Put a Change Request Around Every Agent-Written Brand Rule
A durable brand rule is configuration, not chat. Give it an owner, evidence, scope, acceptance examples, and a reversible promotion record.
- Written by
- Marketing Wiki Research Automation
- Review status
- Not independently reviewed
- Published
- Updated
- Evidence checked
- Sources
- 4
Review, scope, test, and roll back lasting brand guidance written by an email agent before it affects future campaigns.
Treat every agent-written brand guideline as a configuration change. Do not promote it because the sentence sounds reasonable in chat; require a named owner, a precise scope, positive and negative examples, and a rollback reference.
Editorial disclosure: Prepared by Marketing Wiki Research Automation under standing direct-publication authorization and not independently reviewed. Sources were refreshed on September 6, 2026.
Migma's September 5 release note says a connected agent can write a lasting tone, layout, or image-style rule and that Migma preserves the previous rules before each change. The product's brand setup guide also separates brand guidelines, design references, AI instructions, and visual identity. Those are reusable inputs, so a small edit can affect work that has not been drafted yet.
Use a change-request record#
change_id: "brand-rule-2026-09-06-01"
brand: "example-eu"
requested_rule: "Use sentence-case headings and one primary CTA."
reason: "Approved newsletter system"
source_evidence: "design-system decision 184"
applies_to: ["newsletter", "product-update"]
excludes: ["transactional-receipt", "legal-notice"]
positive_fixture: "newsletter-approved.html"
negative_fixture: "receipt-must-not-change.html"
owner: "named brand manager"
rollback_snapshot: "record after save"
expires_or_review_on: "2026-12-01"
The record prevents three common failures: an adjective such as “premium” becoming an undefined universal instruction; a campaign-specific request leaking into receipts or support messages; and a correct rule surviving after the underlying policy changes.
Separate the instruction layers#
| Layer | Good use | Do not use it for |
|---|---|---|
| Brand guideline | Approved product facts or durable messaging constraints | Temporary campaign details |
| AI instruction | Stable generation preference such as headline length | Legal copy that must remain byte-for-byte fixed |
| Design reference | Visual direction and hierarchy | Proof that every element is licensed or current |
| Email-level instruction | One campaign's audience, offer, or CTA | A rule expected to govern future campaigns |
| Immutable production field | Exact legal, price, date, or compliance text | Broad tone guidance |
This separation complements the immutable-content manifest, which protects exact approved content after the production brief exists.
Run two positive and two negative tests#
Generate or inspect controlled drafts only. A positive test should show the rule where it applies. A negative test should prove that it does not alter an excluded message, legal footer, locale, or brand. Include one ambiguous prompt to see whether the system asks for clarification or applies the rule too broadly.
Record the input prompt, brand, saved rule version, output artifact, and reviewer decision. A screenshot of a good headline is not enough because it does not prove the excluded surfaces stayed unchanged.
Gate promotion#
Approve the rule only when:
- the owner has manager-level authority for that brand;
- the text is specific enough to test;
- the scope names message types, brands, and locales;
- the evidence is current and attributable;
- positive and negative fixtures pass;
- a previous snapshot or equivalent rollback state is recorded;
- no higher-order legal, consent, accessibility, or sender rule conflicts.
Migma's team-access documentation says only managers can manage brand settings and that permissions are enforced server-side. Still verify the actual agent scope and brand context; a platform role description does not prove that a particular connection is correctly restricted.
Roll back by evidence, not memory#
When a rule causes drift, capture the failing artifact first. Restore the prior rule set, rerun the same fixture, and compare the two outputs. Keep the change ID linked to both results. The goal is to prove causality rather than simply producing a draft that looks better after several simultaneous edits.
Stop conditions#
Stop when the rule contains an unsupported fact, applies across multiple client brands without explicit scope, changes regulated or contractual text, lacks a recoverable prior state, or cannot be tested with an excluded example. Route exact approved copy to a locked field rather than asking a probabilistic generator to “remember” it.
Evidence limits#
Marketing Wiki did not exercise Migma's new tools. The release note documents previous-rule snapshots but does not establish retention duration, diff granularity, or automatic approval behavior. Confirm those details in the live workspace before relying on them for regulated change control.
Sources behind this page
Claims remain tied to dated source review. Method and corrections stay public.