Role Playbooks4 min read

Build a Claims Register Before AI Writes the Campaign

A practical role playbook for marketing managers who want AI-assisted campaign copy without detaching product claims from evidence, scope and human approval.

Bokili EditorialPublished August 26, 2026· Verified August 26, 2026

Marketing claims register workflow from evidence cards to approved claims, AI-assisted drafting and human sign-off

Generative AI can turn a thin brief into fluent campaign copy in seconds. That speed creates a new control problem for marketing managers: a sentence can sound polished while strengthening a product claim, dropping a qualification or separating a number from the evidence that supports it. The safest place to solve that problem is before the prompt. Build an approved claims register, then let the tool draft only from that register.

This is not a ban on creative work. It is a clean division of labour. People decide what the organisation can substantiate, where a claim applies and who owns it. AI may then explore structure, tone and variants inside those boundaries. The US Federal Trade Commission says advertisers should have a reasonable basis for objective claims before they are disseminated. UK CAP guidance likewise requires documentary evidence before publication for claims consumers are likely to regard as objective.

The marketing manager owns the evidence boundary

A campaign brief usually mixes several kinds of language: verified product facts, qualified facts, aspirations, opinions and creative concepts. If they are passed to an AI system as one undifferentiated document, the model may smooth away the distinction. “Designed to reduce packaging” can become “reduces packaging”. A result from one test condition can become a universal promise. A supplier statement can sound like an independently verified fact.

The manager’s first task is therefore not to write a better prompt. It is to separate usable claims from material that needs evidence, qualification or rejection. This complements a prompt data boundary: one controls what information may enter the tool; the claims register controls what factual language may leave it.

The five-field claims register

Approved wording

Write the strongest exact wording the evidence supports. Treat it as a ceiling, not a suggestion the model may improve.

Evidence

Link the source, test, certificate or approved record. Name what it proves, not merely where the file lives.

Scope and qualifier

Record the product, market, period, test conditions and any qualification that must travel with the claim.

Channel and audience

State where the claim may appear and whether space, format or audience changes the likely interpretation.

Owner and review date

Name the person who can approve changes and the date or event that triggers re-checking the evidence.

Worked example: a reusable bottle campaign

Imagine a fictional bottle launch. The brief contains “keeps drinks cold for 12 hours”, “made with 70% recycled stainless steel” and “planet-friendly”. None of those examples is presented here as a real product fact. They show how a manager would test different claim types.

The first claim needs the relevant test record and its conditions. If the evidence covers one bottle size at a defined starting temperature, those limits belong in the register. The second needs a supplier certificate or another approved record that makes clear whether 70% applies to the bottle body, all steel parts or the whole product by weight. “Planet-friendly” is too broad to release without a specific, supportable meaning. The team should replace it with a narrower verified statement or leave it out.

Reading is a start. Practice makes it stick.

Start learning

	Loose brief	Claims-register brief
Claim	Keeps drinks cold for 12 hours	Use only the approved 12-hour wording and its stated test conditions
Evidence	See product folder	Link the exact test record and identify what it supports
Qualification	Add if needed	Keep the required qualifier beside every version
AI instruction	Make the copy stronger	Vary tone and structure without strengthening factual claims

Draft from the register, then review claims before prose

Give the AI system the audience, channel, objective, tone and only the approved claims. Tell it not to introduce performance, comparison, environmental, price or endorsement claims. Ask it to mark any sentence it cannot produce without an additional fact. This turns uncertainty into a visible review item instead of hidden invention.

When the draft returns, review in two passes. In the claims pass, compare every objective or implied claim with the register. Check numbers, comparisons, superlatives, demonstrations, captions and qualifications. CAP’s rules make clear that qualifications may clarify a claim but must not contradict it or hide material information. In the prose pass, improve clarity, rhythm and fit for the channel. Keeping those passes separate prevents attractive wording from distracting the reviewer from evidence.

Campaign claims gate

Every objective or implied claim appears in the register.
Each claim has evidence that supports the wording actually used.
Scope, conditions and material qualifications remain attached.
The draft adds no unsupported comparison, superlative or environmental benefit.
A named owner has approved the final claim set for this channel.
The final hand-off carries evidence links, review status and stop conditions.

That last item connects the register to a stronger AI workflow hand-off. The next reviewer should not have to reconstruct which lines were checked. A compact record of evidence, ownership and exceptions should travel with the asset. It also helps to define done before delegating work to AI: the approved claim set becomes part of the completion criteria.

Ten-minute claims-register test

Choose one unpublished campaign draft and underline every sentence a customer could read as an objective claim.
For each claim, write the exact evidence source and what it proves.
Add any scope, condition or qualification that must stay with the wording.
Mark claims with missing or mismatched evidence as STOP; do not ask AI to repair them.
Give the remaining approved claims to the AI tool and request two tone variants that may not strengthen them.
Compare the output with the register before reviewing style.

The register is a creative constraint worth keeping

A claims register does not guarantee compliant advertising, and requirements differ by jurisdiction, product and channel. Legal or specialist review may still be necessary. What it does provide is a repeatable operational control: evidence is checked before generation, qualifications are visible during drafting and accountability survives the hand-off. That lets a marketing team use AI for genuine creative range without asking fluency to stand in for proof.