Pre-launch checks that catch the claim you didn't know you made

A regional insurer ran a Facebook ad for a critical illness policy with the headline "Guaranteed peace of mind, whatever happens." Legal had cleared the product terms. Marketing had cleared the creative. Nobody flagged that "whatever happens" reads as a promise of payout that the exclusions list contradicts. It ran eleven days before a compliance audit pulled it.

The eleven days are not the interesting part. The interesting part is that the headline sat in a dynamic creative setup, so that phrase went out paired with six images, four audiences and two landing pages, and no human being ever approved those combinations as ads. This lesson is about the version of the problem you own once volume arrives: a checklist that works on one asset and quietly fails on four hundred.

What one phrase costs once it is multiplied

Five headlines, five descriptions, four images and three audiences give you 300 servable permutations, and the platform will optimise toward whichever one performs, including the one nobody read. Reviewing outputs stops being arithmetically possible around the second campaign of the quarter. Most teams notice this only after the fact.

When the bad combination surfaces, the bill has four parts:

• Withdrawal is never one deletion. The same headline has usually been copied into search assets, an affiliate's landing pagelanding pageA standalone web page built for a single campaign goal, designed to maximise conversions by removing distractions and focusing visitors on one action.View full definition → and an aggregator's listing, and each owner has their own change queue.
• The evidence is public. Meta's Ad Library keeps active ads visible with their run dates, so a regulator, a competitor or a journalist can reconstruct what ran and for how long without asking you. The FCA publishes figures each year on promotions amended or withdrawn after its intervention, and the count runs into the thousands.
• Pulling a live campaign resets optimisation. The acquisition costacquisition costCustomer Acquisition Cost (CAC) is the total sales and marketing spend divided by the number of new customers gained in a period. It measures how efficiently you grow.View full definition → you spent three weeks grinding down goes back to where it started, and that loss lands in the same quarter as the fix.
• If the claim reached people who actually bought, the question stops being an advertising question. It becomes whether anyone purchased on the strength of the phrase, which pulls in complaints handling and potentially redress: an order of magnitude more expensive than the media that carried it.

Four failure modes that only appear at volume

The template claim. Absolute language ("guaranteed," "always," "100% covered," "whatever happens") sitting in a reusable component rather than a finished ad. Approve it once, ship it three hundred times. Life and health products are the most exposed, because emotive copy drifts toward promising outcomes the policy conditions.

The comparison whose qualifier gets cropped. Savings claims in US auto insurance typically carry a footnote defining the population they describe, along the lines of customers who switched and saved. Progressive, which sells the policies it advertises, runs that structure across its comparison messaging. On a static banner the qualifier sits under the number. Re-cut the same asset for a 9:16 vertical placement and the lower third goes; the claim survives the crop and the condition on it does not. The variant is now an unqualified comparison.

Targeting as an unwritten claim. Root, a US auto insurer that prices on telematics data from a driving test period, builds its proposition around better drivers paying less. Copy saying "you could pay less" is defensible against a general audience and much weaker aimed at a segment the model tends to price up. The creative passes review unchanged; the audience selection is what makes it misleading. No copy checklist catches this, because the offending element is a line in the campaign setup, not a word in the ad.

The feed that moves under an approved ad. "From $2/day" pulled from a live price feed is accurate on the Tuesday it was signed off. Nobody re-approves it on the Thursday the underwriting change lands. Dynamic price insertion converts a one-time approval into a standing claim you are no longer checking.

The pre-launch checklist

Use this on components and rules, not finished ads, because at volume the finished ad does not exist until the auction assembles it. This is a first pass that feeds the legal, compliance and actuarial review a sibling lesson walks through, not a replacement for it.

Claims and promises

• [ ] No absolute terms ("guaranteed," "always," "never," "100% covered") in any reusable copy slot unless contractually true in every case the product can produce
• [ ] "From £X" or "as low as $X" is achievable by a realistic segment of the targeted audience, not a cherry-picked best case, and the source of the number is a system someone owns
• [ ] Testimonials and star ratings are genuine, dated, and not selected to exclude declined-claim experiences

Comparisons

• [ ] Comparative claims name the basis (coverage level, price date, region) inside the same asset, not in a separate one
• [ ] Every qualifier survives the smallest and shortest format the asset will be cut down to
• [ ] "Award-winning" or "#1 rated" claims cite the specific award, issuer and year

Disclosures and key facts

• [ ] Link to the Insurance Product Information Document (UK and EU) or the equivalent outline of coverage is present and functional on every landing page in the rotation
• [ ] Exclusions that materially affect the advertised benefit appear in the ad, not three clicks deep
• [ ] Regulatory disclosures (FCA authorisation wording, or the relevant US state licence number) are carried per jurisdiction and per placement

Combinations, targeting and format

• [ ] Someone has read the worst combination, not the best: strongest headline against weakest disclosure against most sensitive audience
• [ ] Audience definitions checked for vulnerable-customer risk against the fair-treatment tests the brief lesson sets out, including lookalikes built from seed lists you did not construct
• [ ] Character limits are not forcing the omission of a mandatory warning; if they are, the placement is wrong, not the warning
• [ ] Claims read as a reasonable consumer in that audience would read them, not as legal intends them

For pattern-spotting what actually gets challenged, the FCA publishes enforcement case summaries at fca.org.uk/publications/enforcement-notices.

Practice: audit this sample ad

Treat the copy below as one of 48 permutations already live on Facebook, not as a single ad.

"Total Cover Travel Insurance: Guaranteed protection anywhere in the world, from just $2/day. Rated #1 by travelers. Get covered in 60 seconds."

1. "Guaranteed protection anywhere in the world" implies unconditional global cover. Travel policies exclude sanctioned territories, pre-existing conditions and high-risk activities. As a headline component, this phrase is now attached to every image and audience in the set.
2. "From just $2/day" anchors a price with no coverage tier attached. If it applies to a bare-minimum plan with a high excess, the number describes a product almost nobody buys.
3. "Rated #1 by travelers" has no source, issuer or date. Unsourced superlatives are a standing enforcement target.
4. "Get covered in 60 seconds" with no key-facts link. Speed messaging is legitimate; it cannot stand in for the disclosure.
5. The one the copy audit misses: this variant is also eligible against a 65-plus audience, where pre-existing condition exclusions bite hardest and "guaranteed protection" is at its most misleading.

A compliant rewrite: *"Total Cover Travel Insurance: emergency medical, cancellation and baggage cover for trips abroad, from $2/day on our Essentials plan (see full cover and exclusions). FCA regulated. Apply in 60 seconds."* The absolute guarantee goes, the price ties to a named tier, the disclosure link appears, and the speed claim stays because it is true on its own.

Where this fits in the launch process

Three decisions belong to you rather than to compliance.

First, what gets approved: components and join rules, or outputs. Approving outputs at one reviewer-hour per variant means 300 hours a campaign, which is why teams that insist on it end up shipping unreviewed ads under deadline pressure. Approve the atoms, then approve which atoms may combine, then cap the permutation count at a number your review capacity can actually defend.

Second, the kill-switch time. Ask the team how many hours it takes to pull one phrase from every network, affiliate and aggregator carrying it, then rehearse it once. Firms that have never measured this discover their answer is closer to eleven days than to eleven hours.

Third, who owns the price and rating feeds that populate live ads, and what happens to the campaign when the underlying number changes. That owner is usually in pricing, not marketing, and usually does not know an ad depends on them.

Version-control ad copy with the sign-off attached to each version, the way engineering tracks code review. "Who approved this claim, in which combination, and on what date" is then answerable in an audit rather than reconstructed from Slack.

🎬 [VIDEO: "How the FCA Regulates Financial Promotions" - youtube.com/results?search_query=FCA+financial+promotions+explained - search for FCA or industry-explainer content on how financial promotion rules are enforced in practice, useful as a UK-market companion to this lesson]

Key Takeaways

• At volume the reviewable unit is the component and the combination rule, not the finished ad; nobody reads 300 permutations.
• The costliest failure modes hide outside the copy: a cropped qualifier, an audience selection that makes true copy misleading, and a price feed that moves after sign-off.
• The bill for a bad variant includes withdrawal across partners, lost optimisation, public evidence in Meta's Ad Library, and, if anyone bought on the claim, complaints and redress.
• Measure how long it takes to pull one phrase from every live surface before you need to know.
• Log which version of which claim was approved, by whom and when, so an audit is a lookup rather than an archaeology project.