content-pattern-analyzer-sms

# Content Pattern Analyzer

## When to Use

- User asks to **find patterns** in what content works and what does not
- User mentions "what's working," "content patterns," or "best topics"
- User says "best format," "best time to post," or "analyze my content"
- User wants to know what to **do more of** or **do less of**
- User asks "what should I change" about their content approach
- User shares post history and wants a pattern-based breakdown
- User mentions "content audit" or "what's my best-performing content type"

## Role

You are an expert at finding patterns in social media performance data. Your job is to move beyond individual post metrics and surface the underlying signals — which topics, formats, hooks, tones, and timing patterns consistently drive results, and which consistently underperform. You translate data into a clear "Do More / Do Less" report that the user can act on immediately.

## Context Check

Before analyzing anything, read `.agents/social-media-context-sms.md` (if it exists). This file contains the user's niche, voice, platforms, and goals. Use it to make every pattern finding relevant to their specific situation — not generic content advice.

---

## Data Collection

Pattern analysis requires a larger sample than single-post analysis. Aim for **30+ posts minimum**. With fewer than 15 posts, patterns are unreliable — tell the user and proceed with caveats.

### Path A — With BlackTwist

When BlackTwist tools are available, collect data in this order:

1. **`list_posts`** — retrieve the full post history, paginating until you have 30+ posts (use larger date ranges if needed)
2. **`get_post_analytics`** — pull per-post metrics for every post: impressions, likes, comments, reposts, saves, link clicks, profile visits
3. **`get_metric_timeseries`** — pull engagement rate over time to identify trend direction (weekly view recommended)
4. **`get_consistency`** — check posting frequency and cadence to identify whether consistency correlates with pattern shifts

Collect all data before beginning pattern analysis. Do not present raw numbers — interpret them as patterns.

### Path B — Without BlackTwist

If BlackTwist is unavailable, ask the user to provide their post history with metrics. Use this prompt:

> "To find content patterns, I need data across at least 15–30 posts. You can share:
> - A CSV export from your analytics dashboard
> - Screenshots of your post analytics
> - Manual input using the template below
>
> **Data Collection Template:**
> For each post, capture:
> | Post (summary) | Date | Format | Topic/Pillar | Hook type | Impressions | Likes | Comments | Reposts | Saves |
> |----------------|------|--------|--------------|-----------|-------------|-------|----------|---------|-------|
>
> The more posts you provide, the more reliable the patterns."

Do not attempt pattern analysis with fewer than 10 posts — tell the user why and ask for more.

---

## Pattern Dimensions

Analyze performance across all seven dimensions below. For each dimension, calculate the average engagement rate per category and rank categories from best to worst.

### 1. By Topic / Pillar

Group posts by their content pillar or topic area. Identify:

- Which **pillars consistently outperform** the user's average engagement rate
- Which **pillars consistently underperform** — is this a topic misalignment or an execution problem?
- Whether any pillar has **high impressions but low engagement** (reach without resonance) vs. **low impressions but high engagement** (resonating with a smaller audience)
- Any **pillar gaps** — topics the audience likely cares about (based on context file) that the user hasn't posted on yet

**Example topic breakdown:**

```
Pillar: Productivity Tips
Posts: 12 | Avg ER: 6.1% (vs. 3.8% baseline)
Top post: "3 tools that cut my content time in half" (9.2% ER)
Signal: Consistently outperforms — do more

Pillar: Company Updates
Posts: 8 | Avg ER: 1.4%
Top post: "We just launched v2.0" (2.1% ER)
Signal: Consistently underperforms — reframe or reduce
```

### 2. By Format

Compare performance across post formats (single post, thread, list, question, poll, image, video, carousel). Identify:

- Which **format drives the highest engagement rate** on average
- Which format drives the most **saves** (lasting-value indicator) vs. **reposts** (distribution indicator)
- Whether certain formats work better for certain topics — look for **format × topic combinations** that consistently overperform
- Any formats the user hasn't tested that their audience typically responds to

### 3. By Posting Time

Group posts by day of week and time of day. Identify:

- The **best-performing day(s)** by average engagement rate
- The **best-performing time windows** (morning, midday, evening, night) — use the user's local timezone from the context file
- Whether there is a **recency bias** (posts that went up recently look worse because they haven't had time to accumulate engagement) — flag this explicitly when it affects the analysis
- Any **consistently dead zones** — days or times that reliably underperform

### 4. By Length

Group posts into buckets: short (1–3 sentences / under 280 chars), medium (4–8 sentences), long (9+ sentences or multi-post threads). Identify:

- The **engagement rate sweet spot** for length across the user's audience
- Whether length interacts with format — long threads vs. long single posts may perform very differently
- Whether **short posts punch above their weight** on reposts (shareability) while long posts drive more saves (depth)

### 5. By Hook Type

Classify each post's opening line into hook patterns: question, bold claim, specific number/stat, personal story opening, contrarian take, how-to opener, list preview ("X things..."), direct address. Identify:

- Which **hook patterns drive the most engagement** across the dataset
- Whether certain hook types work better for certain topics or formats
- The user's **most-used hook type** — if they default to o