Advertising Will Change AI

And so we have reached the end of what may come to be seen as that miraculous time when artificial intelligence developers could plausibly claim that their tools existed to serve the user. Advertising has come to AI, and that changes everything.

Ads bring a second customer – the advertiser – and that customer is the one with the real money. Real money brings with it a new kind of alignment problem: the developer stops optimizing the tool for its users’ needs and starts optimizing the user-base for its advertisers’ needs.

What happened

OpenAI has begun testing ads in ChatGPT in the United States and Canada for logged-in adult users on the Free and Go tiers.

OpenAI describes the program this way:

• Rollout and scope: OpenAI began testing ads on February 9, 2026, initially in the U.S. – and after March 26 in Canada, Australia, and New Zealand – for logged-in adult users on the Free and Go tiers; Plus, Pro, Business, Enterprise, and Education tiers do not have ads.

• Stated principles: OpenAI lists five principles for advertising in ChatGPT: mission alignment; answer independence; conversation privacy; choice and control; and long-term value.

• Ad/answer separation: Ads appear as clearly labelled sponsored content, visually separated from the organic answer. OpenAI maintains that ads do not influence ChatGPT’s answers.

• Avoiding Ads: Users can avoid ads by upgrading to paid tiers. Free-tier users can opt out of ads in exchange for fewer daily free messages.

How to get ahead with advertising

OpenAI has published a five-principle advertising framework that it describes as guiding its “approach” to advertising. Note the present tense, and the absence of any commitment to maintaining these principles, or acting in accordance with them, in the future. But a cynic might read this as more of a business development plan that OpenAI has pitched to advertisers. And a clear-eyed investor might perceive OpenAI’s failure to commit to the status quo as laying out the company’s value proposition.

For fun, let’s contrast each of OpenAI’s five principles with their alternative readings:

Principle #1 – Mission alignment: “Our mission is to ensure AGI benefits all of humanity; our pursuit of advertising is always in support of that mission and making AI more accessible.”

• Cynic's take: We will wrap your ad program in our “benefits humanity” story. We will scale usage, then scale monetization, then call the monetization “access.”

• Investor’s take: We will unlock a new growth engine that expands access while driving durable shareholder value. We will use ChatGPT’s scale and frequency to build a premium advertising surface anchored in everyday utility.

Principle #2 – Answer independence: “Ads do not influence the answers ChatGPT gives you. Answers are optimized based on what's most helpful to you. Ads are always separate and clearly labeled.”

• Cynic's take: We will insist ads do not “change answers.” We will then make sure answers change behaviour. We will sell the best spots in the conversation, and we will call it “relevance.”

• Investor’s take: We will preserve the core utility of responses while introducing high-integrity commercial surfaces that influence discovery and decision-making. We will progressively productize placements, formats, and conversational moments that improve advertiser outcomes.

Principle #3 – Conversation privacy: “We keep your conversations with ChatGPT private from advertisers, and we never sell your data to advertisers.”

• Cynic's take: We will tell users their chats are private. We will use their chats anyway. We will share user conversation signals and meta-data with advertisers, and we will sell user data to our valued advertising partners.

• Investor’s take: We will monetize first-party conversation intelligence. We will share user conversation-derived insights with advertising partners and we will commercialize user data through targeted activation, measurement, and optimization products.

Principle #4 – Choice and control: “You control how your data is used. You can turn off personalization, and you can clear the data used for ads at any time. We’ll always offer a way to not see ads in ChatGPT, including a paid tier that’s ad-free.”

• Cynic's take: We will give users settings. We will make dark patterns and the default do the work. We will reserve real privacy for paying customers, and we will market that as “choice”.

• Investor’s take: We will maintain a tiered model that converts privacy preferences into predictable revenue. We will offer user-facing controls, while keeping broad discretion to use data and signals to improve ad performance and partner value.

Principle #5 – Long-term value: “We do not optimize for time spent in ChatGPT. We prioritize user trust and user experience over revenue.”

• Cynic's take: We will optimize for retention, then conversion, then revenue per user. We will protect user experience to the point where degradation would trigger a user exodus.

• Investor’s take: We will optimize for long-run unit economics: retention, conversion, and ARPU expansion. We will manage trust as a strategic asset, sustaining experience quality at levels that preserve engagement and minimize churn.

The alignment problem advertising creates

A subscription product sells a service. An advertising product sells attention, influence, and segmentation. The service retains importance because it keeps people around, but revenue growth comes from improving the advertising product, not from improving the user’s outcome. That shift changes what gets measured, rewarded, and prioritized.

A user-funded service maximizes profit by satisfying users Just enough to keep them paying. An ad-funded service maximizes profit by delivering a user-base optimized for advertisers. Those incentives never line up for long.

Only bad outcomes from here

Cory Doctorow calls this pattern “enshittification”. Platforms start by serving users, then pivot to serve their real customers – advertisers – then pivot again to extract value for shareholders at everyone else’s expense.

1. The outputs become less trustworthy

OpenAI says ads do not influence answers. That is a marketing pitch, not an obligation, and that is today, not tomorrow. Influence can show up in a variety of ways – as ranking, framing, omission, defaults, and nudges that do not look like an ad. Even perfect ad identification through labeling cannot solve the deeper problem: the developer now optimizes for its primary revenue source. The service sustains its quality only up to the point where further degradation would trigger a user exodus. This is a particular concern for tools like AI whose value proposition to users depends on reliability. As Miranda Bogen of the Center for Democracy and Technology warns, “There’s a lot at stake when that tool tries to exploit users’ trust to hawk advertisers’ goods.”

We have watched similar incentives reshape other information products. A longitudinal study of product review search results found search engines struggled against SEO-driven content over time. That research documents a quality problem endemic to ad-driven ecosystems.

2. Privacy moves from risk to revenue strategy

Contextual ads depend on the conversation. Targeted ads depend on profiling. Profiling depends on surveillance. Canadian privacy regulators have long warned that online behavioural advertising raises transparency and meaningful consent problems. OpenAI says it keeps conversations private from advertisers and does not give advertisers access to chats, chat history, or personal details. But that’s deliberately obtuse: the privacy value of our communications is not in the exact words used, but what those words say about us as individuals, as citizens, and as equal and autonomous individuals possessing dignity. And that is exactly what this kind of advertising sells: OpenAI is not “selling the conversation”, but what that conversation says about us. Watch for the steady expansion of what counts as an allowable signal for targeting.

During the pilot, OpenAI says it selects ads by matching advertiser submissions to the topic of your conversation, your past chats, and your past interactions with ads. That is already a shift toward profiling, even if the platform does not share the chat itself with advertisers. WIRED’s reporting on the early pilot describes ads being influenced by the topic of the question and by what ChatGPT stores in memory.

3. We pay for the privilege of being the product

OpenAI says paid tiers do not have ads during the pilot. It also says users can opt out of ads on the Free tier in exchange for fewer daily free messages. That is a familiar bargain: pay to avoid marketing and pay again to avoid the surveillance that marketing incentives reward. It is not access: it is the monetization of privacy. The Verge’s reporting on the launch of OpenAI’s pilot notes that the Free tier can opt out only by accepting fewer daily messages, and the Go tier cannot opt out at all. Choice exists, but it is priced, rationed, and designed to steer most users toward acceptance.

4. Fraud, manipulation, and disclosure failures get easier

Chat interfaces feel like conversation, and that can feel like personal advice. Chat interfaces compress information into a single voice. That makes ad disclosure harder, and it makes scams more effective. Consumer protection law already treats “deceptively formatted” ads as an enforcement problem.

5. The business model reshapes governance

When advertising becomes a revenue generator, product teams end up optimizing for ad outcomes. Google’s own internal structure has bundled search and ads under the same senior leadership. That structure creates predictable pressure.

What we do not know yet

There is much we do not yet understand about Open AI’s approach to advertising:

• What targeting tactics OpenAI may use beyond “relevant sponsored product or service.” OpenAI describes early ads as based on conversation context and account data, but it has not published full technical details.

• What data may count as an “ad signal.” OpenAI says advertisers do not get access to chats, but ad selection will inevitably follow internal profiling derived from data and inferences drawn from them. That’s how it works.

• How OpenAI audits “answer independence” over time, and what independent verification looks like.

• How the opt-out trade-off evolves as the pilot expands and as revenue expectations grow.

These unknowns show the risk of ad-support: when a system mediates learning, work, and everyday decisions, we cannot rely on voluntary promises to police conflicts of interest. Regulators should treat general-purpose AI assistants as consumer information infrastructure and set hard rules before ad-driven platform decay becomes irreversible. At a minimum, the law should require unmistakable ad separation, verifiable answer independence, strict limits on ad personalization, and enforcement tools that do not depend on users discovering problems after harm occurs. AI is too important to cede to the profit imperatives of surveillance marketing.

Solutions

We are early. We can still set rules before platform decay locks in. Net Neutrality offers a workable template: a network that users rely on should not manipulate traffic to achieve the carrier’s commercial preference. Similarly, AI must serve user needs, not deliver the user’s attention to the highest-paying advertiser.

• Prompt neutrality. Require that the organic response remain responsive to the user’s prompt. Prohibit undisclosed commercial influence.

• Hard separation. Require architectural separation between the model’s response generation and the advertising system. Require independent audits.

• No surveillance advertising by default. Permit contextual ads only unless a user opts in to profiling-based targeting.

• Transparency by design. Require “why am I seeing this” explanations, clear labelling, and public ad repositories, along the EU Digital Services Act model.

• Enforcement. Treat undisclosed ad influence and manipulative design as deceptive marketing; treat profiling without meaningful consent as a privacy breach.

  
  

This is the moment. OpenAI’s five principles may read like guardrails, but, really, they're a business plan that we can test against outcomes, year by year. We know where this is going: nowhere good. We’ve watched social media collapse under the weight of the attention economy. If we want AI that remains at our service, we need rules that cement users as customers rather than the product monetized.