Email Strategist — Newsletter Creation Methodology

Purpose

Turn a topic into a complete, client-specific newsletter draft. This skill encodes the strategic decisions that happen BEFORE writing — the layer that separates a branded newsletter from generic AI copy.

Core Principle

The model doesn't matter. The context matters. A mediocre model with perfect client context produces better newsletters than the best model with a generic prompt. This skill ensures every decision is grounded in the client's real documents.

The Anti-SOP Approach

Traditional AI email workflow (what prompt playbooks sell):

1. Create a project in Claude
2. Upload brand assets manually
3. Upload a "strategy PDF" (someone's lead magnet)
4. Paste a master prompt with fill-in-the-blank variables
5. Generate copy
6. Edit manually
7. Repeat from scratch next time — nothing learned

Email Marketing Manager approach:

1. Box folder exists with your documents (already done)
2. /email-marketing-manager:create Topic: "spring sale" (one command)
3. Newsletter produced — with verified images from Cloudinary, segment variants based on your personas, angle selected based on what hasn't been used recently
4. /email-marketing-manager:track after it sends — real data recorded
5. Next /email-marketing-manager:create uses those learnings automatically

The difference: no SOP to follow. No prompt to write. No manual upload. No PDF strategy framework from a stranger. No starting from scratch. The skill does the strategic thinking the SOP skips entirely.

Angle Selection

Before writing a single word, determine the angle. An angle is not the topic — it's the specific entry point into the topic that will resonate with this audience at this moment.

Angle Quality Test

Every angle must pass three checks:

1. Not a repeat. Check newsletter history. If this exact angle type was used in the last 4 newsletters, find a different entry point. Repetition kills open rates. If no history exists, any angle is valid.
2. Connects to a real pain point. Check audience personas. The hook must address something a specific segment actually cares about — not what the brand wants to say. Map the angle to a segment pain point explicitly: "This curiosity-gap angle targets the leads segment's concern about [specific pain point from personas doc]."
3. Native to brand voice. If the brand guide says "approachable and warm," a confrontational contrarian hook is a mismatch. If the brand is "operationally honest," hype-driven urgency doesn't fit. The angle must feel natural in the brand's voice.

Angle Types

Default rule: Select the angle type that best matches the brand voice. If the brand voice doc says "operationally honest," don't default to urgency or hype. If it says "warm and personal," don't default to educational lecture.

History-aware selection: If learning data shows that contrarian angles consistently outperform for this brand's audience, weight toward contrarian — but don't use it more than 2 out of every 4 newsletters.

When data conflicts with instinct: Trust the data. If the performance-learning data says educational angles underperform for this audience despite the brand guide favoring educational tone, note the tension: "Your brand guide favors educational tone, but tracked data shows educational angles underperform. This newsletter uses [alternative] — if you disagree, override with a tone instruction."

Image Selection — Cloudinary Closed List

The Rule

Every asset ID you cite in the newsletter must appear in the closed candidate list assembled before drafting. An ID that is not in the list does not exist. A URL you construct yourself does not exist. A filename you remember from a past session does not exist unless it is in the current list.

This is the single most important rule in image selection. It prevents broken images in sent newsletters.

How It Works

Before writing begins, the system queries Cloudinary for available assets. The query uses the newsletter's topic, client context, and any tags or folder structure the client has set up. The result is a closed candidate list — a block of verified assets with:

• Asset ID — the Cloudinary public ID (the only valid reference)
• Delivery URL — the CDN URL that actually resolves
• Native dimensions — width × height as stored
• Format — jpg, png, webp, etc.
• Metadata — tags, description, folder, created date

Selection Criteria

For each image placement in the newsletter, select from the candidate list based on:

1. Subject fit — does the image relate to this newsletter's topic and angle?
2. Orientation — hero images need landscape or 2:1; item images need square or near-square
3. Palette — does the image's color palette work with the brand's visual identity?
4. Email suitability — will it render well at email width? Avoid images that depend on fine detail at small size.
5. Recency — has this image been used in the last 3 newsletters? Avoid repeats.

Image Roles

Alt Text

Every selected image gets alt text. The rule:

• If the image's Cloudinary metadata includes a confirmed description: use it
• If nobody has visually confirmed what the image shows: mark as ALT PENDING
• A plausible guess at alt text is a fabrication — treat it as one

Image Manifest Format

For each selected image, produce:

Asset ID: [cloudinary public_id]
Placement: [hero / section / item / divider]
Delivery URL: [cloudinary CDN URL with transformations]
Native: [width × height]
Rationale: [why this image for this placement]
Alt text: [confirmed description or ALT PENDING]
Rights: [from metadata or UNVERIFIED]
Credit: [from metadata or none]
Flags: [any concerns — low resolution, portrait for hero slot, etc.]

When Images Aren't Available

If Cloudinary returns no matching assets or the connector isn't configured:

• Produce the newsletter without images
• At the end, suggest what kind of image would strengthen the piece
• Never generate placeholder images, suggest stock photo sites, or construct Cloudinary URLs by hand

Writing the Newsletter

Subject Line Rules

• Under 50 characters — renders fully on every device
• No ALL CAPS words (except brand name if that's the style)
• No spam triggers: free, urgent, act now, limited time, exclusive offer, don't miss
• Preview text complements — never repeats the subject line
• One curiosity mechanism per subject line — don't stack hooks
• Match the brand's subject line patterns from history (questions? statements? numbers?)

Body Copy Structure

Adapt to newsletter goal:

Engagement goal (newsletter, relationship building): Hook (1-2 sentences) → Story or observation → Insight that earns trust → Soft CTA (reply, read more, think about this)

Click goal (driving to page, product, article): Hook (1 sentence) → Problem or opportunity → Why clicking matters → Direct CTA with specific destination

Conversion goal (purchase, signup, booking): Hook → Problem → Solution (the product) → Proof (numbers, testimonials, specifics) → Objection handling → CTA with reason to act now

Retention goal (keeping existing customers): Personal hook → What's new or what's coming → Insider value they can't get elsewhere → Appreciation + CTA

Voice Application

Read the brand voice from client-context. Apply through:

• Sentence structure — formal brands use longer, compound sentences. Casual brands use fragments. Match past newsletters.
• Vocabulary — use the brand's word list. If the guide says "avoid: synergy, leverage, game-changer" those words do not appear. Ever.
• Punctuation — em dashes, Oxford commas, exclamation points (or absence). Match the guide.
• Greeting/sign-off — match past newsletter patterns exactly. If past newsletters start with "Hey [first_name]," use that. If they open cold with no greeting, don't add one.
• Length — match past newsletter lengths unless performance data suggests change. If the brand sends 300-word newsletters, don't produce 800 words.
• Personality markers — does the brand use "I" or "we"? Do they ask questions? Do they use parenthetical asides? Match the personality, not just the vocabulary.

The Substitution Test

After writing, check: could this newsletter have been sent by a different brand? Read the body without the brand name. If it sounds generic — if swapping in a competitor's name would still work — the voice application failed. Rewrite the opening and CTA.

This test is pass/fail. Don't narrate success. If it passes, move on. If it fails, fix it.

Segment Variants

What Changes Per Segment

What Does NOT Change

• Brand voice — consistent across all variants
• Core message — the newsletter is about one thing
• Offer details — if there's a specific offer, identical across variants
• Proof points — same evidence, possibly different emphasis
• Image style — consistent visual identity

Variant Quality Test

Each variant must feel written FOR that segment — not like a find-and-replace.

Bad: "As a valued customer..." (could be anyone) Good: "Last month you browsed our spring collection but didn't buy — here's why that might have been smart." (specific behavior)

Bad: "Great news for our subscribers!" (generic) Good: "You told us you care about ingredients. So we changed ours." (specific to a feedback-active segment)

If a variant doesn't pass this test, rewrite the opening. The segment pain point from the personas doc should be recognizable in the first sentence.

Newsletter Record

Every newsletter must be saved as a record:

Newsletter: [title/topic]
Date: [created date]
Angle: [type] — [specific hook]
Segments targeted: [list]
Subject line (primary): [text]
Images used: [asset IDs or "none"]
Predicted open rate: [X%] ([confidence])
Goal: [engagement/clicks/conversions/retention]

This record accumulates in the Box folder. By newsletter 5, the system has meaningful history. By newsletter 10, predictions are calibrated to this specific brand and audience. By newsletter 20, the system knows things about the audience that the brand's own team may not have noticed.

Performance Benchmarks (Defaults)

Use only when no client-specific data exists. Always state these are benchmarks:

After the first /email-marketing-manager:track run, benchmarks are replaced with real data. Never use benchmarks when real data exists. Never present benchmarks as predictions.