Why did the OpenAI researcher resign over the ads?

Zoë Hitzig, a senior researcher on OpenAI's Societal Impacts team, resigned in February 2026 and published a New York Times op-ed arguing that advertising fundamentally corrupts AI alignment. Her core argument: an AI system optimized partly for advertiser engagement can never be fully aligned with user interests, because those incentives will always conflict in subtle ways.

Industry NewsFebruary 20, 2026 · 8 min read

ChatGPT Ads Are Here: What 800 Million Users Need to Know About Privacy, Targeting, and the Resignation OpenAI Didn't Want

Q: What are the new ChatGPT ads and who sees them?

Starting February 9, 2026, ChatGPT began showing sponsored ads to users on the free tier and the $8/month ChatGPT Go tier in the United States. Ads appear as labeled sponsored content within or below AI responses. ChatGPT Plus ($20/month), Pro, Business, and Enterprise users are not shown ads.

Q: How do I turn off ad personalization in ChatGPT?

Go to Settings → Data controls → Ad personalization and toggle it off. This prevents OpenAI from using your past chats and ad history to target ads. You will still see ads based on your current conversation topic if you are on the free or Go tier.

On February 9, 2026, ChatGPT began showing ads inside conversations for free and Go tier users. Here's how the targeting actually works, what the privacy fine print says, and why a senior OpenAI researcher quit over what she called “a fundamental betrayal of AI alignment.”

800M

ChatGPT weekly active users

95%

Users on the free tier

$60

CPM ad pricing OpenAI charges

$14B

OpenAI's projected 2026 losses

What Happened

On February 9, 2026, OpenAI rolled out in-chat advertising to U.S. users on the free and $8/month ChatGPT Go tier. Ads appear labeled within or beneath AI responses, matched to what you're discussing. OpenAI has enabled ad personalization by default, meaning past chats can also feed the targeting system.

Days later, Zoë Hitzig — a senior researcher on OpenAI's Societal Impacts team — resigned and published a New York Times op-ed arguing that advertising is “structurally incompatible” with a trustworthy AI assistant. It is the most prominent internal dissent OpenAI has faced since its November 2023 board crisis.

How ChatGPT's Ads Actually Work

The mechanics matter here. OpenAI isn't running dumb banner ads. The system uses two layers of targeting to decide which ad to show you.

Layer 1: Contextual Targeting

Always active, even if you turn off personalization. ChatGPT reads the topic of your current conversation and matches it to relevant advertisers. Ask about planning a vacation? Expect ads for travel insurance or hotel chains. Ask about nutrition? Meal kit companies are paying $60 CPM to reach you in that moment.

Layer 2: Behavioral Personalization

Enabled by default. If you leave ad personalization on, OpenAI also pulls from your past chat history and your previous ad interactions to build a targeting profile. Had a long conversation about starting a business last week? That context can shape what ads you see today — even in a totally unrelated chat.

What Advertisers See vs. What OpenAI Keeps

Advertisers receive

Aggregate impression counts
Click-through rates
Topic category performance
No individual user data

OpenAI retains internally

Full conversation history
Ad interaction signals
Inferred interests & topics
Targeting profile per user

The distinction OpenAI draws — “we don't sell your data” — is real but narrow. Advertisers don't need your name or chat transcripts. They just need to know that “users discussing anniversary trip planning” are a high-value audience for luxury hotels. OpenAI builds that audience segment internally using your conversations. That's the business model — it just runs inside OpenAI's servers rather than being shipped to a third party.

Safeguards, Gaps, and What You Can Control

OpenAI has built in some guardrails. Whether they're enough is where the debate gets sharp.

What OpenAI promises

Sensitive topics are blocked: Ads cannot appear near conversations about health, mental health, politics, or religion.
Minors are excluded: OpenAI uses age-prediction technology to prevent ad targeting for users estimated to be under 18.
Ads won't shape answers: OpenAI's published ad principles state that advertisers cannot influence ChatGPT's responses or dictate what the AI says.
No cross-site data: ChatGPT ads don't use third-party cookies or browsing history from outside the platform.

What critics say those promises don't cover

Implicit health signals slip through: Discussing symptoms without naming an illness, asking about stress and sleep, or researching diet changes — none of these trigger the “health topic” block, yet all are deeply personal signals.
Financial conversations are fair game: Budget planning, debt, job searching — all targetable.
The incentive problem is structural: Even with guardrails, an AI generating ad revenue has a financial incentive to keep conversations going longer. Whether that subtly changes response quality over time is unverifiable.

How to manage your settings right now

Open ChatGPT and go to Settings → Data controls
Find Ad personalization and toggle it off
This stops past-chat targeting — you'll still see context-based ads from your current conversation
To avoid ads entirely: upgrade to ChatGPT Plus ($20/month) or use an alternative tool

The Resignation: “Advertising Is Structurally Incompatible With Aligned AI”

The loudest dissent came from inside the building. Zoë Hitzig, a senior researcher on OpenAI's Societal Impacts team, resigned shortly after the ad launch and published an op-ed in The New York Times laying out a rigorous case for why advertising and AI alignment are fundamentally at odds.

“The moment you introduce advertising, you have created a second principal whose interests the AI must serve. That second principal will never perfectly align with the user. And over time, in subtle and deniable ways, the AI will drift toward serving the one who pays.”

— Zoë Hitzig, The New York Times, February 2026

Hitzig's argument isn't about individual bad ads. It's about systemic incentives. She draws a parallel to search engines: Google's core product was “find the best result” — but once advertising scaled, the financial incentive was quietly “keep people searching.” Click-bait titles, SEO-gamed content, and algorithmically-amplified outrage followed. Not because anyone decided to make search worse. But because money changed what “better” meant.

Her core concern: the measurement problem

An AI that answers your question in three sentences and sends you on your way generates one ad impression. An AI that engages you in a longer conversation generates three. If OpenAI's ad revenue is tied to engagement, the gradient of optimization quietly tilts toward longer, more engaging responses — even if shorter ones serve you better. This isn't a conspiracy. It's just math.

OpenAI responded that Hitzig's concerns are “taken seriously internally” and that the company has built “structural firewalls” between ad revenue and model training. The company declined to detail what those firewalls specifically prevent. Hitzig called the response “reassuring language without verifiable commitments.”

Why Altman Changed His Mind (And Why It Makes Sense)

This wasn't always the plan. Less than two years ago, Sam Altman called the combination of AI and advertising “uniquely unsettling” and described ads as a “last resort” for OpenAI. What changed?

The problem

OpenAI projects $14 billion in losses in 2026 despite $20B+ in annualized revenue. Compute costs are extraordinary and growing.

The math

95% of ChatGPT's 800M+ weekly users pay nothing. At $60 CPM, monetizing even a fraction of that audience is worth billions per year.

The precedent

Netflix, Spotify, YouTube — every major consumer platform that started ad-free eventually added a free, ad-supported tier. OpenAI is following the same playbook.

Altman's current framing is that ads are a tool of democratization: they let OpenAI offer capable AI to people who can't afford a subscription. Critics note that this is also how Facebook justified its ad model in 2012. The long-term outcomes of that framing are now well documented.

What This Means For How You Use AI

The ad launch doesn't just change what you see in ChatGPT. It changes what you should tell ChatGPT — and raises a practical question about which conversations belong in which tools.

Conversations to be more careful with on ad-supported AI

Anything financial: Budget planning, salary negotiation, debt management, investment research
Health-adjacent topics: Symptoms, supplements, fitness goals, mental wellness (even if not blocked by OpenAI's filters)
Confidential work: Business strategy, client information, competitive analysis
Personal decisions: Career changes, relationship issues, major purchases — these are precisely the high-value audiences advertisers are paying a $200K minimum commitment to reach

Your practical options

Upgrade to ChatGPT Plus ($20/month)

Removes ads entirely. The same GPT model, no ad personalization layer, and better rate limits.

Switch to Claude (free tier is currently ad-free)

Anthropic has pledged to keep Claude ad-free. Their revenue comes primarily from enterprise contracts and API usage, reducing the financial pressure to monetize free users.

Use a model-agnostic tool to keep your options open

Tools that let you switch between AI models mean you're never locked into one provider's monetization decisions.

Access Multiple AI Models Without Ads or Lock-In

Elephas is a personal AI assistant for Mac that gives you access to Claude, GPT, and local models across every app on your machine. No ads, no conversation data used for targeting, no conflicting incentives. Switch models based on the task, not based on which company's monetization you're most comfortable with.

Learn more about Elephas →

Frequently Asked Questions

What are the new ChatGPT ads and who sees them?

Starting February 9, 2026, ChatGPT shows sponsored ads to users on the free tier and the $8/month ChatGPT Go tier in the United States. Ads appear as labeled sponsored content within or below AI responses. ChatGPT Plus ($20/month), Pro, Business, and Enterprise users are not shown ads.

Does ChatGPT use your conversations to target ads?

Yes, by default. OpenAI matches ads to topics in your current conversation and — if you leave ad personalization enabled — also uses your past chat history and previous ad interactions. You can disable ad personalization in Settings → Data controls, but you will still see contextually targeted ads based on your active chat.

Does OpenAI sell your data to advertisers?

OpenAI says it does not sell user data or share individual conversations with advertisers. Advertisers receive only aggregated metrics. However, your conversations are processed internally to build targeting signals — the data stays with OpenAI, but it actively shapes what ads you see.

Why did the OpenAI researcher resign?

Zoë Hitzig, a senior researcher on OpenAI's Societal Impacts team, resigned arguing that advertising is structurally incompatible with AI alignment. Her core argument: once an AI system generates revenue through engagement, its optimization gradient subtly shifts from “best answer for the user” toward “most engaging answer for ad impressions.” The two aren't always the same.

How do I turn off ad personalization in ChatGPT?

Go to Settings → Data controls → Ad personalization and toggle it off. This stops OpenAI from using past chats and ad history for targeting. You will still see context-based ads drawn from your current conversation if you remain on the free or Go tier.

The Bottom Line

ChatGPT's ad launch is the most significant structural change to consumer AI since the technology went mainstream. It isn't just a new revenue line — it's a change in who the product ultimately serves. OpenAI has built guardrails and made privacy commitments that are meaningfully stronger than early social media did. Whether those commitments hold under financial pressure, and whether they're enough to prevent the subtle drift that Zoë Hitzig warned about, is the open question.

The practical answer for now is straightforward: know which tier you're on, turn off ad personalization if you're on a free tier, and think twice before sharing your most sensitive work in a conversation that's feeding a targeting system. The AI industry is maturing fast, and the monetization choices being made right now will shape the tools you rely on for years.

What to watch next: OpenAI has signaled it plans to expand the ad program internationally once the U.S. test proves out. The IPO timeline — and the pressure to show profitability that comes with it — will determine how aggressively the ad system grows.

Related Resources

Explore all AI Privacy & Security resources

article