By Gonzalo Mon, Partner at Kelley Drye & Warren LLP.
Many companies want to boast about the AI features in their products. AI is exciting and it makes for good marketing, but the National Advertising Division (“NAD”)—the advertising industry’s self-regulatory watchdog—has made clear that the same truth-in-advertising rules that apply to low-tech products also apply to AI. In recent decisions involving companies in different industries, NAD has sketched out a practical roadmap for how advertisers should (and shouldn’t) talk about AI. Here are four lessons from those cases.
Distinguish Available Features from Ones That Are Coming
Apple likes a splashy launch. In September 2024, it debuted the iPhone 16 with claims that Apple Intelligence was “available now,” touting features like Priority Notifications, Image Playground, Genmoji, and a ChatGPT integration with Siri. There was one small problem: those features were not, in fact, available now. Apple argued that a disclosure at the bottom of the page noting that “some features will become available in software updates in the coming months” clarified the claim. NAD disagreed. An “available now” claim must be supported at the time it’s made, and any limited availability must be part of the main claim itself, not buried in fine print that contradicts the headline.
The NAD decision turned out to be the least of Apple’s worries. Apple faced eight class action lawsuits (later consolidated into one) on the same issue, and in May 2026 plaintiffs asked a California federal judge to approve a $250 million settlement resolving the claims.
The fix isn’t complicated. When Apple later added a prominent disclosure directly under each forthcoming Siri feature explaining that it would be “available with a future software update,” NAD found that acceptable. The takeaway: companies can generate buzz for features that are coming soon, but the ads have to say “coming soon,” clearly and up front. Making that distinction in a footnote may not be enough to escape scrutiny—or a nine-figure settlement.
Don’t Overstate AI Capabilities or Benefits
A consumer electronics company that makes a smart refrigerator advertised that its AI feature “automatically recognizes” what’s in your fridge so you “always know” what’s inside. In reality, the technology recognized 37 specific food items, only in the main cavity, only when they’re unpackaged or in clear containers. Words like “always” and “automatically” convey a message of no material limitations—a message the advertiser could not support. The advertiser agreed to discontinue those claims.
Another consumer electronics company that makes a smart baby monitor advertised that it’s AI features could help parents “rest easy” because the Virtual Fence feature would alert them “if they’re on the move” and that Stand Detection would “ensur[e] their safety.” NAD found that these claims could reasonably convey that the technology itself provides safety assurances, rather than simply providing information for parents to act on. When the testing behind those capabilities was based on limited sample sizes and lacked documented methodology, NAD recommended discontinuing safety-assurance language altogether.
The common thread: AI that works some of the time, or under certain conditions, can’t be advertised as though it works all of the time and under all conditions.
Clearly and Conspicuously Disclose Material Limitations
Microsoft advertised that Copilot can assist users by working “seamlessly” across all their data and that Business Chat helps ground “prompts in work and web data in the flow of work.” NAD thought these phrases conveyed that “actions across apps are uninterrupted or continuous, with less manual steps required.” Although that was true in many cases, manual steps are required for Business Chat to produce the same results as Copilot in a specific Microsoft Office app. NAD recommended that Microsoft disclose material limitations on how Business Chat works. (Notably, NAD did not recommend disclosures for limitations it didn’t think were material.)
Similarly, in the Horizon Brands case, NAD found that camera-distance requirements and minimum-pixel thresholds for the baby monitor’s detection features were material limitations that needed to be clearly and conspicuously disclosed. If an AI feature only works well under specific conditions, the ad needs to say that, prominently and in close proximity to the claim.
Demonstrations and Testing Must Reflect Real-World Use
When Google launched its Gemini AI model in December 2023, it released a slick video that appeared to show Gemini responding in real time to voice prompts and live video input. In reality, the demo used still images and text prompts, and outputs were shortened and latency reduced. NAD suggested that brief on-screen disclosure stating “sequences shortened throughout” was not enough to prevent the video from conveying the message that Gemini could do things it could not actually do the way it appeared to do them. Google voluntarily unlisted the standalone video so it would no longer appear in search results.
The lesson: product demonstrations are a legitimate way to show off AI capabilities, but they must accurately depict what ordinary consumers will actually experience. If you’ve enhanced the demo, that enhancement must be clearly disclosed—or, better yet, eliminated.
The Bottom Line
These rules aren’t new; they’re just being applied to new technology. Distinguish between what your AI can do today and what you hope it will do in the future. Support your claims with evidence that matches the boldness of the promise. Disclose limitations that matter to a buying decision. And make sure your demos reflect reality. AI may be the most exciting marketing story in a generation, but even the best stories have to be true.
