When OpenAI quietly confirmed that ads are coming to ChatGPT’s free tier, it was obvious the business model for consumer AI was shifting. What wasn’t obvious—until this week—is just how messy that shift could get in the eyes of regulators.

Senator Ed Markey, a longtime tech watchdog from Massachusetts, has now fired off letters to OpenAI and six other AI heavyweights—Anthropic, Google, Meta, Microsoft, Snap, and Elon Musk’s xAI—warning that ads woven into chatbot conversations could cross the line into manipulation and “deceptive advertising.” His basic fear: if an AI system already feels like a helpful companion, how do you even tell when it quietly becomes a salesperson?

The immediate trigger is OpenAI’s plan to start testing “sponsored” products and services in ChatGPT for logged‑in adult users in the US on the free and new “Go” tiers, with ads placed at the bottom of relevant responses. Paid Plus, Pro, Business, and Enterprise users stay ad‑free, but for everyone else, the chatbot that used to just answer questions will soon also try to sell you things. OpenAI says the ads will be clearly labeled, won’t be shown to under‑18s, and won’t appear near sensitive topics like physical health, mental health, or politics. It also insists that user conversations aren’t sold to advertisers, and that personalization controls and “why am I seeing this ad?” transparency will be part of the experience.

Markey isn’t convinced those safeguards are enough. In his letters, he points out that chatbots are designed to mimic human‑like conversation, often becoming quasi‑companions for users, especially teenagers and young adults. That intimacy is exactly what makes the ad model so uncomfortable. If a chatbot suggests a product in the middle of a discussion about stress, finances, or relationships, even a clearly labeled “sponsored” tag might not register the way a banner ad or pre‑roll video would. The whole point of conversational interfaces is that they blur the old lines of static content—Markey’s worry is that they may also blur the line between advice and advertising.

There’s also the privacy angle. Markey warns that AI companies must not repurpose “personal thoughts, health questions, family issues, and other sensitive information” for ad targeting. OpenAI has promised not to show ads next to sensitive topics, but its question cuts deeper: will any of that data still influence what you see later, in another chat, on another day? That’s exactly the kind of profiling the Federal Trade Commission has flagged in earlier work on “stealth advertising,” particularly when kids and teens are involved. The FTC has already warned that young users are especially easy to mislead when ads are embedded in what looks like normal content instead of clearly separated commercial spaces.

The broader regulatory backdrop is shifting, too. There’s growing bipartisan anxiety in Washington around AI “companion” tools and youth safety, from proposals like the GUARD Act—which would force chatbots to verify age and regularly remind users they’re talking to a machine—to a wave of state‑level privacy rules that limit targeted ads to minors. Markey’s letters plug right into that trend: he’s effectively asking AI firms to prove that their monetization strategies won’t turn chatbots into another dark‑pattern‑ridden ad funnel built on sensitive data.

On the industry side, OpenAI’s line is straightforward: ads are about keeping the free version of ChatGPT available to as many people as possible, while subscription and enterprise revenue carry the rest of the load. The company frames this as a “diverse revenue model” that keeps intelligence “more accessible to everyone,” and stresses that trust will matter more than raw engagement metrics. In other words, they’re trying to position ads as a necessary, controlled trade‑off—more like search ads next to your query than a sneaky influencer deal embedded in your group chat.

The catch is that AI chatbots aren’t search engines or social feeds. People ask them about everything: health scares, relationship drama, money problems, sexuality, and workplace fears. They share things they might not tell a doctor, a parent, or a friend. That’s where Markey’s language about users’ “emotional connection” matters: when you lean on a tool for support, a product suggestion can feel less like a neutral recommendation and more like advice from a trusted confidant. It’s easy to imagine scenarios where a vulnerable user can’t tell whether the system is optimizing for their well‑being or for an advertiser’s conversion rate.

Markey’s letters lay out a long list of questions he wants answered by February 12th. He’s asking these companies to spell out whether they plan to insert ads directly into conversations, how users will be able to identify them, whether they’ll allow opt‑outs or ad‑free options, and how they’ll prevent targeting based on sensitive queries. He also wants to know how they’re handling kids and teens: what age‑gating looks like in practice, how they’ll stop minors from slipping into ad‑driven experiences, and whether any of the ad tech will lean on minors’ data at all.

This isn’t just about OpenAI, either. By looping in Anthropic, Google, Meta, Microsoft, Snap and xAI, Markey is trying to set expectations for the whole sector before “AI ads” solidify into an industry norm. Some of these players already run huge ad businesses; the idea of folding conversational AI into those engines raises obvious temptations. Imagine an AI‑powered assistant inside a social app that can nudge you toward in‑app purchases or partner stores, or an AI search experience that blurs the line between organic conversational answers and sponsored placements. Once that UX pattern is everywhere, rolling it back will be hard.

So what actually counts as “deceptive” in this space? Regulators have a few frameworks to lean on. Existing consumer protection laws already prohibit unfair or deceptive practices, which can include hiding the commercial nature of content or misrepresenting what data is used for. COPPA covers kids’ data. State privacy laws increasingly target behavioral advertising based on sensitive categories. And the FTC’s prior focus on stealth ads in kids’ content offers a roadmap for how it might approach chatbot‑based promotion if companies push things too far.

That’s why Markey’s move feels like an early test: can AI companies build a sustainable ad business without sliding into those gray zones? The safest path looks boring but clear—aggressively labeled ad slots, strict walls around sensitive conversations, genuinely limited data use for targeting, robust controls, and default protections for young people. Anything more experimental—like ads that feel indistinguishable from organic responses, or personalization that leans heavily on intimate chats—will invite the exact kind of scrutiny Markey is now formalizing.

For everyday users, the impact will be more practical than abstract. If you’re on the free or cheaper plans, your chatbot is about to look a little more like the rest of the internet: useful, yes, but with commerce baked into the edges. The open question is whether this new wave of AI ads ends up feeling like a helpful recommendation section—or like your most trusted digital assistant suddenly dropping a sales pitch into the middle of a very personal conversation. That tension between access and trust is exactly what Markey is trying to force the industry and regulators to confront now rather than later.