---
name: curcuma-email-research
route: "Research how brands handle a specific email type or pattern"
router-terms: research, competitive, analysis, how do brands, what do companies, best practices, examples, trends, benchmark, compare, survey, patterns
description: |
  Research how real brands handle a specific email type or design pattern.
  Searches the corpus, clusters results, extracts common patterns, and
  delivers a structured analysis with stats and examples. Use when the user
  wants to understand a design space before building, needs competitive
  intelligence, or wants best-practice examples.
  Trigger on: "research", "how do brands do", "what do companies use",
  "best practices for", "examples of", "competitive analysis", "trends in",
  "benchmark", "what's standard for".
---

# Curcuma Email Research

Research how 19,795 real emails across 4,367 brands handle a specific design
problem. Deliver evidence-backed analysis, not opinions.

## When to Use This

- User wants to understand an email type before designing ("how do SaaS
  companies do onboarding emails?")
- User wants competitive intelligence ("what do competitors' cart abandonment
  emails look like?")
- User wants pattern prevalence ("how common are dark mode emails?")
- User wants examples for inspiration ("show me great black Friday emails")

## When NOT to Use

- User wants a finished email written → `curcuma-email-design`
- User has HTML to fix → `curcuma-email-improve`

## CRITICAL: Output Behavior

**This skill produces a structured analysis, not an email.**

1. Output findings inline as structured text (tables, stats, examples)
2. Include slugs so the user can study specific emails with get_email_summary
3. Cite prevalence stats from get_pattern where available
4. Don't generate an email — that's the design mode's job

---

## Workflow

### Step 1: Clarify the research question

Before searching, understand what the user actually wants to know:
- Email type (welcome, cart abandonment, re-engagement, etc.)
- Industry/category (SaaS, ecommerce, media, etc.)
- Specific technique (dark mode, bulletproof buttons, preheaders)
- Output shape (overview? deep analysis? just examples?)

If the user says "research welcome emails," they probably want: common
structures, CTA patterns, copy approaches, and 5-10 best examples.

### Step 2: Discover the category space

If the research is about an email type, find the right categories:

```
list_categories()
```

Then search within relevant categories:

```
search_emails(query="welcome", category="onboarding", limit=10, max_per_company=1)
search_emails(query="welcome", limit=10, max_per_company=1)
```

Run 2-3 searches with different angles:
- By email type: "welcome", "cart abandonment", "win back"
- By technique: "bulletproof button", "dark mode", "preheader"
- By industry: search with `company` filter for specific brands

### Step 3: Study a representative sample

Pick 5-8 diverse results and run get_email_summary on each:

```
get_email_summary(slug="<slug-from-search>")
get_email_summary(slug="<another-slug-from-search>")
```

**Hard fail:** only put a slug in “Best Examples” if `get_email_summary`
succeeded with real fields. Skip error/empty rows. Never paste example slugs
from this skill file unless search returned them and summary succeeded.

From the summaries, extract patterns:
- **Layout**: single-column vs multi-column ratio
- **CTA type**: button vs image-link vs text-link distribution
- **Color approaches**: common palette patterns
- **Copy structure**: headline → body → CTA → footer patterns
- **Preheader usage**: how many use it, what copy style
- **Image count**: average images per email
- **Structural complexity**: table count, section count

### Step 4: Get prevalence stats

For any technique the research touches on:

```
get_pattern(name="bulletproof button")
get_pattern(name="responsive")
get_pattern(name="preheader")
get_pattern(name="dark mode")
get_pattern(name="footer")
get_pattern(name="structure")
```

These give **live** hard numbers from the tool response — not opinions and
not numbers memorized from this skill file. Quote prevalence only as returned.

### Step 5: Synthesize findings

Structure the analysis:

```
## [Email Type] Research — [Date]

### Corpus
- Searched [N] emails matching "[query]"
- Studied [M] summaries in detail
- Category: [if applicable]

### Layout Patterns
| Pattern | Prevalence in sample | Industry baseline |
|---|---|---|
| Single-column | X/Y | — |
| Multi-column | X/Y | — |
| Hero image + text | X/Y | — |

### CTA Patterns
| CTA type | Count | Example companies |
|---|---|---|
| Bulletproof button | X | Brand A, Brand B |
| Image link | X | Brand C |
| Text link | X | Brand D |

### Copy Structure
Common pattern: [Headline] → [Subhead] → [Body] → [CTA] → [Footer]
Average word count: N
Preheader usage: X/Y (Z%)

### Color Approaches
- Primary colors trend: [observation]
- Accent colors: [observation]
- Background: [observation]

### Structural Stats (from get_pattern — paste tool strings only)
- Bulletproof buttons: {get_pattern prevalence string}
- MSO conditionals: {get_pattern prevalence string}
- Responsive: {get_pattern prevalence string}
- Preheaders: {get_pattern prevalence string}
- Dark mode: {get_pattern prevalence string}

**Hard fail:** never fill this section with training-memory percentages.
If you skipped `get_pattern`, write "not retrieved" instead of inventing numbers.

### Best Examples
1. [Brand] — [slug] — [why it stands out]
2. [Brand] — [slug] — [why it stands out]
3. [Brand] — [slug] — [why it stands out]

### Key Takeaways
1. [Evidence-backed finding]
2. [Evidence-backed finding]
3. [Evidence-backed finding]
```

### Step 6: Deliver

Present the analysis. Include slugs for every example so the user can
deep-dive with `get_email_summary` themselves.

If the user wants to go from research to a finished email, suggest:
> Ready to design? I can write a [email type] email grounded in these
> findings. Just say the word.

## Grounding

See `skills/GROUNDING.md`. Research output must list **real slugs** for every
example quoted. Prefer quoting actual subject lines / CTA labels from
`get_email_summary`, not paraphrased AI marketing speak.

## Anti-patterns

- DON'T present opinions as facts — every claim should trace to a stat from
  get_pattern or an observation from get_email_summary
- DON'T study fewer than 5 emails — small samples mislead
- DON'T skip get_pattern — it gives the industry-wide baseline that contextualizes
  your sample findings
- DON'T write an email — that's the design mode's job. Deliver research, then
  hand off
- DON'T use `get_email_html` for research — 60KB per email wastes context;
  `get_email_summary` has all the design intelligence in 3KB
- DON'T invent example subjects or “typical” CTAs without a slug
- DON'T deliver research with zero tool calls