Google Ads Rules for AI-Generated Images: What Is Allowed in 2026

Google will make your ad images, then police them

Google Ads allows AI-generated images in 2026. Not grudgingly: Google built an image generator directly into the platform, suggests generated images to advertisers on its own, and says it will not limit how you use those images outside its ecosystem. The catch is simpler than most advertisers expect. There is no special AI policy track. Every image, generated or photographed, faces the same review, the same misrepresentation rules, and the same disapprovals.

That cuts both ways. You do not need to disclose AI in an ordinary commercial ad, and no policy punishes you for generating your creative instead of shooting it. But an AI image that fakes an endorsement, doctors a scene, or promises a result your product cannot deliver gets treated exactly like a Photoshopped one. This guide walks through what Google's built-in tools will and will not create, who gets access to them, how the SynthID watermark follows every image, and the three policy lines that actually catch AI creatives at review.

What Google Ads actually allows, and where the tools live

The starting point is a sentence from Google's own help page that settles most of the debate: all advertising content is subject to Ads Policy, regardless of how it is created. Google adds that generative AI content is held to the same set of content standards, reviews, and enforcement for any policy violations, naming inappropriate content and misrepresentation specifically. In other words, the platform does not care whether a pixel came from a camera or a model. It cares what the pixel claims.

The two ways AI images enter your account

There are two distinct paths, and they carry different rules at creation time but identical rules at review time:

• Google's built-in generative tools. You prompt inside Google Ads, or Google suggests generated images automatically. Creation is constrained by Google's own guardrails (more on those below), and every output is watermarked.
• Images you bring from outside. Anything made with a third-party generator uploads like a normal image asset. Google does not block it for being AI-made; it reviews it under the standard policies, like any AI ad creative.

The practical differences between the two paths sit at creation time, not review time:

Dimension	Google's built-in tools	Outside AI images
Creation guardrails	Google blocks restricted content	Your generator's rules
SynthID watermark	Always applied	Depends on the tool
Eligibility gate	Account must qualify	None, just upload
Policy review	Same Ads policies	Same Ads policies
Election disclosure	Required if in scope	Required if in scope

The distinction matters because advertisers keep asking the wrong question. The question is not "does Google accept AI images?" It is "does this image break a policy any image could break?"

Where the built-in generator shows up

Google surfaces its generative image tools in four places:

• When setting up a campaign
• When editing an ad or asset group
• Inside Asset Library
• In certain recommendations, such as an image asset or Ad Strength recommendation

Automatically generated images are based on either the text assets in your campaign or your landing page, and you choose which ones to add. Nothing publishes without your approval. Google's own example prompts read like a stock-photo brief:

• Elegant dogs on a color backdrop
• A wedding ring with holiday themes in the background
• Vegetables spread across a wooden table with a white tablecloth

The tool is built for backdrop and lifestyle imagery, not for finished ad concepts.

One operational detail trips people up: images you create remain in your Asset Library for only 14 days. To keep an image longer, you have to use it in a campaign. If you generate a batch you like, do not let it sit.

The eligibility wall: who can prompt Google's image tools

Manually prompted image generation is not open to everyone. Google restricts which advertisers get it, and the criteria read like a trust score for your account.

Five requirements most guides skip

If the option to generate images from your own prompts is missing, Google's help page says to check that your account:

• Uses the Google Ads interface in English, Dutch, French, German, Italian, Japanese, Portuguese, or Spanish
• Is active and has been running ads for at least 30 days
• Has a history of policy compliance
• Has not completed Election Advertising Verification
• Is not advertising in a sensitive vertical, with sexual and political content as Google's own examples

Prompts themselves are also limited to those same eight languages. Note the fourth item: political advertisers who completed Election Advertising Verification are excluded from prompting entirely. Google is not just filtering what the model draws; it is filtering who gets to ask.

What newer accounts get anyway

An account that fails the 30-day bar is not fully locked out. Google may still show automatically suggested generated images to review when creating certain campaigns, even when you cannot write your own prompts. And the rollout is gradual: Google warns that eligible advertisers may not have every feature yet, so a missing option is not always a policy judgment on your account.

Why the wall exists

The pattern behind the criteria is risk containment. Google states that its AI is designed to not automatically suggest generated images for sensitive verticals, such as political or pharmaceutical ads. Prompting power goes to accounts with a compliance track record, in categories where a bad generation embarrasses no one. It is the same logic that rewards account hygiene elsewhere in the platform: a clean history keeps doors open, the way invalid-click hygiene keeps budgets intact.

What the generator refuses to make

Google's tools are designed to automatically limit certain content at generation time. Knowing the blocked categories saves you from rewording prompts in circles.

The blocked categories

Three groups of content will not come out of the generator:

• Children and prominent likeness. People and face generation is currently limited to adults, and images of children or recognizable real people are off the table.
• Branded items and logos not provided as input. You cannot conjure a competitor's product or logo into your creative. That is the generation-side cousin of a fight many advertisers already know from search, where rivals bid on your brand name.
• Anything that would violate Ads Policy or Google's Generative AI use policy. If a prompt returns nothing, Google's advice is to reword it away from the content the tool is not designed to create.

Product photos and style references, with conditions

Two features push the tools beyond generic stock-alike imagery. "Feature a product" takes one or more images of your tangible product and builds lifestyle scenes around it, including people interacting with or wearing the product. Style reference images let you upload up to 5 examples per generation so outputs match your brand's look. Both carry fine print: the product input is meant only for tangible goods, and style references and outputs should not contain logos, watermarks, or product images.

Generated does not mean approved

This is the clause that surprises advertisers most. Google states plainly that assets created by generative AI are not guaranteed to be approved by Ads Policy. Its own generator can produce an image its own review then rejects, sometimes in combination with other assets, or in line with local regulation. The help page puts the responsibility where it stays: advertisers should review generated or suggested assets for accuracy before publishing. Treat the generator like a fast junior designer, not like a compliance department.

The watermark that follows every image

Every image generated inside Google Ads is marked as AI-made, in two layers, and neither is optional.

SynthID, and what it survives

The first layer is SynthID, which Google describes as an imperceptible, digital watermark that is resistant to manipulations of the image, such as screenshots, filters, and compression. You cannot crop it out or re-export around it, and anyone can check an image for it at labs.google/synthid. The second layer is an open-standard markup that surfaces on tools like Google Image Search, so the provenance is not just detectable but visible in the wild.

You own the use, inside Google or out

The generous half of the policy: subject to its policies and terms, Google will not limit the use of the content generated using Google's AI tools outside of its ecosystem. Generate a lifestyle image for a Performance Max campaign and you can reuse it on your landing page, your email header, or another ad platform. Google also says it employs multiple technical measures so the tools generate novel and unique content, which matters if you worried about your ad image showing up in a competitor's account.

What provenance means for your strategy

Put the two layers together and the strategic read is clear: AI images from Google's tools are permanently identifiable as AI. Platforms are converging on provenance standards, and disclosure rules that are narrow today keep widening. If any part of your plan depends on AI imagery passing as photography forever, that plan has an expiry date. The durable plan is the boring one: make AI creative that is honest about the product, so the label costs you nothing. The cross-platform state of those rules is mapped in our guide to labeling AI-generated ads.

Bringing outside AI images: the three rules that actually bite

Images from third-party generators skip Google's creation guardrails, so the full weight of policy lands at review. Three policy areas do almost all the catching.

Manipulated media and misrepresentation

Google's Misrepresentation policy states that manipulating media to deceive, defraud, or mislead others is not allowed. Mapped to AI failure modes, that covers:

• The synthetic celebrity or expert endorsement
• The fabricated news screenshot or press mention
• The AI-staged "customer photo" of an event that never happened
• The generated image of a public figure using your product

None of these needed AI to be violations; AI just made them cheap to produce, and review systems have adjusted accordingly.

Clickbait imagery and improbable results

Two neighboring clauses in the same policy family do quiet, constant work against AI creative. Ads that use clickbait tactics or sensationalist text or imagery to drive traffic are not allowed, and neither are claims that entice the user with an improbable result as the likely outcome. Generators are dangerously good at both: shocked faces, impossible before-and-after transformations, products rendered with capabilities they do not have. If the image oversells what the landing page delivers, the ad is the problem, not the generator.

The election exception: the one true disclosure rule

For election ads, Google requires advertisers to disclose all election ads that contain synthetic or digitally altered content by selecting the checkbox in the "Altered or synthetic content" section of campaign settings. In scope is content that inauthentically depicts real or realistic-looking people or events. Inconsequential edits are exempt: Google lists editing techniques such as image resizing, cropping, color or brightening corrections, and defect correction as out of scope.

For some formats Google generates the in-ad disclosure from the checkbox itself; for the rest, the advertiser provides a prominent disclosure. Google's examples of acceptable language include:

• "Altered or synthetic content."
• "This image does not depict real events."
• "This video content was synthetically generated."

The generated-images help page adds that Google Ads Policy requires disclosure of manipulated media uploaded from any third-party tools for certain political ads, closing the "but I made it elsewhere" loophole. If you run commercial ads, this rule does not touch you; if you touch political content, it defines your workflow.

A pre-flight checklist for AI images in Google Ads

Run every AI image through this list before it enters a campaign, whether it came from Google's generator or an outside tool:

• Does the image depict the actual product you sell, accurately? No invented features, sizes, or results.
• Does it show any real person, or someone recognizable as one? If yes, stop unless you have rights and the depiction is truthful.
• Does it include a brand, logo, or trade dress that is not yours? Remove it.
• Would a reasonable viewer feel deceived comparing the image to your landing page? Fix the gap on whichever side is lying.
• Is the imagery sensationalist, or does it promise an improbable outcome as the likely one? Tone it down.
• Is this a political or election ad with synthetic or altered content? Check the disclosure box and plan the in-ad disclosure.
• Is the image still in Asset Library limbo? Remember the 14-day retention window for unused generated images.
• Are you reviewing generated assets before publishing, as Google explicitly asks? Approval is your job, not the model's.

Ten minutes with this list costs less than one disapproval, and far less than the account-history damage that repeated violations cause. If you are still setting up your first campaigns, start with our walkthrough on how to make Google Ads before you optimize creative at this level. And if you run video, the image rules here sit alongside Google's broader creative standards for video ads, which changed in their own ways this year.

How Novoads fits: ad creative built to pass review

Google's generator solves one problem: producing generic lifestyle imagery inside your account. It does not solve the creative that actually wins auctions for e-commerce brands, which is ad-shaped content: product shots staged like ads and UGC-style AI video that reads like a person talking, not a promo. That is the slice Novoads covers. You upload a real product photo, and Product-to-Ad turns it into staged ad images at 0.3 credits per image; AI actors then deliver your script on camera in your buyer's language, with a real regional accent. Because the input is your genuine product and the output is a standard image or video file, everything in this guide applies cleanly: honest depiction in, policy-compliant creative out. You can start for $1 and test it on your own catalog. $1 for 3 days of access. Cancel anytime.

The line was never human versus machine

Read every Google page on this topic back to back and one design choice stands out: Google wrote almost no AI-specific advertising policy. It gated who can generate, watermarked what gets generated, and then pointed the entire existing rulebook at the output. The line Google draws is not human versus machine. It is honest versus deceptive, and it sits in exactly the same place it did before generative models existed. Advertisers who internalize that stop asking whether AI images are allowed and start asking the only question review ever asks: does this image tell the truth about what you sell? Make creative that answers yes, and 2026's rules are not a wall. They are a moat around everyone who cannot.