Why Was My Facebook Ad Rejected? The Reasons AI-Written Copy Gets Flagged

Short answer: most rejections are not about your product. They are about a sentence in your ad text making a claim the platform will not let you make — and AI ad writers produce those sentences constantly, because inventing specifics is what they are good at.

We build an ad generator, so we had to solve this before shipping. What follows is what actually gets copy flagged, in the order it happens.

The claims that get an ad rejected

Meta, Google and TikTok all police roughly the same categories. The wording differs; the substance does not.

Invented social proof. "Over 10,000 happy customers." "Rated #1 by dermatologists." "As seen on Good Morning America." If it is not true, it is a policy violation and often a legal one. This is the single most common thing an AI ad writer will hand you unprompted, because the training data is full of ads that say it.

Personal attributes. Ad copy may not imply it knows something personal about the reader. "Struggling with hair loss?" is the classic — it asserts a health condition about the person seeing it. The safe form is about the product: "A shampoo formulated for thinning hair."

Before-and-after and unrealistic outcomes. Especially health, fitness, finance. "Lose 20lbs in 30 days" is rejected on sight. So is "Double your income."

Superlatives you cannot substantiate. "The best," "the strongest," "#1" — permitted only with a citable source, and an AI will not have one.

Fake urgency. "Only 3 left!" on a product with unlimited stock, or a countdown that resets on refresh. Platforms increasingly treat this as deceptive.

Currency and pricing claims that do not match the landing page. If the ad says $19 and the page says $29, expect a rejection and, on repeat, an account review.

Why AI-generated copy walks into this

An ad generator is optimising for persuasive text. Every one of the above is persuasive. Left alone, a language model will write "trusted by thousands of happy customers" for a store that launched last Tuesday — not from malice, but because that phrasing appears in the millions of ads it learned from.

The failure is worse than a rejection, because a fabricated claim that does get through is a claim your business is now publicly making. A rejected ad costs you an afternoon. An approved false one costs more.

This is why our own generator carries a validator that sits between the model and the merchant, refusing output containing fabricated social proof, invented ratings, unsubstantiated superlatives, and manufactured scarcity. It is not a nicety. An ad tool without one is handing you a liability and calling it a feature.

One honest limitation worth stating, because it applies to every tool in this category: those checks are pattern-based, and patterns are language-specific. A validator written against English phrasing will not catch the same fabrication written in Spanish or German. If you run ads in more than one language, the checking has to exist in each of them — ours does it by restating the banned-claims rules inside the generation prompt for that language, rather than relying on an English filter that cannot see the problem.

The rejections that are not about your copy

Not everything is the text. In rough order of how often they bite:

Landing page mismatch. The ad promises something the page does not deliver, the page has no visible contact information, or it is slow enough to time out during review. Reviewers do open the page.

Restricted categories. Supplements, CBD, weapons-adjacent, adult, gambling, some financial products. These need extra permissions, and no rewording of the ad fixes them.

Text in the image. Meta no longer hard-rejects over the old 20% rule, but heavy text still suppresses delivery. If your creative is a screenshot of a product page, expect it to underperform even when approved.

A new ad account with no history. New accounts are reviewed more aggressively. A first ad rejected is not necessarily a policy problem — it is sometimes just a new account being treated as new.

What to do when it happens

• Read the specific policy cited. The rejection names one. It is usually accurate, and the fix is often a single sentence.
• Change the copy, not the targeting. Most people re-submit the same text with a different audience and get rejected again.
• Remove every claim you cannot evidence. If you cannot produce the source in ten seconds, cut it. "Loved by thousands" goes; "Free returns for 30 days" stays, because you can point at your own policy page.
• Appeal once, calmly, if you believe it is wrong. Automated review does make mistakes, particularly on health-adjacent wording. One appeal, not five.
• Do not repeatedly resubmit near-identical ads. Repeated rejections against one account attract account-level review, which is a much worse problem than a rejected ad.

The rewrite that usually works

Take out the claim, keep the specifics you can actually stand behind.

Before: "The #1 rated phone mount — over 10,000 happy customers can't be wrong! Only 3 left!"

After: "A phone mount that holds through speed bumps. 360° swivel, fits vents and dashboards, free returns for 30 days."

The second one is more likely to be approved, and it is more likely to sell, because it says something a person can picture. The first says nothing about the product at all.

The short version

If an ad is rejected, assume first that a sentence in it claims something you cannot prove. That is the most common cause, it is the one AI writing tools introduce most reliably, and it is the one you can fix yourself in about two minutes — by deleting the sentence.