What an ad pre-check cannot see

It is worth being precise about what this class of tool does, because the gap is where expectations break — and in advertising, broken expectations get measured in suspended accounts.

What it catches

Patterns that published policy themes can be quoted against: "you + condition" personal-attribute framing, absolute promises and number-plus-deadline outcome claims, health and income wording that makes a product restricted-category, sensational and conspiracy framing, all-caps and exclamation density, urgency tactics, testimonials without typical-results context, affiliate framing without disclosure. These are checkable against the pasted text, so they are checked against the pasted text — deterministically, with quoted evidence for every finding, and the same result every run. The reference lists are derived from the themes Meta, Google, TikTok, and the FTC publish, snapshot-dated in the engine version rather than claimed to be live-synced.

What it cannot catch

• Clean copy and a dead offer. A variant can trip zero themes and still not be allowed to run — prohibited categories (some financial products, gambling, crypto in certain markets) are decided by what you sell, not how you phrase it. No phrasing makes a banned product submittable.
• The rest of the funnel. Platforms review the destination, not just the creative: a compliant ad pointing at a misleading landing page gets rejected for "mismatched landing page," and no copy-only check ever sees the page. Image text, video claims, and the profile's category sit in the same review, outside the paste.
• Substantiation itself. "Clinically proven" is a flag because the claim pattern is risky; whether you actually hold the study is a fact the tool cannot know. The FTC's position in its health-claims guidance is that the evidence must exist before the claim ships — a pre-check reminds you of that obligation; it cannot satisfy it.
• Account history and the reviewer. The same sentence passes on a mature advertiser's account and draws a strike on a new one. Enforcement is personalized in ways the published standards do not admit. And policies are interpreted by contractors and models with their own drift — Meta's own appeals process exists precisely because first-pass automated review is imperfect. Nobody — least of all a word-list engine — gets to promise what the final reviewer will do.

Why the boundary matters

A tool that claimed to predict approval would be claiming something nobody can deliver, and advertisers would find that out at the worst possible moment — mid-enforcement. Stating the limit up front is what lets the check get trusted for the things it genuinely knows: that the sentence matches a published rejection theme, that the theme is the kind that escalates, and that the fix is a rewrite, not a resubmit dice-roll. The honest claim is narrower and more useful than the fake one: a scored pre-check moves the odds, per sentence, before submission. The human reviewer stays exactly where an ad check should leave them — holding the context the machine cannot see: the offer, the evidence, the funnel, the account.

How to use the boundary productively

A pre-check is the first layer of a review stack, not the whole stack. Layered deliberately, each layer covers what the one before it cannot: the mechanical scan catches published rejection themes in the copy at paste speed; the creative review checks that the image and video text carry no claim the copy omits — platforms increasingly read both; the landing-page pass confirms the offer matches the ad, because mismatch is its own documented disapproval theme on Google and Meta; the legal pass owns regulated claims, substantiation files, and income or health evidence. Buying these layers separately is how most teams already operate — the mistake is treating any single layer as covering the others. The scan's job is to make sure you never bring a known, mechanical, avoidable problem into the expensive layers.

The correct mental model

Treat a pre-check the way pilots treat the before-takeoff checklist. It does not make the flight safe; it catches, mechanically and repeatably, the handful of items whose omission turns a routine departure into an incident. The judgment — is this runway the right call today — remains entirely human. An ad pre-check is a rejection-theme scan with memory: every variant gets the same published patterns applied, every time, at paste speed, and every finding cites the sentence it came from. It is decision support, not a decision. Anyone selling the decision is selling something that does not exist. The teams that get the most from a bounded tool keep the boundary visible: scan early, escalate what the scan flags, and hold the judgment the scan deliberately declines to make.

Heuristic pre-check only — not platform approval and not legal advice. Final review decisions belong to each ad platform; regulated-claim questions belong to a lawyer.