reels-performance-report

Analyze Instagram Reels and short-form content performance for hook strength, retention/watch time, shares/sends, saves, profile visits, follower conversion, local reach, content cohorts, creative str

SKILL.md

GitHub

Reels Performance Report

Determine which creative patterns deserve more production and which only generate shallow reach.

Workflow

  1. Establish account, exact dates, posting cadence, sample size, and available metrics.
  2. Use exports/connected insights/screenshots/database cohorts. Do not assume every account exposes the same current Instagram metrics.
  3. Normalize content metadata when available:
    • format/franchise
    • topic/service
    • hook type
    • duration
    • real shop footage vs generated/stock
    • voiceover/narrator
    • structurePatternId or equivalent creative cohort
    • CTA, caption, audio, posting time
  4. Analyze distribution and retention:
    • plays/reach where available
    • early retention or skip proxies
    • average watch time and completion proxies
    • replays when supported
  5. Analyze value signals:
    • sends/shares
    • saves
    • comments with intent
    • profile visits
    • follows/follower conversion
    • local/non-follower reach where available
    • website/call/lead outcomes when joinable
  6. Compare cohorts, not isolated viral posts. Require enough observations before declaring a format repeatable.
  7. Identify false wins: high reach with weak retention/value signals, paid/reposted anomalies, or one-off topic spikes that do not generalize.
  8. For Nick's, prioritize useful local automotive education and memorable brand-building that can plausibly increase shop demand, not generic national views alone.
  9. Read references/report-contract.md and follow its output structure.

Rules

  • Do not infer watch-time metrics that Instagram did not provide.
  • Use medians and per-reach/per-play rates when large outliers distort averages.
  • Separate creative effect from topic, distribution, posting time, and sample-size confounds.
  • Do not call a format proven from one or two posts.

Useful triggers

Examples: "run the Reels report", "which formats are winning?", "compare structurePatternId cohorts", "what should we post more of?", "is generated content hurting retention?"