AI and Platform Companies Are Building an Unaccountable Persuasion Pipeline, and They Are Pointing It at Children
FOR IMMEDIATE RELEASE
August 13, 2026
Contact: Katharina Kopp, 202-836-4621, kkopp@democraticmedia.org
Jeff Chester, 202-494-7100, jeff@democraticmedia.org
AI and Platform Companies Are Building an Unaccountable Persuasion Pipeline, and They Are Pointing It at Children
How generative AI changes commercial persuasion, and why privacy law alone cannot fix it
WASHINGTON, D.C. - The transformation of media and marketing by artificial intelligence poses serious risks to the public, and especially to young people. Platform companies are embedding AI capabilities, including generative, conversational, and agentic AI, deep into advertising, programming, and commerce, fueling a new wave of commercial surveillance and manipulation that reaches all of us and that falls hardest on communities of color. For children and teens, this is not a distant prospect. These systems are already embedded in the products they use every day. A report released today by the Center for Digital Democracy (CDD) documents the harms they pose across children's lives, and finds that a category of AI-driven advertising, built on real-time, cohort-level inference rather than on identifiable personal data, falls into a structural gap that privacy and consumer-protection law was not designed to govern.
Generative AI collapses steps that used to be separate. A single system can now produce a commercial message, deliver it as conversation, and adjust it to a particular child in the same moment. It can sort a first-time user into a behavioral cohort, estimate how persuadable she is, and shape what she sees within seconds, drawing on models trained on millions of other people and often without writing any durable record about the child being acted on. Children and teens are among the earliest and heaviest users of these systems, which makes them the most exposed. Sixty-four percent of U.S. teens have used AI chatbots and 30 percent use them daily, with adoption highest among Black and Hispanic teens.
Nothing about the technology required these companies to build advertising into their AI products. They chose to. The report examined nine AI products across nine companies - Meta, Google, OpenAI, Microsoft, Amazon (Alexa for Shopping), Snap, xAI, Anthropic, and TikTok - and found seven of the nine confirmed or likely to serve advertising inside AI chat, to use chat interactions to target advertising elsewhere, or both. Only one, Anthropic, has committed to remaining ad-free. The ninth, TikTok, has no standalone chatbot, though its platform-wide advertising system is AI-driven.
"A teenager talking to one of these systems is on the other side of a model trained on millions of people and optimizing against her as she types," said Jeff Chester, CDD's executive director. "She has nothing to match it with, and no way to see it working. Every exchange teaches the system how to sell to the next child a little more effectively. That is not a loophole someone stumbled into. It is the business model these companies chose."
The Persuasion Pipeline: How Commercial AI Targets, Shapes, and Sells to a Generation of California's Children and Teens, by CDD Deputy Director and Director for Policy Katharina Kopp, Ph.D., traces commercial influence through six connected layers - data collection, profiling, targeting, creative generation, interaction, and measurement - and finds AI now running all six and linking them in feedback loops that get better at influencing children with every cycle.
Where these practices rest on personal data about an identifiable child, privacy law does apply, though first-party exceptions can exempt a platform's own advertising even then. The difficulty is what happens when there is no such record for a right to attach to. Comprehensive federal privacy legislation remains necessary and overdue, but law anchored to the data a company holds about an identifiable person will not by itself close this gap. And while more than three dozen states have adopted over a hundred AI laws, none takes the AI-driven advertising and marketing pipeline as its object, and none requires a company to disclose what these systems are built to maximize.
"These companies have built the most sophisticated persuasion infrastructure in history and are pointing part of it at children," said Kopp. "We still need a strong federal privacy law and we should pass one. But privacy law alone will not reach this. Lawmakers have to govern what these systems are able to do, not only what they store, and judge companies by their effects on children rather than by what they claim not to have known."
The burden is not evenly distributed. The report finds that youth of color and lower-income families face disproportionate exposure at every layer of the pipeline: higher rates of data collection driven by more intensive platform use, profiling that sorts users along racial and socioeconomic lines through behavioral proxies, optimization that steers unhealthy product advertising toward Black and Hispanic audiences, AI-generated creative that embeds raciolinguistic stereotypes, and measurement systems that compound these biases with every cycle. AI did not create these disparities. It formalizes them and runs them at a speed and scale manual marketing could not reach.
What This Does to Children
The report documents eight harms, moving from what these systems hold about a child to what they can do to her in the moment. They include children's private disclosures converted into commercial data, discriminatory outcomes produced by optimization, whatever the intent, manipulation through trust the systems are built to earn, emotional exploitation as a business model rather than a side effect, influence pitched below the level a child can recognize, and identity shaped around commercial values. The last two fall on children as a group - the saturation of childhood by commerce and the erosion of independent choice, whose effects compound across a generation and distort the market itself.
This Is Already in the Products Children Use
In April 2026 Fenty Beauty launched an AI beauty advisor inside WhatsApp, built in the brand's own words to feel "like a text from a friend." A teenager describing her skin concerns receives product recommendations mixed with creator videos and fan reviews, with no obvious advertisement to recognize. What she does not see is the commercial partnership routing her questions through WhatsApp's business tools, or that Meta has used interactions with its AI tools to personalize advertising across its apps since December 2025, with no opt-out.1 Meta says it does not apply that personalization to users it knows to be under 18, whose ad targeting it limits to age and location. That protection depends entirely on the company correctly identifying who is a minor.
In February 2026 OpenAI launched sponsored ads in the free version of ChatGPT, with ad personalization on by default. The company says it will not show ads during this test to users it predicts are under 18, a commitment that covers only the testing phase and rests on its own age-prediction model, which no outside party has audited. In April 2026 Snapchat introduced brand-operated chat agents that sit in a user's Chat tab and behave like any other contact. The only signal that an exchange is an advertisement is a single light-gray "Ad" label. Across all nine platforms the report's appendix examines, the capacity to infer who a young user is, sort them into commercial cohorts, and optimize against them in real time is nearly universal. Almost none of it is visible in what the companies disclose, and the protections that do exist are narrow, self-defined, and concentrated where legal pressure has already been applied. Disclosure should be required, but it is a floor. The harm comes from what these systems are built to do, which is why the law has to reach the capability itself.
What the Report Asks For
The report proposes four principles for legislation, framed as obligations on the companies rather than as restrictions on young people's access or new burdens on parents.
1. Make commercial influence legible. Require companies to disclose what their systems are built to maximize, including the advertising-optimization layer, so that regulators, researchers, and advocates can scrutinize what these systems are designed to do.
2. Prohibit the most harmful capabilities outright. Bar advertising directed at children in these environments, or at minimum advertising targeted or personalized to them, along with optimizing for persuadability, inferring emotional states for commercial purposes, and running the closed measurement loop.
3. Replace knowledge standards with outcomes-based accountability. Trigger obligations on what a system does to minors as a group, not on what a company admits it knew about any one of them.
4. Place accountability on the more powerful party. Cut the closed-loop data flows, so targeting data does not flow in and what is generated inside does not flow back out. Reject consent carve-outs and safe harbors, and require independent audits of commercial influence in model behavior.
The report is addressed to California, home to roughly 8.4 million children and to the companies building these systems, and its framework is written so that Congress and other states can adopt it. A December 2025 survey found 60 percent of Americans, across party lines, want more regulation of AI.
Statement from Senator Edward J. Markey (D-Mass.)
"Big Tech should not be allowed to turn children's vulnerabilities into fuel for an AI-powered persuasion machine," said Markey. "This report makes clear that AI companies are building systems that can surveil young people, predict what will move them, and blur the line between trusted conversation and commercial manipulation. We need strong rules of the road that put children's privacy and well-being ahead of corporate profits. The Center for Digital Democracy has provided an important service by sounding the alarm on these AI-driven dangers to our young people. I am committed to ensuring that AI does not become yet another technology that Big Tech uses to exploit kids for profit."
The Persuasion Pipeline, with an executive summary in English and Spanish and a platform-by-platform appendix, is available at https://democraticmedia.org/reports/the-persuasion-pipeline. Findings are current as of June 2026.
The Center for Digital Democracy is a Washington, D.C.-based public interest research and advocacy organization working on behalf of citizens, consumers, communities, and youth to protect and expand privacy, digital rights, and data justice. CDD's project on commercial AI in advertising and marketing aimed at young people is funded, in part, by the Rose Foundation for Communities and the Environment.
###
Note to editors
1. Fenty Beauty, "Rose Amber" AI advisor, WhatsApp launch, April 2026, as reported by Glossy and Digiday. On Meta's use of AI interactions for ad and content personalization: Meta, "Improving Your Recommendations on Our Apps With AI at Meta," October 1, 2025, effective December 16, 2025, about.fb.com.