AI-generated ads: the policy traps that get ads rejected, and a checklist to run before you submit

An ad policy is the rulebook an ad platform uses to approve or reject your ad. Meta (the company behind Facebook and Instagram) and Google Ads both publish one. Neither page has a rule that says "no AI." Marketing blogs say TikTok rejects undisclosed AI in ads, but TikTok's ad policy was not confirmed, so check it yourself. What they have are general rules about misleading people, using other people's brands, and showing the thing you actually sell. AI tools can break all of those without you noticing, because the tool draws what the prompt describes, whether or not it is true.

This page is the traps list. It is about what you claim and show. The related question of whether to add an AI label is in the platform label rules page, so I keep labels brief here. For the rights side, see can you use AI video in paid ads for clients. This is information, not legal advice, and pages change, so read the current policy for your platform and your product category.

What the platforms actually say

Three official pages shape most of this advice.

• Google Ads, "Misrepresentation" policy. It says you cannot manipulate media to deceive, defraud or mislead. It bars presenting an improbable result as the likely outcome. It bars implying backing from a brand or organisation you do not have. It also says serious violations can lead to account suspension without prior warning. It has no AI-specific wording. A tool being involved changes nothing about the test.
• Meta Advertising Standards. The index lists, among others, Prohibited Commercial Practices (described as content that may deceive or mislead people), Privacy Violations and Personal Attributes, and Third-Party Intellectual Property Infringement. The index has no AI-specific entry and no before-and-after entry. Confirm any claim about a specific Meta rule on its own page, including how it treats an ad that does not match the business.
• The FTC (the US consumer protection regulator) on reviews and endorsements. Its consumer reviews and testimonials rule has been in effect since 21 October 2024. The FTC's Q&A says it does not ban AI avatars as such. It targets testimonials that are false, and it says using a celebrity's likeness without permission to suggest an endorsement violates the rule. The FTC says its staff guidance is not a safe harbour (a guarantee of safety).

Notice the pattern. Every one of these is about truth and permission, not technology. If you hold that in mind, most AI ad rejections become predictable.

A rejection map: symptom, likely cause, fix

Use this when an ad comes back with a rejection or a "limited" status. The exact wording of the notice varies, so match on the meaning, and read the full reason in your ads account.

What the notice suggests	Likely AI-related cause	Fix
Misleading or unreliable claim	The ad shows or says a result the product cannot deliver, such as a dramatic transformation in days	Replace it with a result you can prove. Keep the evidence in a file
Manipulated or deceptive media	An AI image presented as a real photo of a result, a place or a customer	Show the real product. Make clear that the image is an illustration, or remove it
Personal attributes	Copy that implies you know something about the viewer, such as a health condition, money trouble or identity	Rewrite to address the offer, not the person. For example, "Tips for managing blood sugar" is safer than "Are you diabetic?" as a pattern, but check the category rules
Third-party intellectual property	A logo, a famous face, a character or a competitor's style that the model copied	Remove it. See AI content with brand logos, celebrities and competitors
Ad does not represent the business	The product on screen is a generated version that does not match what ships	Use a real product photo as the reference and compare the output to the item
Landing page mismatch	The ad promises an offer, price or look the page does not show	Match the page to the ad before you resubmit
Format or spec error	An AI export in the wrong shape, such as a 16:9 clip pushed into a Reels slot	Export to the placement. Meta Instagram Reels ads: 9:16, 1440x2560, up to 15 minutes, 4 GB, with 14% top, 35% bottom and 6% side safe zones. Meta Feed video: 4:5, 1440x1800, MP4 or MOV, up to 4 GB. Check Ads Manager for the current numbers
Disapproved with no clear reason	Garbled text, a distorted logo or a hand with extra fingers triggers a review	Zoom to 100% on every frame and fix or re-render

Claims, results and before-and-after

This is the biggest trap. An AI tool will gladly draw a "before" with grey skin and an "after" with glowing skin, or a cluttered garage that becomes spotless. If the product did not do that, you have invented a result. Google's policy says presenting an improbable result as the likely outcome is not allowed, and the FTC's endorsement guidance covers typical results. For the image side, the before-and-after page explains what is safe to make from two real photos.

Some categories have stricter rules than the general ones. Health, weight loss, cosmetic procedures, finance and housing all draw extra attention on both Meta and Google. Meta's advertising standards index lists Health and Wellness, Financial and Insurance Products and Services, and Drugs and Pharmaceuticals as separate restricted categories. If your product is in one of these, read that category's page first, because it can be stricter than anything on this page.

A simple rule for claims:

1. Write the claim in one sentence.
2. Write where you can prove it: a test, a study, your own customer data with permission.
3. If the proof is not in your hands, cut the claim or soften it to what you can show.
4. If the image shows more than the claim says, cut the image.

Keep the proof. If you appeal, a screenshot of your substantiation does more than an argument.

Fake people, testimonials and endorsements

An AI spokesperson is not forbidden. An AI person who says "I lost 10 kilos" or "I run a cafe and this saved me hours" is a made-up customer, and that is a testimonial problem, not a technology problem. The FTC's reviews rule bans fake reviews and testimonials, and its Q&A notes the rule does not stop AI avatars in general, only the false statements. Two safe patterns are an AI presenter who explains the product in the brand's own voice with no personal claims, and a real customer's genuine review, read or shown with permission.

Two more traps. A real person's face or voice needs their written consent, which is covered in the consent checklist. And an AI person who looks like a known figure can look like a false endorsement. The longer treatment is in the fake testimonials page. For UGC (user-generated content, meaning ads that look like they were filmed by a customer), see UGC-style ads: real creators or AI avatars, which sets out a hybrid that protects the agency.

Logos, brands and look-alikes

AI models often reproduce things they have seen, including logos, famous characters and the look of a well-known ad. Meta's standards list third-party intellectual property infringement, and Google's policy bars implying backing from a brand you lack. Practical steps:

• Do not prompt with another company's name, product or a celebrity's name.
• Check the background of every frame for stray logos, such as a drink can or a shirt print.
• Build your own logo into the file after generation, rather than asking the tool to draw it. AI tools often misspell text and bend logos, which can itself trigger a review.
• If you want to copy the structure of a winning competitor ad, copy the idea, not the assets. The page on recreating a winning ad shows how to stay on the right side of that line.

Labels and disclosures inside ads

Briefly, because the full matrix is on another page: Meta says it applies an "AI info" label to ads made or significantly edited with its own tools and to ads where it detects AI through industry signals, and when Meta's own tools add a photorealistic AI human the label sits next to "Sponsored" (page updated 1 June 2026). Political, election and social issue ads carry "Paid for by" disclosures. New York and the EU add laws for synthetic performers and deepfakes. An AI person in your ad is the most visible case on Meta, so plan for the label rather than hoping to avoid it. Read the full picture in do you have to label AI content.

Landing page and product match

The platforms review the ad and the page it leads to. The common AI trap is a "perfected" product. A generated bottle with a cleaner label, a bag with a different stitching, a room that looks larger than the real one. A buyer who receives something different may complain or request a refund, and a mismatch is a classic reason for a review. Etsy says the same in its own rules: images must reflect the item's true colour, condition and scale.

Do this for every product ad:

1. Put the real product photo next to the AI output.
2. Check colour, shape, label text, size and extras that came with the item.
3. Check the landing page shows the same offer, price and look.
4. If the AI added anything the buyer will not get, remove it.

Marketplaces are stricter still. The Amazon and Etsy page covers what passes there, and the TikTok Shop page covers what must be real footage. Misleading-image risks for property and product ads are covered in AI-generated property and product images.

Pre-flight review checklist

Run this on every AI ad before submission. Print it or put it in your approval step.

• Claim source: every claim has a proof file saved.
• Results: no outcome shown that a typical buyer will not get.
• Real product: the item on screen matches the real item.
• People: no invented customer or testimonial, no real face or voice without a signed consent form.
• Logos and IP: no third-party logo, character or celebrity, and your own logo added after generation.
• Text: all on-screen words checked at 100% zoom for errors and garbled letters.
• Personal wording: no copy that implies you know something personal about the viewer.
• Category: if health, finance, housing or cosmetic, the category policy has been read.
• Label: a label or disclosure planned where the platform or a law needs it.
• Landing page: offer, price and look match the ad.
• Files kept: source files, prompt, tool and date saved for any appeal.

If you make many variants at once, build the check into the workflow, not the end. The page on testing 20 ad variations shows how to batch reviews without overloading the team.

What to do after a rejection

1. Read the full reason in the ads account, not just the headline.
2. Fix the cause instead of resubmitting the same ad. Repeated resubmission of the same problem looks bad, and Google's page notes that serious violations can bring suspension without warning.
3. Appeal only with proof. Attach the substantiation file, the real product photo or the signed consent. A short, factual appeal works better than a long argument. No appeal is guaranteed to succeed.
4. Check the account as a whole. One rejection can point to a pattern across your other ads. Review them the same way.
5. Ask the account manager if you have one. Platform contacts can give case-specific answers that no general page can.

When a stricter rule or a lawyer beats this list

Health, finance, housing, cosmetic procedures, alcohol, and anything aimed at children each come with their own rules, and those rules are stricter than anything here. Use the category policy page and, for regulated claims, a lawyer or compliance adviser. This page covers the AI-specific ways ads go wrong. It does not replace the rulebook for your industry. If you work for a client in one of these fields, ask for their legal sign-off before you build the ad, not after.

How I'd approach this

I would treat AI as a way to draw faster, and keep a human accountable for every claim, face and logo. Start every ad from the real product and the real claim, let the tool make the scene around it, and check the result against the checklist before it goes anywhere near an ads account. Keep a folder per ad with the proof, the prompt and the date, so an appeal takes minutes.

If you run this for clients, put the pre-flight list inside your approval step so nothing ships unchecked. The first AI video ad page shows the small-business version, with what to film for real. Blur Studio can build the repeatable workflow, an approval portal and a shared asset library around your tools if you want it automated, and you can read more here. Most teams only need the checklist and the habit.