TikTok Algorithm & Faceless Clip Strategy

Trigger

Activate when user says: "tiktok algorithm", "study tiktok", "what's working on tiktok", "improve our clips", "tiktok trends", "tiktok report", "clip strategy", "faceless tiktok", "headless tiktok", or similar.

Purpose

Produce actionable TikTok intelligence for faceless/headless clip campaigns. Every output should answer: "What do we change THIS WEEK to get more views?"

Data File

Store and read cumulative learnings at:

/Users/tori/Documents/Repos/CHI/automations/overnight-pipeline/data/tiktok-learnings.json

Schema:

{
  "last_updated": "YYYY-MM-DD",
  "weekly_reports": [
    {
      "week_of": "YYYY-MM-DD",
      "algorithm_changes": [],
      "top_faceless_clips": [],
      "hooks_tested": [],
      "hooks_results": {},
      "sounds_trending": [],
      "ab_test": { "hypothesis": "", "result": "", "verdict": "" },
      "next_week_test": ""
    }
  ],
  "hook_library": {
    "question": [],
    "shock_curiosity": [],
    "countdown_list": [],
    "before_after": [],
    "controversy": [],
    "pov": [],
    "storytime": []
  },
  "sound_library": {
    "trending_originals": [],
    "evergreen_beds": [],
    "avoid": []
  },
  "account_intel": [],
  "cumulative_insights": []
}

If the file does not exist, create it with empty arrays. Always read it first, append new data, and write back.

Step 1: Algorithm Intelligence Gathering

Run these searches every time this skill is triggered:

WebSearch: "TikTok algorithm update {current_month} {current_year}"
WebSearch: "TikTok algorithm changes {current_year} what's new"
WebSearch: "TikTok for you page ranking factors {current_year}"
WebSearch: "TikTok video length optimal {current_year}"
WebSearch: "TikTok posting frequency best practices {current_year}"
WebSearch: "TikTok B2B content strategy {current_year}"

Extract and record:

• Video length sweet spots (current algo preference: 1-3 min rewarded, but track shifts)
• Posting frequency recommendations
• Hashtag strategy changes (broad vs niche, count, placement)
• FYP ranking signals (watch time, shares, saves, comments — which matters most NOW)
• Any new features being pushed (photo carousels, series, search SEO)
• Creator Marketplace changes affecting reach
• Shadowban triggers or content suppression patterns

2026 Baseline (update as things change)

• 1-3 minute videos getting algorithmic boost over shorts
• Authenticity signals > production polish
• B2B engagement up 340% YoY
• Search SEO on TikTok becoming a discovery channel
• Series/playlist completion rates boost subsequent videos
• Saves and shares weighted heavier than likes
• Comment reply videos get distribution boost

Step 2: Faceless Account Research

Run these searches:

WebSearch: "top faceless TikTok accounts {current_year}"
WebSearch: "faceless TikTok channel examples viral"
WebSearch: "headless TikTok accounts most followers {current_year}"
WebSearch: "AI narrated TikTok accounts growing fast"
WebSearch: "text overlay TikTok accounts viral {current_year}"
WebSearch: "faceless TikTok niche ideas that work"

Categories to Study

1. Motivation/mindset clips - text over stock footage, AI voiceover
2. Product showcase - hands-only demos, unboxing, ASMR product
3. Compilation/curation - "satisfying" compilations, fails, reactions
4. AI-narrated content - Reddit stories, history, true crime, explainers
5. Text-overlay stories - greenscreen text, fake texts, confessions
6. Gameplay channels - Minecraft parkour + story, Subway Surfers + narration
7. Educational/how-to - screen recordings, tutorials, "did you know"
8. Nature/animals - wildlife footage with narration or captions

Per-Account Analysis Framework

For each account found, record:

What Makes a Faceless Clip STOP the Scroll

Analyze and categorize the scroll-stopping mechanics:

• Visual disruption: unexpected color, movement, contrast
• Text hook: bold text with curiosity gap in first frame
• Audio hook: jarring sound, whisper, unexpected voice
• Pattern interrupt: something visually "wrong" that demands attention
• Information gap: partial reveal that forces watch-through
• Controversy bait: statement that triggers "wait, what?" reaction

Step 3: Hook Library Management

Hook Categories & Templates

Question Hooks

• "Did you know [surprising fact]?"
• "Why does nobody talk about [thing]?"
• "What happens when you [action]?"
• "Have you ever wondered why [observation]?"

Shock/Curiosity Hooks

• "This changed everything about [topic]"
• "I can't believe [thing] actually works"
• "Nobody was supposed to see this"
• "They don't want you to know [thing]"

Countdown/List Hooks

• "3 things you're doing wrong with [topic]"
• "5 [things] that will blow your mind"
• "The #1 reason [thing] fails"
• "Ranking [things] from worst to best"

Before/After Hooks

• Side-by-side visual transformation
• "Watch what happens when..."
• Time-lapse progression
• "Day 1 vs Day 30"

Controversy Hooks

• "Everyone's wrong about [topic]"
• "Unpopular opinion: [take]"
• "This is why [common thing] is actually bad"
• "[Authority figure] lied about [thing]"

POV Hooks

• "POV: you just discovered [thing]"
• "POV: you're the only one who [action]"

Storytime Hooks

• "So this just happened..."
• "Story time: [teaser]"
• "The craziest thing happened at [place]"

Hook Performance Tracking

When reviewing our own clip performance, tag each clip with:

• Hook type used
• First-frame text (exact words)
• 1-second retention rate (if available)
• 3-second retention rate (if available)
• Full watch-through rate
• Verdict: KEEP / MODIFY / DROP

Step 4: Sound & Music Strategy

Run these searches:

WebSearch: "TikTok trending sounds this week {current_year}"
WebSearch: "TikTok royalty free music beds faceless content"
WebSearch: "TikTok original audio strategy grow account {current_year}"
WebSearch: "best TikTok sounds for faceless videos"

Sound Decision Matrix

FTC Disclosure for Sponsored Clips: Sponsored clips must disclose #ad — trending sounds still work but the disclosure must be prominent per FTC rules. The #ad overlay and caption disclosure are non-negotiable on any sponsored content regardless of sound choice.

Sound Library Maintenance

Track three lists:

1. Trending originals - sounds blowing up right now (shelf life: 1-2 weeks)
2. Evergreen beds - royalty-free music that consistently works (cinematic, lo-fi, epic)
3. Avoid list - sounds that are overused, copyrighted, or suppressed

Step 5: Weekly Algorithm Report

This report feeds into the Monday morning review. If kill criteria are approaching (see tiktok-clips-biz skill), flag it prominently.

Report Template

Generate this report and post it. Structure:

TIKTOK ALGORITHM REPORT - Week of [DATE]

ALGORITHM CHANGES
- [List any confirmed or suspected changes]
- [New features being pushed]
- [Shifts in content preference]

TOP FACELESS CLIPS THIS WEEK
1. [Account] - [Description] - [Why it worked] - [View count if known]
2. [Account] - [Description] - [Why it worked]
3. [Account] - [Description] - [Why it worked]

TRENDING SOUNDS FOR FACELESS
- [Sound 1] - [Why it works for us]
- [Sound 2]
- [Sound 3]

HOOKS THAT HIT
- [Hook pattern] - [Example] - [Estimated performance]

OUR CLIPS THIS WEEK
- [Clip 1]: [Hook used] / [Views] / [Retention] / [Verdict]
- [Clip 2]: [Hook used] / [Views] / [Retention] / [Verdict]
(If no clips posted, note "No clips posted this week")

LAST WEEK'S TEST RESULT
- Hypothesis: [What we tested]
- Result: [What happened]
- Verdict: [Keep / modify / drop]

THIS WEEK'S TEST
- Test: [What to try]
- Why: [Based on what intelligence]
- How to measure: [Success metric]

STRATEGIC TAKEAWAY
[One paragraph: the single most important thing to act on this week]

KILL CRITERIA CHECK
[If any tiktok-clips-biz kill criteria are approaching threshold, flag them here with current numbers vs thresholds. If all clear, note "All metrics within acceptable range."]

Where to Post

• Read the report back to the user in the conversation
• If user says "post it" or "send to slack", post to the relevant Slack channel

Step 6: Continuous Improvement Loop

Every Time This Skill Runs:

1. Read tiktok-learnings.json
2. Compare this week's data to last week's
3. Identify:
   • Which hook types are trending up vs down
   • Which video lengths are getting rewarded
   • Which sounds are gaining vs losing momentum
   • What new faceless accounts are growing fastest
4. Propose ONE specific test for next week:
   • New hook formula
   • New video length
   • New posting time
   • New sound strategy
   • New caption format
5. Write the updated data back to tiktok-learnings.json

A/B Test Tracking

Each test gets logged:

{
  "test_id": "YYYY-MM-DD-{slug}",
  "hypothesis": "Using countdown hooks will increase 3s retention by 20%",
  "variable": "hook_type",
  "control": "question hooks",
  "variant": "countdown hooks",
  "duration": "1 week",
  "clips_control": 3,
  "clips_variant": 3,
  "result_control": { "avg_views": 0, "avg_retention_3s": 0 },
  "result_variant": { "avg_views": 0, "avg_retention_3s": 0 },
  "verdict": "WINNER: variant / control / INCONCLUSIVE",
  "next_action": "Scale countdown hooks across all content"
}

Cumulative Insights

After each run, append any confirmed learnings to the cumulative_insights array:

{
  "date": "YYYY-MM-DD",
  "insight": "Countdown hooks outperform question hooks by 35% on 3s retention",
  "confidence": "high",
  "based_on": "4 weeks of A/B testing, 24 clips"
}

These insights compound over time and inform all future strategy decisions.

Execution Rules

1. ORGANIC ONLY — never recommend buying views, likes, or engagement. The only paid amplification allowed is TikTok Spark Ads on organically successful clips (>1K views in 24h).
2. Always run the WebSearch queries first. Do not rely on stale knowledge.
3. Produce concrete recommendations, not observations. "Post 1-3 min videos" is an observation. "This week, test a 90-second motivation clip using a countdown hook with [specific trending sound] and post it Tuesday at 11am CST" is a recommendation.
4. When analyzing faceless accounts, focus on REPLICABLE patterns. We cannot replicate "went viral randomly." We CAN replicate "uses bold yellow text on black background with a 2-word hook in frame 1."
5. Track everything in the JSON file. The value of this skill compounds over time.
6. If the user asks for a quick answer (e.g., "what length should our clips be?"), still run the searches to verify the answer is current, then give a direct answer with the source.
7. Never recommend strategies that require showing a face, hiring talent, or doing live content. We are faceless-only.
8. Include sources for all claims. Link to articles, accounts, or data points.