never say rules brand safety ugc

Automated Brand Safety and Never-Say Guardrails for UGC

Never say rules brand safety UGC: build claims libraries, script sign-off gates and a 7-day compliance check that catches violations before launch.

Your brief has a never-say list. The creator read it on a phone, in a hurry, and still opened the video with "this cures my acne in three days." The rule existed. The process did not.

Brand safety in creator work fails at the handoff, not at the policy. A PDF attached to a gig brief is a hope. A guardrail is a sequence of checks that runs before the shoot, before the edit locks, and before money moves.

Why the brief alone cannot hold

Creators work from phones between other jobs. A 12-page brand safety appendix gets skimmed, and the two sentences that matter get lost between the usage rights clause and the payment terms.

The second problem is language drift. A creator who normally says "this thing clears my face" will not notice that "clears" is a medical claim in your category. They are not being careless. They are speaking the way they speak, and your rule was written in compliance English.

The fix is to move the rule out of the appendix and into three places the creator cannot avoid: the gig's first screen, a script outline they must submit before shooting, and a draft check that names the exact clause when something fails.

Build a claims library before you build the brief

A never-say list is half a tool. The other half is the approved replacement, because a creator told only what they cannot say will improvise something worse.

Keep a single document per brand with four columns: the claim you want, the approved wording, the banned variants, and the evidence you hold. Product teams update it, compliance signs it, and every brief pulls from it.

ColumnExample for a skincare brandWhy it exists
ClaimEvens out dark spots over timeThe thing marketing wants to say
Approved wording"My dark spots looked softer after about six weeks"A sentence a creator can actually say
Banned variantscure, clear, remove, treat, heal, permanentTerms that turn a cosmetic into a drug claim
Evidence12-week consumer study, 43 participantsWhat you show a regulator

That table does more work than a page of instructions because it gives the creator a sentence to use. In practice, the rejects that come back are almost always a creator reaching for a stronger verb because the approved line felt weak.

The three gates that catch almost everything

Run the campaign through three gates. Each one is cheap, and each one catches a different class of failure.

Gate one is the script outline, submitted as text or a voice note before the shoot. You are checking structure, not performance: does the hook make a claim, does the demo show the product, does the closing line include the disclosure. A five-minute read here saves a full reshoot. This is also where you catch a creator who plans to say "guaranteed" in the first three seconds.

Gate two is the draft file. This is the expensive check, so give it a checklist so the review does not turn into an opinion. Listen once with the sound on and no video, which forces you to hear claims without being distracted by the edit. Then watch once with the sound off, which is how most of your audience will see it, and read every text overlay.

Gate three is the disclosure and metadata check. Confirm the paid partnership label is on, the caption carries the disclosure line, and the file name matches the version you approved. A correct video with a missing disclosure is still a compliance problem.

GateWhenWho runs itWhat it catches
Claims libraryBefore the briefBrand and complianceWrong or unsupported promises
Script outlineBefore the shootBrand reviewerHook claims, missing disclosure, wrong product
Audio passOn deliveryBrand reviewerSpoken claims, competitor mentions, tonal risk
Muted passOn deliveryBrand reviewerOn-screen text claims, missing captions
Metadata passBefore approvalBrand reviewerDisclosure, tagging, file version
ArchiveAfter approvalBrandEvidence trail for a later audit

Where automation helps and where it lies to you

Transcription tools are good enough to search a draft for banned terms in seconds. Run the transcript against your banned list and you will catch the obvious violations before a human watches.

Optical character recognition on the frames catches text overlays that a transcript misses, which matters because half of Nigerian creator content carries the claim on screen instead of in the voice.

Both tools fail in predictable ways. Sarcasm reads as a claim. Pidgin phrasing slips past an English keyword list, so "e nor dey fail" never triggers a check for "guaranteed". Implied claims pass every filter, and a creator holding up a clear face and saying "you see the difference" makes a visual claim with no banned word in sight.

Treat the machine pass as a filter that reduces the human review load, and keep a person on the final watch. The reviewer's job is to ask one question: if a regulator watched this once, what would they object to?

The 7-day window is a compliance deadline, not a courtesy

When a creator delivers on ugc.ng, the brand gets a 7-day review window. Use it as a scheduled task on day one, not a thing you get to on Friday.

Day one is the audio pass and the transcript search. Day two is the muted pass and the overlay check. Day three is the decision: approve, or reject with a specific clause and a specific fix. If nobody acts by day seven, the submission auto-approves and the creator is paid the next business day in Naira.

That auto-approval is fair to creators and dangerous for brands that treat review as optional. Put a named owner on the gig before it goes live, and give that person calendar time. The workflow detail is worth reading in full at the revision policy and 7-day approval window.

Rejections that actually get fixed

A rejection that says "not on brand" produces a second draft with the same problem. A rejection that says "line 14 says 'clears acne', which is banned variant 3 in the claims library, replace with the approved wording in row 2" gets fixed in one take.

Write rejections in that shape: timestamp, the exact words, the clause, the approved replacement. Then let the creator re-record only the affected segment. Full reshoots are for structural failures, such as shooting the wrong product or missing the demo entirely.

Keep a log of every rejection across campaigns. After twenty gigs you will see a pattern, usually one claim that creators keep reaching for because it sounds natural. That claim belongs in the brief's first paragraph next time, not buried in an appendix.

Guardrails that survive a scale-up

The process above works for one campaign. To run ten at once, add two things: a single owner for the claims library, and a shared rejection log that every reviewer can read. Without those, each campaign manager invents their own rules and the brand's standards drift.

For regulated categories, add the sector rules to the library itself, not to each brief. Fintech language belongs in the claims sheet under CBN and FCCPC compliance for UGC, and health claims belong under NAFDAC rules for skincare and supplements. One source, many briefs.

Then brief the work with the rules already embedded. A gig post on ugc.ng has room for the approved wording, the banned variants and the disclosure line in the first screen of text, which is the only part every creator reads. Put them there, and the check on delivery becomes a confirmation instead of an argument.