What brand rules should I set before AI posts for me?

Write allowed topics, tone, forbidden words, alliances/opponents, and claims you cannot make—then run drafts through those rules and approve a few by hand before any autopilot. This checklist is the practical start of brand-safe rules before Agent Mode. Rules turn “be on brand” from a vibe into checks the draft loop can actually enforce.

Checklist vs “be on brand”

“Be on brand” is useless as a system instruction. People mean different things: warmer tone, no politics, never trash a partner, never invent ROI numbers, never imply FDA-ish claims. A checklist forces those meanings into fields a reviewer (and later optional Agent Mode) can apply the same way every time.

Use rules when:

• Someone else’s name or logo sits on the post
• Legal, compliance, or a client has non-negotiables
• Multiple writers or AI drafts touch the same account
• You plan to schedule ahead or widen autonomy later

Skip the vague brand paragraph. Prefer short, testable lines: what is allowed, what is forbidden, and what must be rephrased or escalated.

Topics: what you will and will not talk about

Start with in-bounds themes—the subjects your archive and pipeline already support (product education, founder lessons, hiring, industry commentary you actually know). Then list off-limits topics explicitly: politics you refuse, competitor pile-ons, unannounced launches, personal drama, speculative M&A, medical or financial advice you are not licensed to give.

Good topic rules are boring and specific:

• Allowed: shipping updates, customer stories with permission, hiring, how-we-built notes
• Forbidden: unverified rumor, attacking named partners, commenting on live legal matters

If a draft hits a forbidden topic, rewrite or kill it—do not “tone it down” into a softer version of the same landmine.

Tone and phrases: how you sound when you are careful

Tone rules are not “sound professional.” They are constraints like: direct but not snarky; witty only if the history already uses wit; no dunking; no fake urgency; no corporate fluff that your real posts never used.

Add forbidden words and phrases—the exact strings that make legal flinch or make the brand sound like someone else:

• Absolute promises you cannot keep (“guaranteed,” “risk-free,” “cures”)
• Internal slang that should stay internal
• Competitor nicknames or meme insults
• Overused AI tells your audience already mocks

Also list preferred phrases when they matter: how you name the product, how you refer to customers, how you describe pricing without inventing discounts. Preferred language keeps drafts aligned with the archive; forbidden language stops the embarrassing exceptions.

People and brands: alliances, opponents, and name-checks

AI drafts love name-dropping. Your rules should decide who may appear and how.

Capture:

• Alliances — partners, causes, or customers you may celebrate (and any approval needed for logos or quotes)
• Opponents / sensitive parties — competitors, public figures, or institutions you will not attack or engage
• Name-check policy — when you may quote a customer, when you need permission, when you anonymize

Facebook Pages (not personal timeline) and LinkedIn company pages often amplify name-check risk because the brand—not a private person—owns the post. Keep alliances and opponents written down so a late-night draft cannot “improvise” a feud.

Claims and disclaimers: what you cannot say

Separate marketing enthusiasm from claims that create liability.

Write:

• Claims you cannot make (outcomes, rankings, certifications, exclusivity, medical/financial promises)
• Claims that need a source or disclaimer before publish
• Numbers policy — no invented stats; only figures you can defend from real data or public pricing pages

If your industry uses required disclaimers, put the short form in the rules so drafts either include them or stay away from the claim entirely. Reviewers should not be inventing compliance language under time pressure.

Five manual approvals before you trust the loop

Rules on paper are not enough. Run a short human streak:

1. Generate drafts under the new rules.
2. Approve or reject at least five posts by hand across the platforms you actually use.
3. Log what you edited: topic drift, tone, forbidden phrases, risky claims, wrong name-checks.
4. Tighten the rules where the same edit repeated.
5. Only then consider scheduled batches—or limited Agent Mode if you use it at all.

Five is a minimum proof, not a magic number. High-risk brands may want a longer clean streak. The point is evidence that rules catch what you care about before volume or autonomy rises.

Phone, web, and desktop can all run the same review loop independently with the same profile—use whichever surface you have when you are focused.

Where rules live in the weekly workflow

Rules are not a one-time onboarding form. Put them where drafting happens:

• Before brainstorm: skim allowed topics so the week’s ideas stay in bounds.
• At generate time: let the profile’s rules constrain first drafts.
• At review: score each draft against the checklist, not against “vibes.”
• After publish: if something awkward shipped, add a rule—not only a Slack rant.

Agencies should keep rules on the client profile, not in a shared team doc that nobody opens. The profile that owns the voice should own the boundaries. Phone, web, and desktop reviewers then apply the same constraints without forwarding a PDF.

When stakeholders argue about tone, settle it in the rules file. Unwritten taste does not survive a busy calendar.

Try Free

Set the checklist, import real posts, and prove the rules with manual approve on Free at https://postmimic.app. Keep Agent Mode off until the feed looks like something you would defend in a client call—then revisit brand-safe rules before Agent Mode before widening autonomy. Soft guardrails beat hoping the model “just knows.”