Campaign

Campaign

Email campaign workflow with safety states. Builds, previews, and sends targeted email campaigns to Webmatrices audience segments. Three-state machine ensures you never accidentally send to 27K users. Performance rules in this skill are derived from real Brevo data (April 2026 analysis, internal opens filtered).

---

STATE MACHINE

DRAFT → PREVIEWED → SENT

State	What happened	What can happen next
DRAFT	Email content built, segments chosen	/campaign preview (sends to the site owner only)
PREVIEWED	The site owner received the test email	/campaign send (fires to full audience)
SENT	Email delivered to segments	Done

Rules:

• /campaign [anything] = always starts in DRAFT state. Never sends.
• /campaign preview = sends to the site owner ONLY (fetch admin user from get_self_personas MCP). Moves to PREVIEWED.
• /campaign send = fires to full segments. If NOT previewed first, asks "Are you sure you want to send without previewing? This will go to [N] users."
• You can re-draft at any point. Going back to DRAFT resets the state.

---

WHAT THE DATA SHOWS — WORKS, KINDA WORKS, DOESN'T WORK

All numbers below come from Brevo events for the trailing 30 days, internal users (is_internal=true plus bishwasbh+* aliases) filtered out. Use these as the empirical baseline. Do not draft a campaign that ignores them.

TIER 1 — CHAMPIONS (run these on rotation)

Pattern	Example	Sent	Open %	CTR %
Personalized lapsed-user story	"5 sites got reviewed this week, [Name]"	49	51.0%	18.4%
Personalized "it took X before Y" hook	"it took 40 posts before google sent a single visitor, [Name]"	48	54.2%	10.4%
Comment notifications on hot threads	"New comment on your post 'getting rejected by AdSense...'"	6	50.0%	50.0%
Reply notifications	"New reply to your comment on '...'"	8	37.5%	25.0%

What unites the champions:

1. First-name personalization in the subject (, [Name] suffix)
2. Concrete number anchored in the hook (5 sites, 40 posts)
3. First-person story framing ("it took X before Y") — not announcement framing
4. Sent in batches of 50–200, never as mass blasts
5. Topic recipient genuinely cares about (AdSense rejection, real engagement on their post)

Run these templates weekly against fresh segments. They are 10–15× more effective than any broadcast.

TIER 2 — KINDA WORKS (good open, weak click — fix the body)

Subject	Sent	Open %	CTR %	What's wrong
"This week on Webmatrices: Vercel breach, vibe coding..."	84	27.4%	0%	Body buries CTA
"Your old content might be killing your new rankings"	80	21.2%	0%	Story told in-email; no reason to click
"A Roblox cheat just took down Vercel..."	57	21.1%	0%	Multiple links compete
"your traffic is down but your rankings are fine"	117	20.5%	3.4%	Working but underperforming the open rate
"a developer let ai write all his code for 4 months"	141	17.7%	1.4%	Same — opens are great, clicks lag

Diagnosis: news-event hooks pull opens but the email bodies leak readers. The reader opens, doesn't see one clear destination, bounces. Fix: single button, single thread, body is a tease not a recap. See "Body design" below.

Solid workhorses (no fix needed, run as-is):

Subject	Sent	Open %	CTR %
"Quick fix — here's the correct link, [Name]"	51	23.5%	3.9%
"Your checklist is waiting, [Name]"	51	23.5%	5.9%
"you joined at the right time — 21 app ideas ready to build"	234	10.7%	1.3%

TIER 3 — DOESN'T WORK (do not draft these)

Subject	Sent	Open %	CTR %	Why dead
Mass blast: "what ai, side hustles, and remote work have in common"	11,409	3.6%	0.25%	Too large; stale list; segment fatigue
"We built something new, [Name]"	116	8.6%	0.9%	Pure announcement frame
"this week on webmatrices — 4 new app ideas + fresh posts"	89	6.7%	0%	Recap framing
"4 new app ideas just landed on webmatrices"	53	5.7%	0%	Recap framing

Banned subject patterns:

• "[X] new posts/ideas/features just landed"
• "This week on Webmatrices" as opener (the actual format, not just words "this week" elsewhere)
• "We built something new"
• "[Number] [things] you might have missed"
• Any subject that telegraphs "this is a promotional update" before the recipient even opens

Aggregate channel state

Cohort	Open rate	CTR
Last 30 days, all sends	5.61%	0.81%
Excluding the one mass blast	20.79%	5.11%
Mass blast in isolation	3.60%	0.25%

Read: the channel is healthy when used surgically. One bad send dragged the average from 21% to 6%. Repeat that mistake monthly and the platform's deliverability degrades for every future send.

---

VOLUME & LIST HYGIENE RULES (NON-NEGOTIABLE)

The MCP enforces a hard cap of 500 recipients per send. There is no override. To reach more users, batch multiple campaigns and chain excludeUserIds so the second campaign excludes everyone the first reached.

These rules exist because the 2026-04 mass blast (sent before the hard cap was tightened) hit 2.93% hard bounce rate — above the ESP threshold of 2%, throttling above 5%.

Volume strategy

Goal recipient count	How to do it
≤ 500	Single send_targeted_email call. Sweet spot.
500–1,500	2–3 sequential calls, each ≤ 500, chained with excludeUserIds so no overlap
1,500+	Reconsider. Splitting a topic across that many users almost always lowers per-send performance. Narrow the segment instead.
all_active (~27K)	Blocked by MCP. Only allowed for critical TOS/policy updates after explicit override.

What the MCP does automatically (don't duplicate this work)

• Excludes is_internal=true users — no skill-level filtering needed
• Excludes users with emailNotifications=false
• Verifies internal links (/post/[slug]) exist; blocks send on broken links
• Personalizes {{username}} and {{firstName}} placeholders
• Hard-caps at 500 recipients (501+ is blocked)
• Returns the list of sentUserIds so you can excludeUserIds in the next batch

What the skill must add

• Hard-bouncer exclusion. The MCP doesn't track hard bounces (yet). After any send, manually note who bounced via Brevo dashboard and add their IDs to excludeUserIds for future sends.
• 12-month-silent exclusion. Use inactive_users segment with days: 365 to surface them, route to re-permission flow rather than including in promo campaigns.

List rotation

Do not email the same segment more than once per 7 days unless it's a transactional event (comment notification, reply, account confirmation). Repeated hits cause segment fatigue.

---

SUBJECT LINE PATTERNS (use these formulas)

Formula A — Personalized story hook (CHAMPION)

[Concrete number] [outcome verb] [time/scope], [First Name]

Examples that work:

• "5 sites got reviewed this week, [Name]"
• "it took 40 posts before google sent a single visitor, [Name]"
• "$143 in adsense bills last month, [Name]"

Why it works: name signals 1:1 intent, number signals real story, lowercase signals personal.

Formula B — News-event hook (Tier 2 — works for opens)

[Specific event/entity] [verb] [outcome]. [Optional context.]

Examples that work for opens (need body fix for clicks):

• "Google just confirmed GEO is real. They put it in the ads team."
• "A Roblox cheat just took down Vercel."
• "Your old content might be killing your new rankings."

Use only when there's a real news hook AND the body is single-thread, single-button.

Formula C — Pain-point question (Tier 2)

[State of affairs] but [contradicting outcome] — [framing question]

Examples that work:

• "your traffic is down but your rankings are fine — and nobody can explain why"
• "you joined at the right time — 21 app ideas ready to build"

Banned subject formulas

• ❌ "[Number] new [things] just landed on Webmatrices"
• ❌ "This week on Webmatrices: [recap]"
• ❌ "We built something new, [Name]"
• ❌ "[X] you might have missed"
• ❌ All-caps anywhere
• ❌ Excessive punctuation (multiple exclamation marks, ellipses)
• ❌ Emojis at the start of the subject (mid/end is fine in moderation)

---

BODY DESIGN RULES

The skill provides inner body HTML only — paragraphs, optional inline links, optional <strong> for emphasis. The MCP wraps it with the email template (header, footer, unsubscribe) and appends the CTA button when you pass ctaText + ctaUrl. Do not write the button HTML yourself. Do not write <table> wrappers. Do not write the email template chrome.

The Tier 2 problem (high open, zero CTR) is bodies leaking readers. These rules exist to fix that.

The single-destination rule

Every email has one destination — passed as ctaUrl. Do not put multiple inline links inside the body. If you absolutely must mention a secondary item, put it in a small <p> "P.S." line at the bottom with a single secondary inline link. The CTA button (rendered by the MCP from ctaText/ctaUrl) must be the dominant action.

Body structure (inner HTML only)

What the skill writes:

<p>hey {{firstName}},</p>

<p>[2-3 sentence personal frame — what made you write this email today.
Real and specific, first person, not corporate.]</p>

<p>[1 sentence specific tease about the thread — a number, a name,
a turn of phrase from the post. Not a summary. Leave the reader curious.]</p>

<p>[1 sentence why this matters to the recipient specifically — segment relevant.]</p>

<p>— bishwas</p>

The MCP appends the CTA button below this body using ctaText and ctaUrl parameters. The button styling, white-text span fix, and centering are all handled by the MCP. Don't write button markup in the body.

Body length

• Total stripped text: 80–180 words. Below 80 reads as spam. Above 180 the reader drops off before the button.
• The body is a tease, not a summary. If the email tells the whole story, there's no reason to click.

Personalization

Use {{firstName}} and {{username}} in body and subject. The MCP substitutes them per-recipient. Fall-back behavior: if firstName is null, the MCP uses username.

CTA parameters (passed alongside body)

• ctaText: action-oriented, max 5 words. ("Read the thread →", "See what changed →", "Drop your URL below →"). Never generic ("Click here", "Learn more").
• ctaUrl: relative path (e.g., /post/your-slug) or absolute. The MCP verifies the slug exists before sending.

Personal tone markers

• First person ("I"), not corporate "we"
• Lowercase greeting ("hey {{firstName}}")
• One specific detail that signals real authorship (a number, a date, a personal admission like "i sat at my desk for a while after that")
• Sign-off with first name only

---

INPUT

Input	What it does
"recent"	Finds last 2-3 published posts, builds campaign around them
Post ID	Builds campaign promoting that specific post
Post slug	Fetches post via get_post_by_slug, builds campaign around it
Description text	Builds campaign based on described theme/promotion

---

WORKFLOW

Step 1: Gather content

If "recent":

1. Call list_posts to get the 5 most recent posts
2. Pick one standout post (highest comment count, freshest news angle, strongest hook)
3. Verify the slug exists

If postId/slug:

1. Fetch the post via get_post or get_post_by_slug MCP
2. Verify it exists

If description:

1. Determine the theme
2. Search for relevant posts using search_posts_by_content or list_posts
3. Pick one post that best matches

CRITICAL: Single destination only. The data is unambiguous — multi-link emails get 0% CTR even with 21% opens. Pick one thread.

Step 2: Choose segments

Use preview_audience MCP to check segment sizes before choosing.

Segment selection rules:

1. Match segment to topic. AdSense post → AdSense segment, never "all users."
2. Default segment size target: 200–800 recipients. The Brevo data shows performance peaks in this range.
3. Cap any single send at 1,500 unless explicitly justified.

Available segments (real MCP names):

Segment	What it targets	Best for
commenters	Users who have commented at least once	High-engagement campaigns. Most reactive segment.
post_authors	Users who have published posts	Engagement boosts on related new posts
lapsed_contributors	Posted/commented but inactive days+	Reactivation of warm users (set days: 14-30)
inactive_users	Inactive for days+ (default 30)	Re-engagement (set days: 30/60/90/365)
new_users	Joined in last days (default 7)	Welcome / starter-content campaigns
checklist_users	Created project checklists	Tool/workflow related campaigns
premium_users	Has isPremium=true	Premium-only announcements
all_active	Everyone active	Blocked by MCP. Critical TOS only.

Segment chaining: when running multiple sends to overlapping segments, the MCP returns sentUserIds after each send. Pass that array as excludeUserIds on the next send so no user gets duplicates.

Time-window parameter (days): for inactive_users, lapsed_contributors, and new_users, the days argument tunes the window. Use it to surface the right cohort (e.g., lapsed_contributors with days: 30 = posted/commented but quiet for a month).

Step 3: Internal user exclusion (handled by MCP)

The MCP automatically excludes is_internal=true users from every send_targeted_email call. You do not need to fetch persona IDs and add them to excludeUserIds. This was previously a skill-level requirement; it's now baked into the MCP. Just use excludeUserIds for chaining across multiple sends and known hard-bouncers.

Step 4: Pre-flight quality check

Before showing the draft, run this internal check. Score 1 point per ✅. Block any draft scoring below 6/8.

Check	Pass criteria
Subject matches a Tier 1 or 2 formula	First-name + number + story OR clear news-event hook
Subject is NOT a banned format	No "X just landed", no "this week on", no "we built"
Subject is under 70 characters	Mobile preview cuts off
Body provides single ctaUrl destination	One thread, one button
Body word count 80–180 (stripped)	Sweet spot for opens-to-click conversion
Body does NOT contain manual button HTML	MCP renders the button from ctaText/ctaUrl
Body does NOT contain template chrome	No <table> wrappers, no header/footer markup
Segment matches topic and ≤ 500 recipients	MCP hard-caps at 500

Step 5: Present the DRAFT

Show:

• Subject line (with character count)
• Target segment(s) with estimated user count
• Excluded segments/users count
• Full HTML body
• Pre-flight score (e.g., "9/9 — ready to preview")
• STATE: DRAFT

Ask: "Ready to preview? Run /campaign preview to send a test to the site owner."

NEVER auto-send. NEVER proceed past DRAFT without user action.

---

SUBCOMMAND: /campaign preview

1. Send the email to the site owner ONLY using send_email MCP with:
   • userId: 1 (the site owner's user ID)
   • Full subject and body as built in the draft
   • Do NOT use send_targeted_email for preview — it picks from the segment, not the site owner
2. Move state to PREVIEWED
3. Tell the user: "Test email sent to the site owner. Check your inbox. When ready: /campaign send"

---

SUBCOMMAND: /campaign send

1. Check state is PREVIEWED
   • If YES: proceed
   • If NO: warn "This campaign hasn't been previewed. Are you sure you want to send to [N] users without testing? Say 'yes send it' to confirm."
2. Re-run pre-flight checks (in case anything changed since draft)
3. Send via send_targeted_email MCP with all exclusions applied
4. Move state to SENT
5. Report: "Campaign sent to [N] users across [segments]. Subject: [subject]. Track results in Brevo dashboard after 48 hours."

---

CAMPAIGN TYPES & WHEN TO USE EACH

Reactivation (CHAMPION — use weekly)

Target: lapsed users (no login 30–90 days, opened at least one email in last 12 months). Format: Tier 1 personalized story hook. Subject: "5 sites got reviewed this week, [Name]" / "it took 40 posts before google sent a single visitor, [Name]" / variant. Send size: 100–300 per batch. Cadence: weekly.

Hot-thread reply boost

Target: users who commented on related threads in last 30 days. Format: Tier 2 news-event hook OR Formula C pain-point. Subject: news angle from the post. Send size: 80–250. Cadence: 1× per major post.

Onboarding (transactional)

Target: just-registered users. Format: short, single CTA to a starter thread. Subject: "Confirm your email" / "Welcome to Webmatrices!" — these already perform well (38–55% open).

Re-permission (long-silent users)

Target: 12+ months no opens. Format: explicit "still want to hear from us?" with single yes/no link. Send size: 500 per batch max. Cadence: once per stale segment, then drop them if no reply.

NOT a campaign type

• ❌ Generic platform recaps ("this week on Webmatrices")
• ❌ Feature announcements ("we built something new")
• ❌ Mass blasts to all 27K users for any reason except a critical TOS/policy update

---

SAFETY CHECKLIST (before any send)

• Post slug in ctaUrl is verified to exist (the MCP will block on broken slug — but check yourself first)
• Subject matches a Tier 1 or Tier 2 formula
• Subject is NOT a banned format
• Body is inner HTML only (paragraphs, no template chrome, no manual button markup)
• Body word count 80–180 (stripped)
• Single ctaUrl destination
• Send size ≤ 500 (MCP hard cap)
• Segment matched to topic (not all_active)
• Hard-bouncers added to excludeUserIds (manual until MCP supports it)
• Multi-batch sends chain excludeUserIds from previous sentUserIds
• Pre-flight score ≥ 6/8
• Preview sent to site owner before full send (or explicit override with reason)

---

QUARTERLY REVIEW

The performance numbers in this skill are from the trailing 30 days as of 2026-04-27. Re-run the Brevo analysis quarterly:

1. Pull /v3/smtp/statistics/events for last 90 days
2. Filter out is_internal=true users (and bishwasbh+* aliases)
3. Bucket by template (strip , [Name] suffix)
4. Update the TIER 1 / TIER 2 / TIER 3 tables in this skill if patterns shift
5. Update the Wiki diagnosis at wiki/Forum-Engagement-Diagnosis-YYYY-MM.md

If a pattern stops working, retire it from TIER 1 immediately. The empirical data leads the rules.

---

IMPORTANT

• State machine is non-negotiable. DRAFT → PREVIEWED → SENT.
• The empirical performance tiers lead the rules. Do not draft against patterns proven dead.
• Single destination per email. One ctaUrl. Multi-link emails get 0% CTR even at 21% opens — proven repeatedly in the data.
• MCP hard-caps at 500 recipients per send. Larger reach = multiple chained sends.
• Body is inner HTML only. Paragraphs and inline tags. The MCP wraps the template, adds the button, handles white-text fix, personalizes, and excludes internal/opted-out users automatically.
• Personal tone, not corporate. First person, lowercase greeting, one specific detail that signals real authorship.
• Quarterly performance review. Update tiers when data shifts. Reference wiki/Forum-Engagement-Diagnosis-YYYY-MM.md.