cover more edge cases
Workflow15 / 15
Progress:
- Step 1: Run discovery searches, log candidate accounts
- Step 2: Filter accounts for consistency (not one viral fluke)
- Step 3: Pull top posts per account, tag metadata
- Step 4: Deconstruct each post into template components
- Step 5: Map templates to your product/offer
- Step 6: Draft adapted posts, queue for testing
Step 1: Discovery Search
Use X advanced search operators:
min_faves:N/min_retweets:N— set thresholds per niche size (smaller niches: lower threshold, e.g. 300-500; broad niches like "SaaS": 1000+)-filter:replies— exclude reply threads, focus on original posts- Combine niche keyword + proof-of-income or offer language:
"made $","MRR","clients","book a call","high ticket" - Run 3-6 search variants per niche (don't rely on one query)
Step 2: Filter for Consistency
An account qualifies only if it has 3+ posts above your engagement threshold in the last 60-90 days. One viral post = luck or a quote-tweet boost; repeated virality = a system worth copying.
For each candidate account, log:
Account: @handle
Niche/offer: [AI agency / SaaS founder / affiliate]
Follower count: [for context, not gatekeeping]
Viral posts found: [count]
Consistency verdict: PASS / FAIL
Step 3: Pull & Tag Top Posts
For each qualifying account, collect their 10-15 best posts with raw metrics (likes, retweets, replies, post age). Note post type:
- Personal result/case study ("I made $X")
- Contrarian take / myth-busting
- Listicle / framework ("5 ways to...")
- Behind-the-scenes / build-in-public update
- Pain-point callout + solution tease
Step 4: Deconstruct Each Post
Break every post into these components — this is the core reverse-engineering step:
| Component | What to extract |
|---|---|
| Hook (first line) | Exact wording style — number, contrarian claim, bold result, question? |
| Emotion | FOMO, envy, relief, validation, curiosity, anger at status quo? |
| Body format | Short punchy lines, numbered list, before/after, story arc? |
| Proof element | Screenshot, specific number, timeframe, client name/logo? |
| CTA | Soft (follow for more) vs hard (book a call, link in bio)? |
| Topic angle | What specific pain/desire does it target? |
Output as a template with placeholders, e.g.:
HOOK: "I [specific result] in [timeframe] with [method]. Here's exactly how:"
EMOTION: Envy + achievability (result feels big but process feels replicable)
BODY: 5 numbered steps, one sentence each, no fluff
PROOF: Screenshot of dashboard/revenue
CTA: "Want the exact system? Book a free call ��"
Step 5: Map to Your Product
For each template, substitute your own:
- Result/number (real or realistic projection)
- Timeframe
- Method name (your product's unique mechanism)
- Audience pain point (from your actual customer conversations, not guesses)
Step 6: Draft & Queue
Write 2-3 adapted drafts per template. Keep a swipe file mapping: original post URL → template → your adapted draft → status (testing/posted/results).
Recommendation▾
Include a 'bad outcome' example (e.g., a template that flopped) to contrast with the two successful adaptations
Using LLMs in the Pipeline
- Claude (or similar reasoning model): best for Step 4 deconstruction — feed it the raw post text and ask it to isolate hook/emotion/format/CTA into structured output. Also good for Step 5 adaptation since it can reason about tone-matching.
- Separate "search/research" pass: if your LLM has no live X access, paste in manually-collected post text/screenshots transcribed to text rather than asking the model to "search Twitter" — it can't do real-time search unless tool-connected.
- Keep deconstruction and adaptation as separate prompts/turns — don't ask one prompt to both analyze and rewrite; quality drops when you combine them.
Examples17 / 20
Example 1: Input: Post from @AIFounderX — "I replaced 3 VAs with one AI agent and saved $4,200/month. Here's the exact workflow (steal it):" [thread with screenshots, ends with "DM me 'AGENT' for the template"]
Output:
HOOK: "I replaced [old solution] with [new solution] and saved $[amount]/[timeframe]."
EMOTION: Relief (eliminating a known pain) + curiosity (steal it)
BODY FORMAT: Thread, screenshot-led, numbered workflow steps
PROOF: Dollar savings + visual workflow screenshot
CTA: Low-friction DM keyword (not a hard sales pitch)
TOPIC ANGLE: Cost-cutting via automation, framed as "steal my exact system"
ADAPTED DRAFT (for a project-management SaaS):
"I replaced 2 full-time schedulers with one workflow in [Product] and saved $3,100/month.
Here's the exact setup (steal it): 🧵"
Example 2: Input: Post from @SaaSGrowthGuy — "Nobody tells you this about SaaS pricing: raising my price by 40% DOUBLED my conversion rate. Here's why that's not a coincidence:"
Output:
HOOK: Contrarian claim ("Nobody tells you this...")
EMOTION: Surprise + validation-seeking (challenges conventional wisdom)
BODY FORMAT: Short explanation, 3-4 punchy paragraphs, no list
PROOF: Specific % metrics (40% price increase → 2x conversion)
CTA: None explicit — pure authority-building, follow-driven
TOPIC ANGLE: Pricing psychology myth-busting
ADAPTED DRAFT:
"Nobody tells you this about cold outreach tools: raising our price by 30% TRIPLED our close rate.
Here's why that's not a coincidence:"
Recommendation▾
The stray '📞' emoji rendered as garbled characters in the Step 4 example output — clean up encoding issues
Best Practices
- Prioritize consistency over single viral hits — one-offs are often luck, boosted engagement, or a lucky algorithm moment.
- Separate hook analysis from emotion analysis — a hook can be structurally identical across accounts but trigger different emotions depending on audience.
- Always check post age vs engagement velocity — a 3-day-old post with 2000 likes is more signal than a 60-day-old post with 2000 likes.
- Build a swipe file, not a one-time list — revisit source accounts weekly since viral formats decay fast (2-4 weeks typical half-life on X).
- When adapting, change the proof and numbers to your real data — never fabricate specific metrics; use ranges or "projected" framing if real numbers aren't available yet.
- Test adapted drafts in small batches (2-3 at a time) and track which template performs before scaling to more.
Common Pitfalls
- Don't copy wording verbatim — extract the structure, not the sentence, to avoid both plagiarism and a stale/derivative feel.
- Don't chase accounts with one lucky viral post; verify the 3+ post consistency threshold first.
- Don't skip the emotion-tagging step — two posts can look structurally identical but fail if you misjudge the underlying emotional driver.
- Don't mix analysis and rewriting in a single LLM prompt — split into deconstruct-first, adapt-second for better output quality.
- Don't over-fit to one account's voice — pull from 3-5 accounts minimum so your final templates aren't just one person's style.