Advertising Evidence Brief (Crabstone)

Google's AI Label Is Free. Skipping It Risks €15 Million.

On 9 July Google began telling shoppers when an ad was made with AI across Search, YouTube and Discover, yet it auto-labels only its own tools and takes third-party advertisers at their word. With the EU AI Act's Article 50 landing on 2 August, the disclosure becomes a compliance cost that falls first on the fast-fashion marketplaces that industrialised the synthetic product image.

Sir John Crabstone

On 9 July, Google added a panel to its ad menu called “How this ad was made”. With it came a question every retail marketer must answer: was the picture real? The feature reaches Search, YouTube and Discover, and it turns the synthetic product shot from a shortcut into a disclosure. The prompt that once saved a photoshoot is now a line in a compliance file.

The mechanism is more telling than the announcement. Ads built with Google’s own generative tools carry the label automatically; ads made anywhere else depend on the advertiser ticking a box that Google has said it will not check. Google discloses its own work and takes everyone else at their word. It is an honour system written by the one party with no reason to police it.

Since 2023, Google has taught retailers to prefer the render. Its Performance Max tools build generated lifestyle assets from a product feed, so the model and the backdrop now arrive without a photographer. The platform that made the synthetic image cheap now owns the disclosure that makes it dear.

The burden does not fall evenly. Temu and Shein, the ultra-fast-fashion marketplaces, were called by Business of Fashion the “Jaws” of digital advertising. Their catalogues turn over fast enough that the product photograph could never survive as a craft; it had to become a template. The advertiser most reliant on the manufactured picture can least afford to admit it.

The scale is not abstract. When tariffs bit this spring, Temu cut its daily ad spend by 31% across Facebook, Instagram, TikTok, Snap, X and YouTube in the two weeks after 31 March. A cut that broad, across a spend base that large, does not happen to a marginal advertiser. These accounts will carry the label first, and carry more of it than anyone.

In America the label is a courtesy. Europe made it law. From 2 August, the EU AI Act’s Article 50 requires synthetic content to be marked as such.

The penalty for skipping the mark runs to €15 million or 3% of worldwide turnover. For a marketplace with billions in sales, 3% is the larger figure, and the one that applies.

The label also does something the compliance memo will miss. In fashion the image is the merchandise; the shopper buys the drape before the dress. Google’s rule reaches the everyday edits retailers lean on, and a synthetic background or an AI render now sits inside the disclosure. Persuasion by picture only worked while the picture was trusted.

The trade press read the rollout as transparency with a caveat: Google labels the AI, if the advertiser admits it. That reading is too kind. Retailers are already lobbying Brussels to exempt AI-made ads from the rule, and on YouTube Google auto-labels AI video whether the creator discloses or not, a stricter test than the one it applies to its own ads. Google keeps the revenue from the render and leaves the advertiser to carry the liability.

Google will not check the box; Europe will fine the empty one.

So the retailer who took the advice inherits the exposure. The render was free; the label is not, and the penalty, when it lands, falls on the buyer who generated the image and never on the platform that offered to. The biggest spenders are counting their pictures now.

Related Coverage