Ensemble Content Scorer

Wisdom of crowds, but for AI. This skill scores your content ideas using multiple AI models, then aggregates for consensus. More reliable than single-model predictions.

WHAT IT DOES

            Content Idea
                 │
┌────────────────┼────────────────┐
│                │                │
▼                ▼                ▼

[Claude] [GPT-4o] [Gemini] Score Score Score │ │ │ └────────────────┼────────────────┘ │ ▼ [Aggregator (Claude)] │ ▼ Consensus Score + Verdict

WHY MULTI-MODEL?

Single Model Ensemble

May have biases Biases cancel out

One perspective Multiple perspectives

Black box score Transparent reasoning

May miss nuances Catches different angles

TRIGGERS

Use this skill when you say:

• "Score this content idea"

• "Is this topic worth pursuing?"

• "Rate my video concept"

• "Predict if this will go viral"

• "Ensemble score: [topic]"

USAGE

In Claude Code (Recommended)

"Ensemble score: Statins myth-busting for Indian audience"

"Score this video idea: Why your LDL target depends on your risk"

"Rate these ideas and rank them:

1. GLP-1 agonists explained
2. Heart attack warning signs
3. Is coconut oil heart-healthy?"

CLI Mode

Score single idea

python scripts/score_content.py --idea "Statins myth-busting for Indian audience"

Score multiple ideas

python scripts/score_content.py --ideas "GLP-1 explained" "Statin myths" "CAC scoring"

Use specific models

python scripts/score_content.py --idea "Topic" --models claude,gpt4o,gemini

SCORING DIMENSIONS

Each model scores on these dimensions (1-10):

Dimension What It Measures

Relevance How relevant to target audience (Indian patients/doctors)

Novelty How fresh is the angle? Been covered before?

Expertise Match Does it match your expertise as interventional cardiologist?

Engagement Potential Will it capture and hold attention?

Share-ability Will people share this? Controversy potential?

Evergreen Factor Will this be relevant in 6 months?

Total Score: 0-60

OUTPUT FORMAT

ENSEMBLE CONTENT SCORE

Idea: Statins myth-busting for Indian audience - why most "side effects" aren't real

Date: 2025-01-01

INDIVIDUAL MODEL SCORES

Claude (Anthropic)

GPT-4o (OpenAI)

Gemini (Google)

CONSENSUS SCORE

VERDICT

🟢 STRONG PURSUE (Score: 49/60, Consensus: High)

All models agree this is a strong content idea. The combination of:

• High relevance to your audience
• Perfect expertise match
• Good controversy factor
• Evergreen potential

Makes this a priority topic for your content calendar.

RECOMMENDATIONS

1. Angle Enhancement: Focus on the "nocebo effect" - most statin "side effects" are psychosomatic
2. Hook Suggestion: "90% of statin side effects aren't real - here's the data"
3. Format: 12-15 minute deep dive with studies
4. Hinglish Tip: Use "side effect ka drama" for relatability

DISSENT ANALYSIS

• Gemini scored lower on engagement (7 vs 8-9)
• Suggests: May need stronger hook to maximize viral potential
• Consider: Adding patient testimonial or counter-narrative

SCORING TIERS

Score Range Verdict Action

50-60 🟢 STRONG PURSUE High priority, create immediately

40-49 🟡 WORTH PURSUING Good idea, add to calendar

30-39 🟠 NEEDS REFINEMENT Has potential, needs angle work

20-29 🔴 RECONSIDER Weak idea, low priority

0-19 ⛔ SKIP Not worth the effort

CONSENSUS INTERPRETATION

Std Deviation Interpretation

< 3 High consensus - models agree

3-5 Moderate consensus - some disagreement

5 Low consensus - divisive idea (may be worth exploring!)

INTEGRATION

Enhances:

• viral-content-predictor

• More reliable predictions

• youtube-script-master

• Validate topics before scripting

• content-repurposer

• Know which content to repurpose

Workflow:

Idea Generation → Ensemble Score → [High Score?] → Create Content ↓ [Low Score?] → Refine or Skip

MODELS USED

Model Provider Cost Notes

Claude Sonnet Anthropic Subscription Your primary

GPT-4o OpenAI API Strong analysis

Gemini Pro Google FREE Good for fact-checking

Grok xAI API Twitter trend awareness

Minimum required: 2 models (Claude + one other) Recommended: 3+ models for robust consensus

DEPENDENCIES

anthropic>=0.18.0 openai>=1.0.0 # For GPT-4o google-generativeai>=0.3.0 # For Gemini python-dotenv>=1.0.0 rich>=13.0.0

API KEYS NEEDED

Key Purpose Status

ANTHROPIC_API_KEY Claude Already have

OPENAI_API_KEY GPT-4o Already have

GOOGLE_API_KEY Gemini Already have

XAI_API_KEY Grok (optional) Already have

BATCH SCORING

For scoring multiple ideas at once:

python scripts/score_content.py --batch
--ideas "GLP-1 for heart failure"
"Statin myth-busting"
"CAC scoring guide"
"Why LDL matters"
"Exercise for heart health"

Output:

NOTES

• Speed: ~30 seconds for single idea (parallel API calls)

• Cost: Minimal - short prompts to each model

• Reliability: Consensus typically more accurate than single model

• When to ignore: If YOU have strong conviction, trust your expertise

This skill helps you invest your time in content that's more likely to succeed.