Google's New AI Image Rules Will Penalize Brands in Shopping Results — Here's What Marketers Need to Know Before August

Google's February 2024 clarification on AI images in Merchant Center feeds isn't about whether you use generative AI. It's about whether your files carry the right machine-readable metadata, and a missing tag can trigger product disapproval or ranking demotion.

The required standard is the IPTC DigitalSourceType metadata tag. Google recognizes three values: TrainedAlgorithmicMedia (a generative model trained on sampled content), CompositeSynthetic (real images with synthetic elements blended in), and AlgorithmicMedia (purely algorithmic, no training data involved). This applies across image_link, additional_image_link, and lifestyle_image_link in both Merchant Center Classic and Merchant Center Next.

Here's where most teams trip up: lossy image compression strips IPTC metadata silently. A properly tagged image exits your design tool compliant and arrives at Google's systems naked of any disclosure tag. Google reads C2PA, IPTC, and EXIF data on every upload, so the stripping happens invisibly, and the resulting disapproval can arrive with no obvious cause.

What actually gets penalized — and what doesn't

Google doesn't penalize AI-generated images for being AI-generated. The three failure modes the rules actually target are fully synthetic main images with no photography involved, images that misrepresent the physical product a customer receives, and undisclosed AI styling applied to real product shots, things like generated shadows, reflections, or staging elements added without metadata disclosure.

That third category is the hardest to catch, and it tends to go unnoticed until a disapproval surfaces. AI imagery stays fully permitted across Display Network, YouTube, and Discovery creative. The restrictions are specific to Merchant Center feeds and Shopping placements.

How Google's enforcement actually works — three signals brands need to understand

Google combines IPTC/C2PA/EXIF metadata checks, algorithmic detection of visual characteristics common to AI-generated images, and user-reported misleading listings. Each path leads independently to the same consequence ladder: product disapproval first, ranking demotion second, and account-level review for repeat violations beyond that. A brand can be metadata-compliant and still trigger review through user reporting if the image misrepresents what shoppers actually receive.

The promotional overlay grey zone Google created by shipping Asset Studio

Merchant Center's image_link spec has long prohibited promotional overlays: text, price badges, discount banners. Then in 2025, Google launched Asset Studio at Google Marketing Live, a first-party AI creative tool that generates product images including ones with promotional overlays. PPC practitioners have documented "20% OFF" banners and "NEW PRODUCT" badges staying live without disapprovals. Brand logos carry the highest disapproval risk; promotional text overlays are widely tolerated right now but carry no official sanction. Move promotional messaging into feed attributes like sale_price and promotion_id rather than burning it into the image.

The EU AI Act deadline arriving August 2, 2026 — a separate layer on top of Google's rules

Article 50 of the EU AI Act creates a buyer-facing labeling obligation under law, separate from Google's feed metadata requirement. Any brand serving EU customers is in scope, including UK businesses with EU users. Penalty exposure reaches up to €15 million or a percentage of worldwide turnover. Satisfying Google's IPTC metadata rule does not make you Article 50 compliant; the EU requirement adds a consumer-visible disclosure on top of the machine-readable tag.

A pre-August compliance audit: what to check, in what order

Work through this in sequence:

• Inventory every AI-touched image in your feed, separating fully synthetic from AI-assisted real-product images
• Verify the DigitalSourceType tag is present on each file, then confirm your export and compression workflow preserves it
• Review images against the misrepresentation standard: does what's shown match what the customer receives?
• Pull Merchant Center diagnostics for existing disapprovals and monitor after re-uploading corrected files

What brands should prioritize if they can't fix everything before the deadline

Start with fully synthetic main images missing IPTC metadata; they face the clearest, most direct enforcement path. AI-generated images misrepresenting the actual product come second, because Misrepresentation policy violations escalate faster than disclosure violations. Lossy compression workflows silently stripping correctly tagged images are third, since the damage stays invisible until a disapproval wave hits mid-catalog. A disapproval event during peak trading is significantly harder to recover from than a pre-deadline audit done right, which is why content platforms like Letterstory, built around strategy-first production workflows, increasingly treat compliance checkpoints as part of the publish cycle rather than an afterthought.