The Thinking Behind Ads in ChatGPT

OpenAI is doing the thing every AI-first company swore it wouldn’t: putting ads in ChatGPT. But the way Asad Awan frames it, the decision isn’t about monetization. It’s about who gets to use the best AI.

The Episode

This is Episode 13 of OpenAI’s own podcast, hosted by Andrew Maine. Asad Awan, one of the ad leads at OpenAI, lays out the reasoning, the principles, and the guardrails behind advertising in ChatGPT. The conversation is methodical, almost defensive in tone, clearly designed to get ahead of the backlash. What makes it interesting is the internal rubric Awan reveals and the specific technical separation between ads and the model.

Why Ads, Why Now

The argument is straightforward: 800 million+ people use ChatGPT, and the best way to take the best version of the product to all of them is an ad-supported model. The alternative, Awan argues, is the frustrating version where free users hit five questions and get cut off.

“If there is this elitist view that some people get to use it and some don’t get to use it based on who can pay, I think that itself is a pretty fork in the road in terms of how AI can be valuable to people.”

Ads fund higher usage limits on the free tier. Pro, Plus, and Enterprise users see no ads at all. This is the core trade: you either pay with money or with attention, but either way you get access to capable models with generous limits.

The Rubric: Trust > Value > Revenue

The most revealing part of the conversation is the internal decision-making rubric OpenAI built through “hundreds of rounds” of company-wide debate:

User trust > User value > Advertiser value > Revenue

This ordering sounds obvious, but Awan gives a sharp example of what it means in practice: if an ad is highly relevant and the user clicks and buys, but then wonders “was this app listening to me? Is the mic on?”… that’s a user value win but a user trust failure. Under OpenAI’s rubric, that scenario is not acceptable.

“Is creepy okay if it is good? It’s not.”

The rubric cascades downward through the organization. When any team proposes a product change, the first question is: does this fit the rubric?

The Wall Between Model and Ads

The technical separation is absolute:

• The model has no knowledge of what ad is being displayed. If you ask ChatGPT “what is this ad about?”, it will say it doesn’t know.
• Ad matching happens internally at OpenAI. Advertisers never see user conversations.
• Visually, ads appear as a clearly distinct bottom banner, separate from the model’s response.
• Users can explicitly press a button to “ask ChatGPT about this ad”, which functions the same as pasting any external link into a conversation.

When pushed on whether this wall might erode over time as an ad division grows in power, Awan’s answer pivots to the structural argument: OpenAI’s entire product, from consumer to enterprise to future devices, is built on trust. Unlike companies that are purely ad-supported, OpenAI has multiple revenue streams, so ads never become the sole incentive driver.

Sensitive Conversations and Guardrails

Conversations about health, politics, violence, and other sensitive topics are automatically detected and excluded from ad serving entirely. The data from these conversations is never used for ad matching either.

OpenAI uses its own models for this classification, and Awan claims the precision is the highest he’s seen in any product in his career. The policies are defined by a dedicated internal team and reviewed by both internal and external partners.

User Controls

The spectrum of control available to users:

• See what data OpenAI has on you for ad purposes (transparency)
• Clear your ad-related data entirely, so OpenAI forgets it
• Disable past chat usage for ad matching, while keeping click-based personalization
• Turn off personalization completely, getting only context-based ads
• Upgrade to Pro/Plus to remove ads entirely

The “clear your data” option is notable. Awan points out that essentially no other ad platform offers true data erasure for ad targeting purposes.

The Small Business Vision

Awan sketches a future where running ads is as simple as prompting. A shoe company founder could say “sell more shoes in the Midwest” and the system would run experiments, find optimal bids, identify the right niche audience, and report back with results.

He cites two examples: Allbirds found that every designer at tech companies loved their shoes and targeted that niche precisely, winning Silicon Valley on foot. A vegan instant ramen company found its seemingly impossible niche audience through smart targeting. Both cases required teams of performance marketers. The vision is that AI makes this accessible to any small business without that overhead.

“Today literally a small business has to hire performance marketers which could be one of the biggest costs… the vision would be that it is as easy as just steering and telling what you need from your business.”

The Agentic Future of Ads

Looking ahead, Awan outlines three stages:

1. Current: traditional banner-style ads, conservative placement, clearly separated
2. Next: conversational ads where users can interact with and understand products within the chat
3. Future: agentic ads where ChatGPT proactively discovers products and aggregates deals based on learned preferences, functioning as a marketplace matchmaker between consumer intent and merchant offerings

The ramen example comes back: if ChatGPT knows you like ramen, it could surface a vegan ramen brand you didn’t know existed, with the best available deal, without you ever searching for it.

Closing Notes

A few observations worth sitting with:

• The rubric hierarchy (trust > value > advertiser value > revenue) is easy to state and hard to maintain. The real test isn’t year one; it’s year five when the ad team has 500 people and quarterly targets.
• The “model knows nothing about the ad” architecture is a genuinely novel design choice. In traditional ad-supported products, personalization and content are deeply intertwined. OpenAI is betting they can keep them fully decoupled.
• The competitive framing is interesting. When asked about rivals mocking the ads decision, Awan pivots to mission scope: if you don’t serve hundreds of millions of free users, it’s easy to say you don’t need ads. It’s a subtle dig at Anthropic and others.
• The small business vision is the most compelling long-term argument. If AI can truly democratize ad creation and optimization, the economic impact on SMBs could be significant, far beyond what it means for OpenAI’s revenue.
• The elephant in the room: this is an OpenAI-produced podcast. Awan is making the best possible case. The absence of hard skeptical pushback is by design.