Marketing Email Campaign

Marketing Email Campaign

Purpose

Run a marketing email campaign end to end: an HTML template that renders correctly in Outlook, Gmail and mobile clients, a compliance layer that satisfies Australia's Spam Act, a seed-test protocol that validates the exact construction before the real send, header-level deliverability diagnosis, and a warm-up plan that builds sender reputation without burning it.

When to use

Before the first campaign from a new sending domain, when building or fixing an email template (layout breaking on phones, backgrounds misbehaving, unstyled previews), when a send lands in junk and you need to know why, or when speccing the tool stakeholders will use to send. Pair with resend-email-sending for the provider layer, m365-email-authentication for DKIM/DMARC, and supabase-marketing-backend for the forms campaigns point at.

Inputs expected

Partial inputs are fine — state assumptions.

• Brand assets (logo, palette, display font) and the campaign's copy blocks (subject, preheader, body, CTAs)
• The sending domain/subdomain, provider, and authentication state (SPF/DKIM/DMARC)
• The audience: list source, sectors (consumer vs corporate/public-sector domains), and expected volume
• The published privacy stance (tracking or no tracking) and who operates the send tool

Guiding principles

• Size the card with the fluid-hybrid pattern, never percentage max-width. A table sized width:600px; max-width:100% resolves back to 600px on phones inside nested tables (percentage max-width is circular in auto table layout) and overflows the viewport — confirmed on both iOS and Android. Use width="100%" style="width:100%; max-width:600px" on the card, wrapped in an MSO conditional ghost table so desktop Outlook stays pinned at 600. See Fluid-hybrid card in reference.md.
• Never overlay text on a background-size:cover cell. The artwork re-proportions whenever content height changes (mobile text wrap grows the cell and the colour boundary slides into the text), so it will collide at some width. Structure it as a normal full-width <img> transition strip whose bottom edge ends in one flat solid colour, followed by a separate bgcolor cell of that exact colour holding the text: seamless join, legible at every width, Outlook-safe.
• Draw flat-vector brand splash art programmatically; bake the wordmark, keep sublines live. Exact hex values and compositing the real logo beat AI generation for flat art (exact palette, crisp small-size rendering, reproducible from a committed script); AI art wins only for organic complexity on large canvases. Bake the wordmark into the header image using the brand's display font (a Google Fonts link helps Apple Mail only; everyone else gets the fallback stack), and keep audience-specific sublines as live text so campaigns can vary them.
• A preview page on the product site needs its own CSP. A site CSP of style-src 'self' blocks ALL inline style attributes — the entirety of an email template's styling — so the preview renders unstyled (serif, blue links, visible preheader) while real sends are unaffected (email clients never see the site's headers). Give the preview route a CSP allowing 'unsafe-inline' styles, mark it noindex, and keep any analytics beacon off it.
• The Spam Act attaches to every commercial email sent, not to having "subscribers". Cold outreach still requires sender identification (legal entity line) and a functional opt-out — a reply-"unsubscribe" line is the minimal compliant form, and the regulator's largest fines have been for unsubscribe failures specifically. If the owner directs removing it for a one-off send, record the direction and date in the template's comments, and keep unsubscribe mandatory in any repeatable sending tool (unique signed link, suppression list enforced at send time, List-Unsubscribe header).
• The visible unsubscribe link goes to a page that only confirms on GET and acts on POST. Mail-security scanners (Outlook Safe Links) prefetch every link with a GET and would silently unsubscribe recipients — render only a confirmation button, fire the unsubscribe POST on explicit click. Make the endpoint idempotent (200 for unknown or already-removed tokens, no info leak), treat the token as opaque and never log it, and serve the page with Referrer-Policy: no-referrer. The RFC 8058 List-Unsubscribe header keeps pointing at the endpoint directly; one-click needs no page.
• Seed-test the exact construction before any campaign. Send the real email through the real tool to a Gmail, an Outlook.com and a corporate-tenant address, then read the raw source for SPF/DKIM/DMARC verdicts, multipart/alternative, and the List-Unsubscribe headers. In the same pass verify content assembly: subject appears once, preheader is independent of the subject, and there is exactly one greeting — tools that prepend their own greeting mean the body copy must not include one (and check what renders when the recipient name is blank). Run a second seed after any fix so the send validates the exact construction the campaign will use.
• Read Microsoft's placement header before blaming content. X-Microsoft-Antispam-Mailbox-Delivery states WHY placement happened: dest:I vs dest:J plus an OFR reason, where TrustedSenderList means the recipient's own Not-Junk click whitelisted that one mailbox (global SCL can still be 5) and SpamFilterAuthJ means authenticated-but-junked on reputation. Third-party filtering gateways (relay MX in front of a tenant) break SPF by forwarding, but an aligned DKIM signature keeps DMARC passing, so mail lands junk-but-deliverable instead of rejected; gateways also stamp bulk classification and shared-IP ratings you cannot fix. For high-value targets, a personal one-to-one email from the normal org mailbox carries none of the bulk fingerprints and bypasses that entire classification path.
• Warm up in small sector-aware batches with a human on the button. Batches of 20–30 per day, warmest recipients first; audience sector maps directly to gateway risk (private practices sit on consumer domains that accept early sends; public-sector bodies share one strictly-filtered domain — cap around 5 recipients per email domain per batch and let the batch fill from the next group). Watch three signals between batches: replies (answer them, they build reputation), bounces (a cluster means pause), and complaints (report-spam via feedback loop — the single most damaging signal, which frictionless unsubscribe exists to divert). Prefer a manual send-next-batch button plus a run sheet over automated scheduling until reputation is established and auto-pause guardrails exist.
• A stakeholder-operable send tool needs exactly this shape: a vendored template whose header/footer are not editable in the UI; compose limited to subject, preheader, constrained rich-text body and 0–2 CTA pill buttons (label plus URL restricted to https on owned domains); recipient groups from CSV with validation, case-insensitive dedupe and visible counts; [TEST]-prefixed test sends to configurable addresses before an explicit confirmed real send; a per-recipient delivery audit storing the exact rendered email; and open/click tracking disabled when the published privacy stance is no tracking.

Process

1. Build the template — fluid-hybrid card in an MSO ghost table, <img> transition strip + solid-colour text cell for the footer, programmatic header art with baked wordmark, live sublines.
2. Host the preview — stable URL on the product site with its own 'unsafe-inline'-styles CSP, noindex, no beacon.
3. Compliance pass — sender identification line, functional opt-out (page + RFC 8058 header for a tool; reply-line minimum for one-offs), any owner direction recorded in template comments.
4. Stand up the unsubscribe endpoint — confirm-on-GET page, act-on-POST, idempotent, opaque unlogged token, Referrer-Policy: no-referrer.
5. Seed test — real tool → Gmail + Outlook.com + corporate tenant; check auth verdicts, multipart, headers, single subject/greeting, independent preheader, blank-name rendering; fix and re-seed.
6. Plan the warm-up — batch sizes, sector caps, run sheet, the three watch signals, manual send button.
7. Send and watch — advance batches only after replies/bounces/complaints review; divert complaints with frictionless unsubscribe.
8. Audit — per-recipient record of the exact rendered email; placement notes (headers) for any junked seed or report.

Output format

1. Template — the HTML with structural notes (card, ghost table, footer strip)
2. Preview config — URL, route CSP, noindex confirmation
3. Compliance block — sender ID line, opt-out mechanism, recorded directions
4. Seed-test report — per-client auth verdicts, rendering checks, fixes applied, second-seed result
5. Warm-up run sheet — batches, caps, dates, watch-signal log
6. Send-tool spec (when tooling is in scope) — the constrained-compose shape above
7. Delivery audit — where the per-recipient rendered copies and placement notes live

Quality checklist

• Card uses width="100%" ... max-width:600px + MSO ghost table — no percentage max-width sizing
• No text overlaid on background-size:cover; footer is img strip + solid bgcolor cell
• Preview route has its own CSP, noindex, no beacon
• Sender identification present; opt-out functional; any removal direction recorded with date
• Unsubscribe page confirms on GET, acts on POST, idempotent, token never logged, no-referrer
• Seed test passed on Gmail + Outlook.com + corporate tenant, re-run after fixes
• Exactly one subject, one greeting; preheader independent; blank-name rendering checked
• Warm-up batched 20–30/day with per-domain caps; manual advance; replies/bounces/complaints reviewed between batches
• Tracking disabled if the published privacy stance is no tracking

Avoid

• max-width:100% sizing on nested email tables — it resolves to fixed width on phones and overflows
• Text over background-size:cover artwork — it collides at some width; use the strip + solid cell
• Judging deliverability without reading X-Microsoft-Antispam-Mailbox-Delivery — TrustedSenderList and SpamFilterAuthJ mean different fixes
• A GET-acting unsubscribe endpoint — Safe Links prefetch will silently unsubscribe recipients
• Shipping a campaign whose only test was a self-send — seed across Gmail, Outlook.com and a corporate tenant
• Automated warm-up scheduling before reputation and auto-pause guardrails exist
• Letting stakeholders edit header/footer or paste arbitrary URLs in the send tool
• Removing unsubscribe from a repeatable sending tool, whatever a one-off direction said

Example usage

"New marketing.example.com subdomain is authenticated and verified with the provider. Build the campaign template (our card design breaks on iPhones and the footer artwork slides into the text), get us Spam-Act compliant, seed-test it against Gmail and a corporate tenant, and give me the warm-up run sheet for a 400-contact list that's half public-sector."

Source: This skill is sourced from the Matrix Skills library. Learn more at the AI Agent Skills Library.