Do You Need to Disclose AI-Generated Images in Ads? A 2026 US Advertiser’s Guide

Do you need to disclose AI-generated images in ads? Not under a single rule. As of mid-2025, no comprehensive federal AI disclosure law had been enacted. The real test is whether the image could mislead a buyer about a material fact, such as how the product looks, how it performs, or who is shown using it. A label on the wrong image can still leave a misleading ad in market, while an unlabeled stylized illustration may pose no real risk at all. This guide gives US advertisers a practical way to decide. It explains where the Federal Trade Commission’s deception standard applies, how major ad platforms treat synthetic media, and a two-axis method for deciding when a disclosure is required, recommended, or unnecessary.

Key Takeaways

• No single rule says every AI image needs a label. The question is whether the image could mislead a reasonable buyer about a material fact.
• Score each image on two axes. Materiality asks whether a buyer would rely on what the image shows. Realism asks whether a typical viewer would take it for a photograph of reality. High on both calls for a clear disclosure placed close to the image.
• Stylized, obviously illustrated images with low materiality usually need no label. A cartoon mascot or a clearly painted scene rarely creates a false impression about a purchase.
• A disclosure cannot fix a misleading product image. If an AI render shows features the real product lacks, a small “AI-generated” caption does not make the ad truthful. The product claim itself is still deceptive under the FTC’s standard.
• Platform labels and captions do different jobs. A platform’s AI-info label is not automatically your legal disclosure, so check what each one covers and add a note in the creative where the risk sits.
• Record your reasoning for every high-risk asset. A short note showing how you scored materiality and realism shows good-faith judgment if a regulator, platform reviewer or customer challenges the ad.

Table of Contents

• What Counts as an “AI-Generated” Image in an Ad?
• Does US Law Require You to Disclose AI-Generated Images?
• When a Disclosure Is Required Versus Recommended
• Platform Rules: Meta, Google, TikTok, and Amazon
• Conclusion and Summary
• Frequently Asked Questions

What Counts as an “AI-Generated” Image in an Ad?

The label “AI-generated” covers a wide range of production methods, and the disclosure question depends on which one you used. Before you decide whether an image needs a label, identify where it sits on that spectrum.

Fully generated images versus AI-edited photos

A fully generated image is one where a model produced the entire frame from a text prompt or a reference, with no camera capture of the scene. A product shot created this way, or a lifestyle scene with invented people, belongs here.

An AI-edited photo starts with a real photograph and changes part of it. The product was photographed, but the background was replaced, the lighting adjusted, a scratch removed or a feature added. The result is partly real and partly synthetic.

This distinction matters because the two carry different risks. A fully generated image has no physical counterpart, so every visible detail is a claim the advertiser made. An AI-edited photo inherits the real product’s appearance, so the risk concentrates on the parts the model changed. A hyper-real render of a sofa and a photograph of the same sofa with an AI-invented fabric pattern can both mislead, but in different ways.

Where generative fill, upscaling and background removal fit

Many everyday tools now use AI without the advertiser thinking of the output as “AI-generated.” Background removal, automatic color correction and upscaling usually preserve the real subject and alter only its surroundings or resolution. These rarely change a fact a buyer would rely on, so they sit at the low end of concern.

Generative fill and generative expansion are different. These tools invent pixels that were never captured, such as extending a table, adding a second product, or filling in a missing part of a garment. Once a tool invents content that a viewer would read as part of the product or scene, disclosure should be considered.

A practical test: ask whether the edit adds, removes or changes something a buyer would notice and rely on. If yes, treat it as a generated element. If the edit only tidies a photo the buyer would see the same way with a studio lamp and a clean backdrop, it probably does not change the impression.

AI-generated people, products and locations

Each category carries different stakes.

• People: A synthetic person presented as a real customer, user or expert can mislead about who is endorsing the product. This is the highest-stakes category, because viewers often assume a face represents a real experience.
• Products: A generated product image can overstate features, color accuracy, size or finish. Buyers make purchase decisions on exactly those attributes.
• Locations: A generated beach, store or hotel room can imply a place the business does not operate in, or a service environment that does not exist.

The more a viewer would use the image to judge something real about the offer, the more the image is a material fact. A stylized illustration of a generic person in a cartoon setting makes no factual claim. A photorealistic person giving a testimonial makes a very specific one.

Our position: Do not sort images by the tool used. Sort them by the claim the image makes to the viewer. A tool-based rule (“we only label fully generated images”) misses edited photos that mislead. A blanket rule (“we label everything touched by AI”) produces meaningless labels on harmless edits and dilutes the warnings that matter. Our approach uses two axes, materiality and realism, which is why it works where the simpler approaches do not: it asks what the image does to a buyer’s decision rather than how it was made.

Does US Law Require You to Disclose AI-Generated Images?

As of mid-2025, no comprehensive federal AI disclosure law had been enacted, so no US rule says to label an AI image in an ad as such. Advertisers still need to disclose AI-generated images in ads when the image could mislead a buyer about a material fact. The answer depends on the existing rules against deception, which already cover visual claims whatever tool produced them. The practical test is whether the picture misleads, not whether a model made it.

The FTC’s truth-in-advertising standard and why it applies

The FTC enforces Section 5 of the FTC Act, which prohibits unfair or deceptive acts or practices in commerce. The agency’s Advertising FAQs guidance says an ad is deceptive if it contains a representation or omission likely to mislead consumers acting reasonably, and that the representation is material. The medium does not change that test. A photograph, a painting, a video render and a generated image are all representations to the Commission.

The FTC has already addressed visual misrepresentation directly. Its Endorsement Guides and its guidance on product demonstrations both make clear that an advertiser cannot show a result the product does not reliably produce. A visual shown to sell a product must reflect how the product actually performs or looks. An AI image that shows a feature the real item lacks falls under that principle, whatever tool created it.

The FTC does not need a new rule to reach AI imagery. It applies the existing one.

What “deceptive” means when a picture is involved

Three elements determine whether an image is deceptive, and each can be assessed without knowing how the image was made.

• Net impression. The FTC looks at the overall impression an ad leaves, not just whether each element is literally true. A picture of a phone with a screen that looks sharper than the real one creates an impression about performance, even if the caption is accurate.
• Materiality. A misrepresentation is material when it would affect a buyer’s decision. Claims about performance, features, safety and efficacy are the kind of claims the FTC generally treats as material. An AI image of a product’s appearance is material when a buyer could reasonably rely on it to judge what they are getting.
• Reasonable consumer. The standard is how a typical viewer would interpret the ad, not how an expert or the advertiser would. If most viewers would read a picture as a photograph of the real product, the image carries factual weight.

Realism is where this becomes useful in practice. A stylized, clearly artistic graphic rarely creates a net impression about a specific, real-world fact. A near-photographic render of a product, or a realistic person presented as a real customer, does. Realism is the main variable, and it links directly to the Deception Risk Matrix covered later in this guide.

Consider a worked case. A skincare brand runs a paid social ad with a generated image of a woman whose skin looks flawless after one application, while the real product produces a modest change. Materiality is high because the result claim concerns efficacy. Realism is high because the face looks like a real photograph. The image misleads on a material fact. A caption saying “results may vary” does not cure a picture that overstates what the product does.

State laws and how they interact with federal rules

State attorneys general and state consumer protection statutes add a second layer. Most states have their own unfair and deceptive trade practices acts, often called UDAP statutes, and many follow the FTC’s interpretation of deception. Because these laws are broad, a state enforcer can pursue an ad the FTC declines to challenge, and private plaintiffs can sue under some of them.

Federal and state rules interact in three ways:

1. Harmonized standards. Most state UDAP laws track the FTC’s reasonable-consumer test, so an image that fails the federal standard usually fails the state one too.
2. Independent enforcement. A state can act on an ad even when no federal action follows. Federal silence does not mean an ad is safe.
3. Emerging state AI measures. Several states have passed or proposed rules on AI-generated content, particularly around political advertising and deepfakes of real people. These provisions vary in scope and effective date. Advertisers should check the law in each state where an ad runs, not rely on a single national summary.

State rules change often, so the durable advice is to meet the federal deception standard and keep records showing you did. An advertiser who can document a good-faith assessment of each image is better placed under any state regime than one who assumed the rules did not reach generated content.

When a Disclosure Is Required Versus Recommended

Many advertisers ask whether a rule says “label this image.” When the answer is “not always,” they stop there. A better question is whether the image could make a reasonable buyer believe something false about the product, the results, or the people shown. That question has a workable answer, and it turns a vague legal duty into a decision you can make in minutes with a creative brief open.

Decision framework: materiality, realism and reliance

The Deception Risk Matrix uses a two-axis test. Score each AI-generated image on two questions.

Materiality: Does the image show a fact a buyer would rely on when deciding to purchase? Product appearance, measured results, the identity or experience of a person, and the location of a service all count. A background of a generic beach does not.

Realism: Would a typical viewer believe this is a photograph of reality? A glossy, photographic render of a sofa in a sunlit room scores high. A flat, clearly illustrated character scores low.

The combinations map to actions:

Materiality	Realism	Risk level	Recommended action
High	High	High	Clear disclosure placed next to the image
High	Low	Medium	Short caption or alt-text note
Low	High	Medium	Short caption or alt-text note
Low	Low	Low	No label needed

Two cases sit in the middle and need judgment. A high-materiality image that looks stylized still needs a note, because a stylized picture of a dishwasher’s cleaning results can still mislead about performance. A photographic-looking image of a generic office with no product claim is medium at most, since realism alone does not make an ad deceptive.

The matrix keeps the reliance test at the center. The FTC’s general standard asks whether an ad is likely to mislead consumers acting reasonably, and whether that misleading element is material to the purchase decision. Materiality and realism are how you apply that standard to a picture. The baseline appears in the FTC’s advertising FAQs and its Policy Statement on Deception.

Product demonstrations and results imagery

Product demonstrations carry the highest risk because buyers treat them as evidence. If an image shows a blender crushing ice in seconds, the viewer assumes that performance is real. An AI render that makes the product look more capable than the physical item fails on materiality, because the performance claim is the reason to buy.

Results imagery needs the same discipline. An AI “before and after” skin result, a cleaned carpet, or a larger engine output is a claim about what the product does. If the real product cannot produce that result, a label does not rescue the ad. The image is the claim, and the claim is false.

Use this check before launch:

1. Does the image show the product doing something? If yes, treat it as high materiality.
2. Could a buyer compare it against the real product and notice a difference? If yes, treat it as high realism.
3. Does the substantiation file hold up if the image were the claim? If not, remove the image or replace it with real footage.

Where the AI image only improves presentation, such as a tidier background behind a real product photo, materiality is lower. The product is still real, so the buyer’s reliance is intact.

Realism and consumer trust

Consumers respond differently to synthetic imagery depending on how it is presented. Realism raises the stakes because a viewer who cannot tell a render from a photograph is more likely to rely on it. . Consumer trust figures for AI-generated ad imagery should be checked against current published research before use, since surveys in this area change quickly.

Testimonials, influencers and synthetic spokespeople

Testimonials are where the matrix meets an additional rule. The FTC’s Endorsement Guides require that endorsements reflect the honest opinion of a real endorser who used the product. A synthetic person who “loved the product” is not a real endorser, so the endorsement is fabricated, and a label does not make a fabricated testimonial acceptable.

Influencer content adds a second layer. If a real creator posts an AI-generated image of themselves using the product, the material question is whether they used it. A disclosure of the AI edit does not fix a testimonial the creator never gave.

Synthetic spokespeople clearly presented as characters, such as a cartoon mascot or an obviously fictional brand persona, are low in both materiality and realism for most categories. A photographic-looking AI model presented as a customer is high on both, and the label should sit close to the figure, not in the footer.

Ready-to-use disclosure phrasings for high-risk cases:

• Synthetic model: “Image shows a computer-generated model. Not a real customer.”
• AI product render: “Product image is digitally rendered and may differ from the physical item.”
• AI results imagery: “Result shown is simulated. Individual results will vary.”

The rule of thumb: if the figure is implied to be a real person with real experience, the image is high materiality. Disclose, and check whether the endorsement itself is honest before you publish.

Platform Rules: Meta, Google, TikTok, and Amazon

Each major platform sets its own policy on AI-generated and manipulated media, and the rules differ. A platform label is not a legal disclosure. Passing ad review does not show that an image is accurate for the buyer. Check current terms before you publish, because these policies change often.

Meta’s AI-info labels and ad review requirements

Meta applies an “AI info” label to content it detects or is told was made or edited with AI. For ads, two questions matter. Does the label appear on the creative? Will ad review reject an ad that shows a realistic person or event that did not happen? Meta’s Advertising Standards and its Community Standards on manipulated media set out what it treats as misleading. Review the current versions before launch, since these policies are updated periodically and the latest wording is the only one that governs your submission.

The gap is this. A platform label says the content used AI. It does not say which part is synthetic or whether the product looks the way it is sold. An unlabelled ad can pass review and still mislead. A labelled ad can still mislead if the depicted product differs from the real one.

Google Ads policy on synthetic and manipulated media

Google Ads updated the Unacceptable business practices portion of its Misrepresentation policy in March 2024 to cover impersonation. Its Misrepresentation policy prohibits ads that misrepresent a product, service or the advertiser. Google can require a disclosure for some synthetic content and can disapprove creatives that show results or features that are not accurate. Read the current Google Ads policy center before you submit, because enforcement wording shifts over time.

On search and display, the image is often small and seen beside text, so a label is easy to miss. Treat the image as the claim, and make sure the claim is accurate.

TikTok and Amazon

TikTok’s advertising policies restrict misleading and manipulated content, and the platform has labelling tools for AI-generated material. TikTok ads are short, vertical and fast. A disclosure must be legible on a phone within a second or two. An overlay or caption usually works better than small print.

Amazon’s advertising policies and its product detail page rules govern what a listing and its sponsored placements can show. Amazon is strict about images that do not represent the actual product, because shoppers compare the listing photo with the item they receive. A generated lifestyle image is usually acceptable as context. A generated product image that adds features, changes color or enlarges the item invites rejection and returns.

Disclosure settings and what they actually do

Most platforms let advertisers toggle an AI disclosure or synthetic-media setting when creating an ad. Ticking the box typically adds a label and feeds the platform’s own records. It does not change what the image says, and it does not make a misleading image accurate.

Mechanism	What it does	What it does not do
Platform AI label (automatic or toggled)	Tells viewers the content used AI	Explain which part is synthetic or whether the product matches
Ad review approval	Confirms the creative meets platform policy at review	Certify that the image is not misleading to a buyer
Caption or on-image note	Gives specific, readable context at the point of decision	Replace an accurate depiction of the product
Metadata or file provenance	Records how the file was made for internal and audit purposes	Reach the consumer viewing the ad

The working rule: use the platform setting as a minimum. Where the Deception Risk Matrix places an image in the high-materiality, high-realism quadrant, add a specific on-image or caption disclosure. Never rely on the setting to justify an inaccurate product image. Record the reasoning behind each choice, so you can show good faith if a platform or regulator asks later.

Platform rejection case: a disclosed AI image that was still refused

Each ad review rejection should be documented with the platform, the review ticket reference and the date it occurred. In this kind of review, a disclosed synthetic image is typically judged on whether the depicted product or result matches what is sold, not on the label alone. Record the reviewer’s stated reason and the change that resolved it, if any, so the next creative avoids the same issue.

Conclusion and Summary

Our position: disclose whenever the image could change a purchase decision

Most advertisers ask the wrong question. They ask, “Does a rule require me to label this image?” The better question is, “Could this image lead a reasonable buyer to believe something false about the product, the results, or the people shown?” If the answer is yes, disclose. If not, a label adds little and may only add clutter.

The Federal Trade Commission judges deception under Section 5 of the FTC Act by the net impression a reasonable consumer takes from an ad. The test does not ask whether the picture is technically real. A label does not cure a misleading image. A missing label does not automatically create liability for a harmless stylized illustration. The disclosure should match the risk.

Use the Deception Risk Matrix from earlier in this guide as the working tool:

Quadrant	Materiality	Realism	Recommended action
A	High	High	Clear disclosure, placed next to the image and readable on mobile
B	High	Low	Short caption or alt-text note, especially if the image shows a result or a person
C	Low	High	Review the image for hidden material claims, then decide whether a note is needed
D	Low	Low	No label required, provided the image is obviously stylized

Three rules follow:

1. Treat product and results imagery as material by default. A buyer relies on what a sofa looks like, how a supplement affects skin, or how a cleaner removes a stain.
2. Do not treat a label as a substitute for accuracy. If an AI image shows a feature the real product lacks, the ad misleads even with a label. Fix the image.
3. Document the decision. Record the quadrant, the reason, and who approved the call.

Advertisers who follow this approach avoid two errors. Over-labelling everything weakens the disclosures that matter. Under-labelling images that shape purchases, then relying on a platform setting or a tiny caption as a defence, creates the real risk.

Frequently Asked Questions

Do I have to label every AI-edited photo in an ad?

No. Minor edits such as color correction or removing a stray object rarely change what a buyer relies on. Disclose when an edit alters a material fact, such as the product’s shape, features, size, results, or the people depicted. Use the matrix to decide case by case.

Does a small AI background change need a disclosure?

Usually not, if the product and its context are unchanged. If the new background makes the product look used somewhere it cannot be used, such as a beach setting for a product that is unsafe in water, the change may be material and should be assessed.

Are AI-generated models in ads treated differently from real ones?

Yes, when the model is presented as a real customer, user, or expert. A synthetic person giving a testimonial or claiming personal results can mislead buyers about endorsement, so treat that case as high materiality. A clearly stylized brand mascot is different.

Do platform labels satisfy the FTC’s requirements?

Not necessarily. A platform label tells viewers the content may be AI-generated. It does not confirm the ad is accurate or that the advertiser has assessed deception risk. Use platform labels alongside your own disclosure, and check each platform’s current policy before launch.

What happens if I forget to disclose an AI-generated image?

The outcome depends on whether the image misled anyone. An undisclosed image that did not affect a purchase is a low-risk oversight, though it may still breach platform policy and lead to rejection. An undisclosed image that materially misrepresented a product can draw an FTC inquiry, consumer complaints, or a platform suspension, and the advertiser bears responsibility.

Who is responsible when an agency produces the ad?

The advertiser is generally responsible for the claims its ads make, even when an agency or vendor creates the creative. Written briefs, provenance records, and an approval step with a named owner protect both parties and show who made each decision.