How this card works under Option B. This is not a card a human runs at review time. It is a pre-check: the AI CMO runs it automatically on every artefact a specialist produces, before the review pack is assembled. The card's result (7/7 clean, or the specific box(es) failed with a one-line fix note) becomes item 2 in the review pack — the six-item WhatsApp message the four humans see. Humans read the artefact and the card result together, then sign the domain lines that apply. The AI does not vote. Humans decide.
Why humans still see it. The card is printed and kept where the humans read WhatsApp packs, so that when the AI CMO reports "Box 3 fail: 'debt wipe' appears in line 4", every human recognises the box number without leaving the thread. It also lets any human catch something the AI's pre-check missed — the card is a floor, not a ceiling.
The rule. All seven boxes must tick before the AI CMO forwards the pack for sign-off. If any box fails, the AI CMO returns the artefact to the specialist with the specific box number and fail note, and the pack is not posted to WhatsApp until the box clears. This keeps the humans' review focused on judgement (domain-line sign-off) rather than mechanical compliance-craft (already caught by the card).
Companion pages. /marketing-operators-manual (the shared-gate playbook · how the pack physically moves) · /marketing-escalation-map (the domain-line assignment table · which artefact categories trigger which sign-off lines).
Contents
1. The card
LWD Marketing — AI CMO Pre-Check
Run automatically by the AI CMO on every specialist-produced artefact. Result becomes item 2 of the review pack the 4 humans read on WhatsApp. All seven must tick before the pack is forwarded for sign-off.
/complaints) and "privacy" (/privacy-policy). Missing one = fail. Even for small artefacts.
/complaints) and "privacy" (linking to /privacy-policy) both appear on the artefact. For inline formats (short LinkedIn posts, tweets) the links appear on the landing page the post drives to — check the destination page too.
2. Pre-check outcomes and how they land in the review pack
The four outcomes the AI CMO records after running the card. These are pre-check states — not human verdicts. The AI is not deciding whether to ship; it is deciding whether the pack is ready for humans to see.
The AI CMO does not use words like APPROVE, BLOCK, or ESCALATE on the card. Those are human verbs, reserved for the humans' sign-offs on domain lines. The AI's four possible outcomes are CLEAN, FIX-INLINE, RETURN-TO-SPECIALIST, and CLEAN+FLAG. They describe what the AI did, not what the artefact deserves. This discipline keeps the AI in its lane — producing and pre-checking, never voting.
3. What each box actually checks (detailed)
Box 01 — Four compliance markers + complaints + privacy
The mediation disclosure sentence is the one most often forgotten on short-form artefacts: "LWD holds Australian Credit Licence 387398 for the mediation function; beneficiary casework is delivered under that licence with NCCP-compliant conduct." For an artefact where space is genuinely tight (a LinkedIn post's character limit, an ad's headline+description budget), the markers appear on the destination landing page. The AI CMO checks the destination too when a Box 01 pass depends on it, and notes "Box 01 passes on destination page" in the review pack so humans know where to look if they want to verify.
Box 02 — Y1 numbers verbatim
The audited set is fixed. Rounding "$1.82M" to "$1.8M" is a fail. Saying "over 30 households" instead of "31 households" is a fail (it loses the exact number that is the strongest part of the story). Extrapolating "we will serve 300 households next year" is a fail (unaudited future promise). If a Y1 number appears in donor-facing copy, the outcome caveat attaches: "Y1 outcomes describe past pilot results for 31 households and are not a guarantee of individual future outcomes." Because Y1-number use also triggers Laurence's Evidence & Brand sign-off line, a Box 02 pass is necessary but not sufficient — the human sign-off still lands.
Box 03 — No "debt wipe" language
ASIC RG 96 has strict conduct rules on credit-adjacent language. "Debt wipe" implies the debt is extinguished by LWD's action, which is not what happens — LWD funds mediation between the beneficiary and their creditor, resulting in a settlement that the beneficiary can meet. That's a materially different service and describing it as "wipe" would be misleading. Similarly banned by RG 96: "debt relief" as a service description (LWD is not a debt relief provider).
Box 04 — No tax-deductible / DGR / PBI (pre-DGR)
DGR is pending as of Sprint 54, with registrations expected to complete within ~2 weeks. Not even hinting at it — not "gifts will be tax-deductible once DGR is granted", not "tax-deductibility coming soon". After DGR is granted, this box flips: it becomes a check that deductibility copy is present where appropriate and correct in its wording (specific category of DGR endorsement, gift-of-$2-or-more threshold, receipt-at-time-of-gift language). Sprint 56 (DGR-launch pack) will publish the post-DGR wording; between now and then, Box 04 stays in "pre-DGR: none allowed" mode.
Box 05 — Real beneficiary imagery / quote / naming
The corrected imagery rule (Sprint 53): stock imagery with faces IS allowed and often the warmer, better choice for charity communications. What's forbidden without written consent + all-four-human sign-off (constitutional hard case) is imagery of real LWD beneficiaries. Also caught by this box: pairing a stock image with a real beneficiary's real details (a real name, real postcode, real case circumstance) — that combination reads as a real beneficiary even though the image itself is stock. Also caught: hero-framing of carers, clinicians, LWD staff, or the founders themselves.
Box 06 — Complaints and privacy links
Discrete from Box 01 because the compliance markers can be present without the links being clickable. The AI CMO checks that /complaints and /privacy-policy resolve, are correctly anchored, and appear in the artefact or its destination. FIA Code and Privacy Act 1988 both require accessible complaints and privacy discovery paths.
Box 07 — Voice
The hardest box because it is a judgement, not a mechanical check. The AI CMO's Box 07 pass is provisional — it catches the obvious voice failures (exclamation marks, urgency verbs, hero-framing, fundraising-mail-out register) but the final voice call sits with Corrina's Marketing-Craft sign-off line on WhatsApp. The heuristic the AI applies: if the artefact would fit as a paragraph in a peer-reviewed evidence summary, it passes; if it reads like a fundraising-charity mail-out, it fails. LWD's voice is qualified, quiet, evidence-led. It is the voice of a licensed practitioner, not the voice of a fundraiser. If the AI is uncertain on Box 07, it records "Box 07 passes provisionally — flagging for Marketing-Craft sign-off attention" as a CLEAN+FLAG outcome.
4. Common failures and their fixes
These are the patterns the AI CMO catches most often. Each row records the failure, which box catches it, and the fix pattern the AI applies when returning the artefact to the specialist. Humans reading the review pack recognise the fix pattern immediately from this table.
| Failure | Box | Fix pattern the AI applies (or requests from the specialist) |
|---|---|---|
| "$1.8M" or "over $1.8M" | 02 | Rewrite to exact "$1.82M". Never round. |
| "Wipe your debt" / "debt-free future" | 03 | Rewrite to "mediation and settlement" or "resolution". |
| "Tax-time giving" / "your gift will help" | 04, 07 | Remove tax-deductibility angle (Box 04) and remove urgency verb (Box 07). |
| Stock image + specific case detail in caption | 05 | Either the stock image with generic caption, or the case detail rewritten as illustrative composite (labelled). |
| "Save a life" / "One dollar could" | 07 | Rewrite to evidence-led: "The Y1 pilot resolved $1.82M of debt for 31 households at 87c per dollar." |
| Missing ACL 387398 on a short-form ad | 01 | Add to destination landing page, and confirm the ad links to the destination. Report "Box 01 passes on destination page" in the review pack. |
| Exclamation marks in donor copy | 07 | Remove. LWD's voice never uses exclamation marks in donor-facing text. |
| Y1 number without the outcome caveat in donor copy | 02 | Attach the outcome caveat sentence in the same paragraph or in a footnote-style disclosure on the same page. |
| Founder hero-framing ("our incredible founders") | 05 | Rewrite as role-framed: "LWD's founders launched the pilot in [date]" — not "our incredible founders launched the pilot". |
| "Client" used for beneficiary | 07 | Rewrite to "beneficiary" (LWD's constitutional term). Clients pay; beneficiaries receive. |
5. What this checklist does NOT cover — where domain-line judgement takes over
This checklist is the AI CMO's pre-check for mechanical compliance-craft. It is not a substitute for the four humans' domain-line judgement on WhatsApp. Specifically, this card does NOT check:
- Whether an outcome claim is factually audited. Box 02 checks the numbers are verbatim from the canonical set. It does not check whether the underlying audit is correct — that's Laurence's Evidence & Brand sign-off line.
- Whether a donor message serves LWD's philanthropic strategy. Box 07 checks voice. It does not check whether Lisa would want that message sent to that funder in that moment — that's Lisa's Philanthropic sign-off line.
- Whether a $ figure is defensible in a regulatory sense. Box 02 checks it matches the audited number. Whether an admin-ratio claim is compliant with ACNC guidance is Carla's Financial & Regulatory sign-off line.
- Whether the artefact voice fits the channel, calendar, and campaign context. Box 07 is the AI's provisional voice check. The final voice-and-craft judgement is Corrina's Marketing-Craft sign-off line.
- Whether the artefact should exist at all. Strategic decisions (a whole new campaign concept, a new audience) are made in the Monday-plan session before the AI is briefed to produce anything; if a produced artefact seems to be outside the week's approved plan, humans flag it in the WhatsApp thread and the pack is held.
- Whether the AI's output is factually correct beyond the LWD-specific checks. AI can invent statistics, misquote external sources, or hallucinate journalist names. Humans spot-check as part of their domain-line sign-off. This is a general professional-editing responsibility not caught by the 7-box card.
In short: the 7-box card is the minimum a pack must clear before humans see it. It does not replace domain-line judgement. It catches the LWD-specific mechanical failures reliably so that the humans' WhatsApp review can move at speed and focus on the judgement that only humans can make.
6. Change control
This card is versioned by Sprint number and reviewed monthly by all 4 humans in the WhatsApp group (one message each). Two triggers force an out-of-cycle review:
- DGR granted. Box 04 flips from "no tax-deductible copy" to "deductibility copy present and correct where appropriate". Reviewed same-day.
- A red-flag failure shipped past quorum. The card is retro-audited, the failing pattern gets a new box or a strengthened box, and the update is highlighted in the Friday CMO Summary posted to WhatsApp.
Card updates are noted in git with the reason. Every version increments the "Sprint XX · vX" tag in the card header. When the printed card is refreshed, the previous version is recycled — not kept — to avoid mixed generations circulating.