The question is not whether AI touched the file. The question is whether a reasonable person could treat the result as real.
The AI label is not the policy.
On August 18, the Interactive Advertising Bureau released Version 2 of its AI Transparency and Disclosure Framework. MarTech reported that the document tells advertisers, agencies, publishers, platforms, and technology companies when to tell consumers that AI was used in advertising and marketing content (MarTech).
The line IAB draws is materiality.
Disclose when AI changes authenticity, identity, or representation. Do not put a label on every color correction, crop, or background track.
That is a useful distinction.
It is not a finished operating rule.
The Law Already Moved
IAB first published this framework in January. Version 2 exists because the legal map filled in around it.
Marketing Dive reported that New York's synthetic-performer law, California's SB 942, South Korea's labeling rules, and Article 50 of the EU AI Act have all taken effect since that first version (Marketing Dive). IAB's own page dates the update August 18 and presents it as a practical protocol for consumer-facing advertising and marketing content (IAB).
Those regimes do not share one icon, one threshold, or one enforcement date.
MarTech noted that the IAB framework offers U.S. advertisers a standardized sparkle icon or clear text, and that those options are recommendations, not a substitute for the law. The EU requires disclosure for covered AI-generated content and deepfakes under Article 50. It does not mandate a specific icon (MarTech).
A company that treats "we added the sparkle" as done has confused a mark with a decision.
Inventory The Change, Not The Tool
Most teams still sort creative by software.
This file came from an image model. That file went through an edit suite. This voiceover used a clone. That landing-page draft started in a chat window.
IAB's test is different. Marketing Dive reported that Version 2 still focuses disclosure on consumer-facing cases: synthetic images or video, digital twins of living or deceased people, some synthetic voices, and conversational agents used in advertising. It still says standard editing and clearly stylized or fantastical imagery do not need a label (Marketing Dive).
The useful inventory is therefore a change log, not a tool log.
For each customer-facing asset, record four facts:
- What the model changed.
- Whether a reasonable person could treat the result as a real person, place, product, or endorsement.
- Which markets will see it.
- Which statute or platform rule applies in those markets.
If the change is color, crop, or layout, it usually stays off the label. If the change is a face, a voice, a body, or a conversation that looks like a person, it usually does not.
That is classification.
It is also the only way to keep one creative system from generating four conflicting disclosures.
A Label Without An Owner Becomes Label Fatigue
Marketing Dive reported that IAB warns against overusing disclosures because they can produce label fatigue and make the marks less effective (Marketing Dive).
That warning is an operations problem dressed as a design problem.
If every asset gets a mark, the mark stops meaning anything. If no asset gets a mark until legal reviews a one-off, the first market with a live statute will write the rule under a deadline.
The middle path is an owner.
Name one person who can decide, for a given asset and market, whether the AI change is material. Give that person the inventory, the IAB test, and the local statute. Do not give them a mandate to decorate every file.
IAB is explicit that its framework is industry guidance. It does not override New York, California, South Korea, or the EU. A company that copies the sparkle icon into every market has not read the document it is citing.
The Operating Rule
The old default was silence. If a human approved the ad, the file was treated as human work.
That default is gone in several markets and under review in the rest.
The new default cannot be "label everything." IAB already rejected that, and Marketing Dive reported the trade body is trying to keep transparency from collapsing into noise.
The usable rule is narrower.
Classify the asset by the change. Disclose when the change would alter what a reasonable person believes is real. Keep a market-level record of which mark, text, or metadata the statute actually requires. Leave routine production unlabeled on purpose, not by neglect.
That is not a creative flourish.
That is a disclosure policy.
Magnet helps marketing teams turn platform guidance and local rules into an asset inventory, an owner, and a decision record that survives the next icon change. We map the creative system, the markets it ships to, and the proof a reviewer can inspect.
Build the disclosure policy with Magnet
The Icon Is A Reminder, Not The Decision
IAB gave the industry a clearer line than it had in January. The line is worth using.
It does not replace a statute. It does not replace a market-by-market owner. It does not replace a record of what the model changed.
Treat the sparkle as a reminder that a decision was made.
If no one can show the decision, the label is decoration.
Sources
- Interactive Advertising Bureau, AI Transparency and Disclosure Framework V2
- MarTech, IAB updates standard for disclosing AI in ads
- Marketing Dive, IAB revisits AI disclosure in ads as legal requirements multiply
