A Friend — Distribution

references__shortform-virality

14 min read

Synced from references/shortform-virality.md in content-extraction skill on 2026-05-18. Edit upstream in the skill; this file is overwritten on next sync.

Short-Form Virality Canon: Founder-Monologue Clip Extraction

Programmatic knowledge base for identifying, scoring, and extracting ship-worthy clips from founder transcripts. Built on 2025–2026 authoritative research, sharpened by real-review feedback (see refinement-log.md). Each rule is testable; each signal a script can apply.

Substance over Hook-Aggression — the meta rule

Before any hook-archetype matching, before any Tier-1/2/3 scoring, this rule wins:

A clip exists to deliver substance. The hook is the doorway, not the room.

Three sub-rules, all encoded from real review failures (see refinement-log.md 2026-05-01):

Rule M1 — Hook without setup is a dead hook

A specific-number hook (“Approximately 14 hours.”) sitting at clip-position-0 with no preceding context is not a hook — it’s a riddle. The viewer has no scaffold to receive the number. They swipe.

A hook archetype must be received against context the viewer can ground. Either:

  • The archetype is self-supplying (a contrarian categorical reframe like “We are not merely building toys” carries its own context). OK to lead with.
  • Or the hook needs a prior segment supplying the topic. Number-only hooks are the most setup-dependent — never lead with a bare number unless an immediate visual overlay supplies the unit AND the topic.

When in doubt: add the setup segment. A 5-second context-prefix is cheaper than a swipe.

Rule M2 — Substance over runtime

Aggressive reduction (2 segments, 17 seconds, hook-and-close) optimizes for the swipe boundary and sacrifices what makes founder content actually pull: depth, layered reasoning, the texture of a real thought. A 60-second clip with substance pulls harder than a 17-second clip without.

Default target_duration_seconds is 30–90 seconds, not 15–60. Reduce only when reduction sharpens substance — never when it just shortens runtime. YouTube Shorts caps at 60s; TikTok and Reels accept 60s+. Use the headroom.

Rule M3 — Compose, don’t just cut

A clip is a story arc — setup → hook → body → reframe → close — not “find the hook, find the close, concat.” The cut plan’s job is composition, not selection.

Move sentences across distance in the source if it builds a tighter arc. Pull a Q&A answer forward of its question. Glue a reflection at minute 5 to a vision-statement at minute 12 if they’re thematically the same beat. Plan the arc, then pick segments to fill it. Not the other way around.

Valid arc shapes (pick the one that fits the substance, not a default template):

Arc shapeWhen to use
Cold-openHook self-supplies context (contrarian, named-product); body deepens; close promises
Setup-then-hookHook needs context; build curiosity in setup; deliver hook mid-clip
Question-then-answerQ&A material; pose then resolve in 2-3 segments
Number-then-contextNumber hook with overlay support; immediately ground in topic
Vision-then-vulnerability-then-promiseLong-form (60-90s); emotional pull; rare but powerful
Vulnerability-then-recoveryTier-1 Pattern A; hardest pull; deepest stake

Hook Archetypes

Six hook types that work reliably in 2026 short-form, with concrete failure modes a script can detect.

1. Product / Outcome Showcase

Definition: Lead with the finished result, no setup. “You lead with the thing — the product in action, the before-and-after.”

Pattern: Segment opens with a concrete outcome, demo result, or “here’s what we shipped” statement. Zero preamble before the reveal.

Performance: Highest-performing hook type across 34,635 analyzed clips (avg 6,037 views).

Scriptable signal: First sentence contains a tangible result + present/near-past tense + visual proof potential.

Fails when: Setup precedes outcome, or “outcome” is abstract (“I learned something valuable”). Requires tangible proof in first ~2 seconds.

2. Expert Explainer Setup

Definition: Establish authority → tease insight → promise payoff.

Pattern: “I’ve worked in [role] for 12 years. Here’s what nobody tells you about [topic].” Specific number + credential + contrarian claim.

Scriptable signal: Opening contains: role/credential keyword (built / shipped / years / founded) AND a contrarian framing (“nobody tells you,” “the real reason,” “what actually matters”) within first 3 seconds.

Fails when: Credential vague (“I work in tech”) or claim lacks punch. Weak contrarian = audience assumes generic advice.

3. Contrarian Take

Definition: Reject widely-held belief in opening sentence. Cognitive dissonance — brain can’t leave contradictions unresolved.

Pattern: “Everyone says [common advice]. It’s wrong.” Or “Stop doing [popular practice].”

Performance: 156% more For-You-page appearances when framed as a pattern interrupt.

Scriptable signal: Opening contains negation frame (don’t / stop / wrong / myth) + specific belief or action. Must deliver substance within 5 seconds or credibility collapses.

Fails when: False contrarian without supporting evidence. Once trained that a creator makes empty contrarian claims, audiences ignore — and the failure is cumulative.

4. Specific Number Hook

Definition: Odd, specific numbers signal authenticity. “23 pounds in 67 days” beats generic claims.

Pattern: “I wrote 3,247 cold emails. Here are the 4 subject lines that worked.” Or “We grew from 0 to $1.2M ARR in 14 months because of [one insight].”

Scriptable signal: Transcript contains a concrete, non-round number paired with a claim or result. Examples: revenue, user counts, days elapsed, dollars spent, components iterated, errors fixed.

Fails when: Number lacks context, or isn’t paired with insight. Random statistics feel like bragging; paired statistics feel like proof.

5. Curiosity Gap Hook

Definition: Incomplete information that creates a desire to discover the answer.

Pattern: “I tested this one weird trick…” “The real reason most founders fail isn’t what you think…” Or end opening sentence with an unanswered question.

Performance: 234% more completion, 178% higher shares vs direct-statement openings.

Scriptable signal: Opening sentence leaves a void — open question, teased fact, or promised reveal not yet delivered. Must deliver answer by 8s or retention collapses.

Fails when: Gap too wide (payoff feels distant), payoff underwhelming, or viewer guesses the answer in first 3s.

6. Vulnerability Beat (Personal Stakes)

Definition: Founder admits struggle, failure, or counterintuitive experience. Creates emotional permission for honesty.

Pattern: “I almost shut down the company.” “I’ve made this mistake 4 times.” “I was terrified to [X] but here’s what happened.”

Performance: 89% retention improvement when vulnerability is specific and time-stamped (“in 2021,” “after our Series A”).

Scriptable signal: First-person past tense about failure, near-miss, fear, or surprise. Phrases: “I almost,” “I remember thinking,” “I was wrong about,” “I was scared,” “we almost didn’t.”

Fails when: Vulnerability without lesson. Generic struggle (“starting a company was hard”) without specific detail. Or perceived as fishing for sympathy without action.

Retention Mechanics: Second-by-Second Rules

What must happen across the arc for TikTok / Reels / Shorts to hold attention. Algorithmic thresholds, not artistic goals.

0–3s — The Swipe Boundary

  • Target retention: 80% min (YouTube Shorts benchmark). TikTok algorithm makes its first distribution decision at 1.5s.
  • Required: Hook communicates in 1.5s (text + visual + spoken word aligned). Pattern interrupt OR immediate value promise.
  • Signal: Does the segment open with one of the 6 hook archetypes above?
  • Failure: Slow intro / preamble / context-setting before hook. Every second of pre-hook ≈ 15% retention loss.

3–8s — Hook Delivery & First Context

  • Target retention: 70%+ (maintenance phase).
  • Required: Hook promise clarified (question answered, claim explained, direction revealed). Transition feels inevitable, not forced.
  • Signal: By 8s, viewer should know the clip’s thesis. Ambiguity here kills retention.
  • Failure: Hook delivered, then silence/delay. Pacing collapse.

8–15s — Proof or Story Escalation

  • Target retention: 60%+ (midpoint).
  • Required: Specific evidence (number, screenshot, testimonial quote, before/after) OR narrative escalation (stakes rise, contradiction deepens). Pacing tightens, not plateaus.
  • Signal: Concrete fact, metric, quote, or plot turn — not filler.
  • Failure: Repetition of hook without new info. Talking head without visual support. Common retention cliff at 10–12s.

15–30s — Value Delivery / Climax

  • Target retention: 55%+ (most viewers leave before completion; this is the acceptable-loss phase).
  • Required: Core lesson, insight, or emotional beat delivered. Setup for ending: resolution or CTA.
  • Signal: The “why” for watching is satisfied. Viewers now have a reason to complete or share.
  • Failure: Anticlimactic payoff, generic advice, trailing off.

30–60s — Conclusion / Loop / Finish Strong

  • Target retention: 70%+ average percentage viewed (short-form-overall benchmark; completion is secondary to APV).
  • Required: Rewatch hook (loopability) OR clear CTA OR micro-ending (closed-loop, punchline, emotional landing).
  • Signal: Last 3s feel intentional.
  • Failure: Rambling into silence, no closure, unclear takeaway.

Founder-Content Levers: What Makes Startup Content Pull

Founder monologues differ from lifestyle / narrative / comedy content. These specific angles make founders go viral in 2026.

L1 — Named Numbers + Founder Attribution

Specificity signals authenticity. “We grew a lot” reads as marketing; “we hit $1.2M ARR because of the 3-email cold outreach sequence” reads as earned data.

Signal: Precise revenue / user / growth / ROI figure + paired with a specific founder action (“we did X, got Y”) + ideally non-round number (23 vs 25, 1.2M vs 1M).

Fails: Numbers without context, or generic attribution.

L2 — Technical Detail as Credibility Proof

When a founder explains implementation detail (API choice, algorithm, schema pattern), it signals they actually built something. AI-startup founders especially: “we prompt-chained X to Y” reads as expertise.

Signal: Brief technical term, tool name, or build step that only an actual builder would know. Examples: “we switched from GPT-4 to Claude for latency reasons,” “the bottleneck was inference cost, so we quantized the model.”

Fails: Buzzword soup without specificity. Technical detail that doesn’t connect to business outcome.

L3 — Building-in-Public Moment

Live decisions, launches, milestones trigger FOMO + credibility simultaneously.

Signal: Present tense or near-past (“we just shipped,” “we’re about to,” “as of yesterday”) + concrete artifact (new feature, user testimonial, metrics movement).

Fails: Vague future tense (“we’re planning to”) without shipped proof.

L4 — Vulnerability + Recovery Pattern

Admitting mistakes or near-deaths (and what was learned) earns trust.

Signal: Specific failure or crisis with timeframe (“in month 4,” “before we pivoted”) + decision in response + outcome.

Fails: Humble-brag without real stakes. “We learned a lot from our mistakes.”

L5 — Customer Quote or Social Proof (With Names)

A testimonial with a specific use case or first-name reads as proof.

Signal: Direct quote or paraphrase from a named customer, partner, or user, with a specific use case.

Fails: Attributed quotes with no context. Generic praise.

L6 — Contrarian Bet Explained

Founders explaining why they chose the unpopular path signal conviction.

Signal: Bold claim (“we decided to go vertical instead of horizontal”) + reasoning in 2–3 sentences. High-velocity, not lengthy.

Fails: Contrarian without reasoning.

Platform Deltas: Same Clip, Three Platforms

SignalTikTokInstagram ReelsYouTube Shorts
Algorithm priorityWatch-through in first 1.5s; completion rateSaves & shares > likes; aesthetic consistencyWatch time + search keywords (title/description)
Native styleRaw, trend-aware, quick cutsPolished, aesthetic, visual consistencyEducational, clear metadata, “how-to” friendly
Optimal length15–60s (avg swipe at 41s)15–30s for highest completion30–60s; keyword in title matters more
Hook requirementCritical; lands in 1–2sImportant but secondary to visual polishImportant but searchable title compensates
AudioTrending sounds boost discovery; original audio now performs 156% better than beforeCan use trending audio, but watermarked TikTok reposts get punishedTitle/thumbnail > audio
CaptionsIn-app overlays; on-screen text +30% retentionVisual-overlay consistency; on-screen text expectedHighly important; many watch muted
Repurposing costHighest; TikTok-watermarked Reels get punishedHigh; IG detects and deprioritizes TikTok watermarksLow; can upload clean vertical video
Failure modeHook misses first 2s → cold-start, no distributionProduction-value + weak hook = siloed to followersPoor metadata (generic title) → no search traffic

Per-platform adapt:

  • TikTok version: Use trending sound when relevant, max contrast + quick cuts in first 3s.
  • Reels version: Polish visual consistency, strip watermarks, lean save/share CTAs.
  • Shorts version: Optimize title + description with keywords (founder/builder/startup topic), lean educational (“How we X’d our Y by Z%”).

Brand-Voice Tension: When to Bend, When to Hold

Founder monologues carry brand voice — but platforms reward platform-native voice. Tradeoff: earned authenticity vs algorithm score.

Hold brand voice when:

  • Founder has accumulated audience trust (100K+ followers, >1% engagement across 20+ clips).
  • Contrarian position already established.
  • Niche audience expects long-form depth (B2B SaaS, deep-tech, academic — more patient).

Bend toward platform when:

  • Clip is first/early in a series (audience doesn’t yet know the founder’s voice).
  • Goal is discovery/reach over brand tone.
  • Founder’s brand is itself about being authentic-to-platform.

Hybrid (recommended for A Friend / Daniel):

  • Hook (0–3s): Use platform-native archetypal hook (pattern interrupt, curiosity gap, specific number).
  • Body (3–30s): Founder’s voice — authentic vocabulary, pacing, examples. Daniel: keep German/English code-switching here.
  • Ending (30–60s): Reinforce brand position or learning, not a virality hack.

For A Friend specifically: brand voice rules from _brand/voice-preservation.md are non-negotiable in body + ending. Hook can be slightly bent toward platform-native if it makes the clip land — but the bend must not violate banned vocabulary or break the two-beat cadence.

Segment-Selection Heuristics: Scriptable “Ship-Worthy” Signals

Given a transcript, these signals identify segments to clip.

Tier 1 — High Confidence (script SHOULD flag)

Combinations nearly guaranteed to work.

Pattern A — Vulnerability + Specific Number + Outcome

  • Contains: “I was [scared/wrong/nearly]” + timeframe/dollar/count + “so we [did X]” + result.
  • Example: “I almost shut the company down in month 4 when we were burning 500K ARR.”
  • Recommendation: SHIP.

Pattern B — Bold Contrarian + Expert Credential + Specific Example

  • Contains: “Everyone says [belief]” / “common advice is wrong” + credential + concrete example.
  • Example: “Everyone says you need VC funding to scale. I’ve bootstrapped 4 companies. We grew this one to $2M ARR without a single dollar of venture capital.”
  • Recommendation: SHIP.

Pattern C — Shipped Artifact + Named Metric

  • Contains: “We just launched [X]” / “as of yesterday” + early-user metric or % delta.
  • Example: “We shipped the new vector search this morning. Early beta users say it reduced their inference latency by 60%.”
  • Recommendation: SHIP.

Pattern D — Technical Detail + Business Outcome

  • Contains: tool/architecture/API choice + reason + outcome.
  • Example: “We switched from GPT-4 to Claude 3.5 Sonnet because we needed faster outputs for our chat interface. Turns out our users prefer the response style anyway.”
  • Recommendation: SHIP.

Tier 2 — Medium Confidence (script SHOULD CONSIDER)

One or two signals; likely to work but not guaranteed.

Pattern E — Specific Number Alone: single claim with precise non-round number. Medium-high — needs narrative arc or contrast to be great.

Pattern F — Contrarian Claim Alone: bold opening, proof in second sentence. Medium — risky if explanation fails.

Pattern G — Customer Moment: quoted/described customer experience. Medium — engaging but generic without name + use case + revenue impact.

Tier 3 — Lower Confidence (script MAY skip or mark CANDIDATE)

Pattern H — Lesson/Advice Standalone: pure teaching without personal story or proof. Low-medium — depends on existing credibility.

Pattern I — Vague Aspiration / Future Tense: “we’re planning to,” “our goal is” without shipped proof. Low.

Pattern J — Generic Gratitude / Filler: thank-you, conversational filler. Skip.

Scoring Rubric

Signals presentRecommendation
3+ Tier-1 signalsSHIP — high confidence
2 Tier-1 signalsSHIP — strong clip
1 Tier-1 + 2+ Tier-2SHIP — good clip
2+ Tier-2 onlyCONSIDER — needs review
1 Tier-2 onlyHOLD — re-evaluate in context
Tier-3 onlySKIP unless niche-specific

Mapping to hook_strength (1–5) in cut-plan.json

  • 5 = Tier-1 Pattern A/B/C/D AND opens with a clear hook archetype (1.5s rule met).
  • 4 = Tier-1 pattern present but hook is in the body, requires reorder to lead.
  • 3 = Tier-2 pattern with narrative arc; needs supporting B-roll to hold retention.
  • 2 = Tier-2 alone, no clear hook; CONSIDER but expect lower pull.
  • 1 = Tier-3; skip unless contextual reason to ship.

Refinement Backlog

Topics for deeper research as the skill matures:

  1. Niche-specific hook variance — does founder/builder content respond differently to curiosity-gap vs contrarian vs vulnerability hooks across niches (fintech / climate / deep-tech)? Early signal: vulnerability beats outperform in deep-tech. Needs 5+ niche validation.
  2. Cross-platform serialization — when a founder posts the same clip TikTok → Reels → Shorts, does sequence order matter for audience journey? (Reels performs better if they’ve seen TikTok first?) Compounding effect unexplored.
  3. AI-startup content premium — does AI/LLM/agentic-systems content have higher retention floors? Sample data shows 8.2% engagement vs 2.8% TikTok baseline. Needs control for founder-credibility confound.
  4. A Friend brand-voice eval — once 20+ clips are shipped from this skill, build a feedback loop scoring which clips Daniel approved unchanged vs which he edited. The edit pattern reveals where the skill’s brand-voice match is weakest.
  5. German-language short-form deltas — most virality data is English-centric. Daniel’s clips will be German/English mixed. Open question: do the same hook archetypes pull on German-speaking TikTok / Reels audiences? Needs separate research pass.

Sources

Graph View

Backlinks