Your agency runs paid campaigns across Meta and Google for a dozen-plus clients. Some of that creative was touched by AI somewhere in production — a background generated in Meta Ads Manager, a product shot cleaned up in Photoshop's generative fill, a voiceover drafted with an AI tool before a human recorded the final read. Right now, there's a good chance nobody on your team could tell you exactly which assets those were, or whether any of them were supposed to carry a disclosure label.

That gap is about to matter a lot more. Two platform policies are already live, and two new laws land on the same date: August 2, 2026. This isn't a "watch this space" trend piece — it's a dated compliance problem with a client-trust problem sitting right behind it.

What Actually Changed: Meta, Google, and Two Laws Converging on August 2

Four separate things are happening at once, and it's worth being precise about which are platform rules and which are actual law.

Meta already labels ads automatically when a creative team uses Background Generation, Image Generation, or Add Animation inside Ads Manager, and it detects third-party AI tools — Photoshop's generative fill, DALL-E, Canva AI — through C2PA metadata embedded in the file itself. Undisclosed AI content is now a live reason for ad rejection, not a future one, and it's become one of the more common rejection triggers agencies are running into this year, according to AuditSocials' 2026 Meta ad-policy tracking.

Google moved second. On July 9, 2026, Google added a "How this ad was made" panel inside My Ad Center, visible globally across Search, YouTube, and Discover. Ads built with Google's own generative AI tools get the disclosure automatically. Ads built anywhere else depend on the advertiser checking a box — and Google has said outright it won't verify whether that box was checked truthfully, which shifts the actual burden of getting this right onto agencies and brands, not the platform.

Then there's the part that isn't optional. California's AI Transparency Act (SB 942, as amended by AB 853) becomes operative on August 2, 2026, requiring embedded, machine-readable "latent disclosure" metadata in AI-generated image, video, and audio content, with penalties running $5,000 per violation, per day. On the same date, Article 50 of the EU AI Act becomes applicable, requiring that AI-generated or AI-manipulated audio, video, or images be made clear to users, with a stricter standard for anything resembling a deepfake. The EU is still finalizing exactly what the consumer-facing label should look like — a proposed interim "AI" icon is circulating — so treat that specific visual as unsettled for now rather than final.

Two things worth being clear-eyed about here. First, platform labels and legal requirements are different animals: Meta's or Google's label doesn't automatically satisfy California or EU law, and passing the legal bar doesn't exempt you from a platform's own rules. Second, whether Meta and Google end up aligning their label formats with the EU's eventual standard icon is still an open question — worth checking again closer to August, not something to treat as settled today.

Why This Is a Client-Trust Problem Before It's a Compliance Problem

The fines are real, but for most agencies, the bigger risk lands somewhere else first: a client finding out AI touched their creative from a platform label, a rejection notice, or a rumor — instead of from you.

That's already happened in less serious versions. When Meta's AI tools altered ad images without approval, brands including REI and a small business called Quite Literally Books ended up with product photos and campaign creative changed in ways they hadn't signed off on, and Meta's response was essentially that reviewing AI output is the advertiser's responsibility. If your agency is the one running that account, "Meta did it" is not a sentence a client wants to hear after the fact.

The opposite failure mode is just as costly. One Shopify brand paused an entire AI video campaign for two weeks in early 2026 after a Reddit thread wrongly claimed Meta was banning AI-generated ads — only to find out later that their ads had been fully compliant the whole time. That's two weeks of paused spend and a client conversation nobody needed to have, caused entirely by not having accurate information to act on.

Both scenarios point to the same fix: agencies need to own this conversation proactively, on their own timeline, rather than reacting to whatever a platform or a rumor surfaces first. Legal commentary on the broader disclosure landscape puts it plainly — agencies should be the ones explaining this shift to clients and aligning on who owns each compliance decision, not waiting for the client to ask.

How to Audit What You're Already Running

You don't need forensic tooling to start this, and you don't need to review every asset with equal urgency. The more useful first move is auditing what your own team actually used to make things, not trying to reverse-engineer AI usage from the outside.

Start with your team's tool list, not your ad library. Ask everyone who touches client creative — designers, editors, freelancers, contractors — which AI tools they've used on live campaigns in the past few months. This is faster and more reliable than trying to visually inspect every asset for AI fingerprints.

Prioritize by risk, not by client size. Photorealistic AI-generated people, synthetic voiceovers, and any creative depicting something that didn't actually happen carry the highest disclosure risk under the IAB's materiality standard, described below. Review those first. Minor edits — cropping, color correction, background cleanup — generally don't trigger a label at all, so don't burn review time there.

Check what's already live. For Meta accounts, verify labels directly in Meta's Ad Library against what you know was AI-touched. For Google accounts, spot-check the My Ad Center panel on a sample of live ads, and confirm whether the third-party-AI checkbox was set correctly for anything not built with Google's own tools.

Update your creator briefs now. Add a required field asking any contractor or creator to flag AI tool use before delivery. This is the cheapest fix on this list and it prevents the next three months of assets from creating the same backlog you're auditing today.

If your account list is too long to review individually before August 2, that's a real constraint — the risk-based prioritization above exists specifically so you're not trying to do all of it at once.

How to Bring It Up With Clients Before They Find Out

There isn't yet a well-established script specific to agencies disclosing AI use to advertising clients — this is new enough that one hasn't settled. What does transfer is a pattern from adjacent professional-services fields, like legal, that have been navigating client-facing AI disclosure for longer: name the tool or process used, name the human review step that happened after it, and confirm who owns responsibility for the final asset.

Applied to an ad account, that's a short, factual note — not a big meeting for every client. Something like: which AI tools touched this campaign's creative, what your team reviewed before it shipped, and that your agency is tracking the August 2 requirements on their behalf. Putting this in writing, as part of a regular campaign update or a one-time account note, scales far better than scheduling individual conversations across every client relationship — which is the real barrier for agencies managing a large account list under a hard deadline.

The clients who find this out from a rejection notice or a platform label instead of from you are the ones most likely to ask what else they weren't told.

Building Disclosure Into Your Creative Workflow Going Forward

August 2 is a deadline, not a finish line. AI tool use in ad creative isn't going to slow down, so the fix that holds is a workflow change, not a one-time cleanup.

Three additions make this durable: an internal, written definition of what counts as "AI-generated" versus "AI-edited" for your team specifically, so people aren't guessing case by case; a short decision log per asset noting what AI tool was used and what a human reviewed before launch; and a review checkpoint before any client creative goes live, the same way a legal or brand-safety check would already sit in your process.

For deciding what actually needs a disclosure label in the first place, the IAB's AI Transparency and Disclosure Framework offers a useful filter: disclosure matters when AI materially affects authenticity, identity, or representation in a way that could mislead someone — not for routine production tasks or background tooling. That's a reasonable internal standard to adopt even before it's formally required everywhere your clients run ads.

Frequently Asked Questions

What's the difference between "AI-generated" and "AI-assisted" for disclosure purposes?

Under the IAB's materiality standard, disclosure is generally required when AI meaningfully affects authenticity, identity, or representation — a synthetic voice, a photorealistic person who doesn't exist, a scene depicting something that never happened. Minor AI-assisted edits like resizing, color correction, or background cleanup generally don't require a label. This isn't just an industry norm: Article 50 of the EU AI Act itself carves out the same distinction, exempting AI that "performs an assistive function for standard editing" or doesn't "substantially alter" the underlying content from its marking requirement.

Does this apply to campaigns outside California and the EU?

Platform labels from Meta and Google apply globally, regardless of where the campaign runs. Legal exposure is jurisdiction-specific: California's penalties apply to covered providers with a large enough California user base, and EU AI Act obligations apply to campaigns reaching EU users. A US-only agency running US-only campaigns still has to meet the platform-level requirements even without direct EU or California legal exposure.

What actually happens if we don't disclose?

On Meta, undisclosed AI content is grounds for ad rejection, and campaigns can be retroactively flagged after launch. California's penalty structure runs $5,000 per violation, per day, once the law is operative. Under the EU AI Act, Article 99 sets the penalty for Article 50 transparency violations at up to €15 million or 3% of a company's total worldwide annual turnover, whichever is higher — the same tier as failing to meet obligations as a provider, importer, or deployer elsewhere in the regulation.