Video summary

How i automate 30 ai videos a day that print $100k/month (full guide)

Main summary

Key takeaways

Business

Business model & goal (Organic AI drop shipping)

  • Strategy: Create AI “organic” videos (no paid ads) to go viral on social platforms, then funnel viewers to a website to drive orders.
  • Core operational automation: Use Claude + Claude MCP (HixField) to rapidly generate and iterate creatives while maintaining quality and reducing creative fatigue.

Framework / process (“6-step” playbook)

1) Connect MCP

In Claude settings:

  1. Go to General → Connectors
  2. Add custom connector
  3. Use this connector link: https://mcp.hixfield.ai/mcp

2) Give Claude context (build a reference library)

  • Instead of relying only on generic prompts, collect and feed Claude an “ethos” plus clip frames.
  • Inputs include (as applicable):
    • Hook concepts (e.g., in-store shots, 3-2-1 reveals)
    • CTA clips
    • Context clips
    • Audio/music style, text style
    • Character/model references where relevant

Actionable example:

  • Screenshot 2–3 frames from an in-store hook (e.g., person enters store → grabs product → puts it in cart), and submit them as context.

Result:

  • Claude “logs” the concept and later reuses it as structured building blocks, reducing the need to regenerate from scratch when adding new variations.

3) Test ample creatives (get data → find outliers)

  • Post enough videos to establish a baseline and identify winners.
  • Rule of thumb: test ~7–20 videos, typically expecting outliers around 7–10 posted.

Example metric:

  • If most videos average ~800–1,500 views (avg ~1,100) and one hits ~2,500 views, that’s roughly a 2x outlier.

Variables to vary across tests:

  • Hook
  • Text
  • Music/audio
  • Concept/angle
  • Product presentation
  • Model/character
  • Controversy
  • Marketing angle (e.g., niche-specific phrasing; “walk don’t run”; in-store vs gifting, etc.)

4) Analyze posts → find outliers

  • Identify what likely drove the outlier (hook vs music vs product novelty vs controversy).
  • Emphasizes critical variable isolation: outliers can stem from one variable or from interactions (e.g., text + music).

5) Analyze outliers → lock “working variables”

  • Break the outlier into what to keep vs what to change.
  • Example (same concept, different model):
    • Model 1: ~7,000 views
    • Model 2: ~30,000 views
  • Inference: model is likely the main variable → double down on model 2 and test variations around it.

6) Automate + automated split testing (reuse assets, remix variables)

  • Automation isn’t “press a button for viral guaranteed results”; editing and manual tweaks (e.g., text/audio adjustments) still occur.
  • Benefits claimed:
    • Avoid reusing identical clips → reduces creative fatigue
    • Reduce risk of running out of credits (including risks like shadow ban / stopping due to payment/top-up issues)
    • Increase creative throughput by generating variations from “Lego-like” assets

Mix-and-match approach (example):

  • Generate from a combination like: “Hook concept 1 + bridge clip 4 + context clip 3 + CTA clip 3”
  • Then generate new versions by changing one variable at a time (model, background, CTA angle, etc.).

Automated split testing: mapping variables to performance signals

  • Use retention/watch-time insights to locate where drop-off occurs:
    • If retention drops sharply at CTA, adjust CTA variables (shot type, pacing, framing, audio, etc.).
    • If drop-off occurs at a context clip, change that segment’s framing/music/pace/audio.

Actionable “variable tweak” examples:

  • Change model outfit (e.g., LOTR-themed shirt style instead of green jacket)
  • Change background/theme (e.g., medieval convention with fans/merch booths)
  • Change CTA filming style (e.g., from prop-up phone to selfie with one hand)
  • Principle: keep the vibe, change one element to isolate impact

Tools & utilities mentioned (operations enabling execution)

  • Claude + Claude MCP (HixField): main automation layer for generating prompts/creatives from provided clip context.
  • VidIQ (Chrome extension) for outlier spotting:
    • Purpose: identify account creatives that are outliers and show outlier multipliers (e.g., “3.6x”).
    • Visual cue mentioned: color intensity indicates the strength of the outlier.

Metrics / KPIs referenced

Top-of-funnel KPI

  • Views
    • Baseline example: ~800–1,500 views (avg ~1,100)
    • Outlier example: ~2,500 views (~2x)
    • Strong outlier example: 30,000 vs 7,000 views (~4.3x)

Cadence expectation

  • Test 7–20 videos to find outliers (with “7 to 10” referenced as typical discovery range)

Retention KPI

  • Watch time / retention drop-off points
    • Used to decide which segment to split test (hook/context/CTA)

Note: The transcript claims an outcome like “$100k/month” and “consistent 1K days,” but does not provide specific revenue or conversion details (e.g., CAC/LTV/churn) in the subtitles.

Concrete recommendations from the presenter

  • Do not skip context: Results degrade if you don’t provide the correct “organic ethos” via video frames and structured clip library entries.
  • Test enough creatives to find a baseline + outlier: Don’t iterate off “nothing.”
  • Double down on outlier variables: Identify what changed and keep the highest-impact component(s).
  • Automate by remixing assets (not regenerating every time): reduces repetitive clip reuse and creative fatigue.
  • Use retention diagnostics to patch weak segments: adjust CTA/context where drop-off occurs.
  • Use VidIQ to speed outlier identification from competitor/benchmark accounts.

Presenters / sources

  • Presenter: “Smith” (spoken as “Smith here”)
  • Tools/Sources referenced:
    • Claude with HixField MCP
    • VidIQ (Chrome extension)
    • Other social analytics apps mentioned (e.g., Instagram/“Instagram edits”)

Original video