Episode 13 - The Thinking Behind Ads in ChatGPT

Key Concepts

• AGI Democratization: OpenAI’s core mission to make advanced AI accessible to all of humanity.
• Ad Principles: A framework prioritizing user trust, conversation privacy, transparency, and user control in ad implementation.
• Model-Ad Separation: Maintaining a strict separation between the core language model and the advertising system to preserve answer integrity.
• Rubric for Decision-Making: A hierarchical system prioritizing user trust > user value > advertiser value > revenue.
• Agentic Advertising: Future potential for AI-driven, conversational, and proactive ad discovery and personalization.
• Personalization & Control: Providing users with transparency and control over data used for ad personalization, including options to clear data or opt-out.

Ads in ChatGPT: A Deep Dive into OpenAI’s Approach

This discussion with Assad Awan details OpenAI’s strategy for integrating advertisements into ChatGPT, focusing on balancing monetization with maintaining user trust and accessibility. The core driver behind this decision is OpenAI’s mission to “bring AGI to all of humanity,” and ads are viewed as a viable model to fund widespread access and enhanced capabilities.

The Rationale for Ads

Awan explains that introducing ads allows OpenAI to extend the benefits of its most advanced models – including higher usage limits – to a broader audience. Currently, over 800 million people use ChatGPT. Without a sustainable revenue model, providing a high-quality experience to this many users becomes challenging. The alternative – limiting access to powerful models or reducing usage – was deemed less desirable than a carefully implemented advertising system. As Awan states, “If the goal is to truly democratize access, I think ads is a good model.”

Core Principles & Safeguards

OpenAI has established a set of core principles to guide ad implementation, centered around preserving user trust. These principles are:

1. Independence of Answers: Ads must not influence the model’s responses, both in terms of visual presentation and the underlying training data. The model itself is unaware of the presence of ads. If asked about an ad, it will state it has no knowledge of it unless explicitly prompted.
2. Conversation Privacy: Sensitive conversations will remain ad-free, and user data from these conversations will never be shared with advertisers. Matching ads to conversations happens internally, protecting user privacy.
3. Transparency & Control: Users will have clear visibility into how their data is used for ad personalization and will have granular control over this process, including the ability to clear their data entirely.
4. Incentive Alignment: Internal incentives are structured to prioritize user value over simply maximizing time spent on the platform. A single, effective ad is preferred over numerous, irrelevant ones.

Technical Implementation & Separation

A key technical aspect is the complete separation between the language model and the advertising system. Visually, ads will appear as distinct banners, clearly differentiated from the model’s responses. The model itself is “blind” to the ads; it doesn’t know they exist unless a user explicitly asks about them. This separation extends to the data used for training and operation.

Users can actively engage with ads by selecting a button to ask ChatGPT about the ad’s content, treating it like a link from the internet. This interaction is crucial for maintaining the model’s independence.

Decision-Making & Governance

OpenAI employs a rigorous internal process for ad-related decisions, rooted in a company culture that prioritizes research and debate. A “rubric” has been developed, ranking priorities as: User Trust > User Value > Advertiser Value > Revenue. This rubric serves as a guiding principle for all ad-related development and ensures alignment with OpenAI’s core values.

The company has dedicated teams focused on defining and enforcing policies related to privacy and safety, with clear forums for discussion and review.

Ad Experience & Personalization

Ads will be personalized based on user data, but with a strong emphasis on transparency and control. Users can:

• View Data Usage: See what data is being used to personalize ads.
• Control Data Sharing: Specify which past chat data can be used for personalization.
• Clear Data: Completely erase their data, preventing its use for ad targeting.
• Opt-Out: Disable personalization entirely.
• Upgrade: Subscribe to a paid plan (Pro or Plus) to remove ads altogether.

The goal is not to maximize ad impressions but to deliver useful and relevant ads. If a good match cannot be found, no ad will be displayed.

Future of Advertising: Agentic Ads

Looking ahead, Awan envisions a future of more “agentic” advertising, where AI proactively discovers and presents relevant offers to users. This could involve ChatGPT identifying user preferences and suggesting products or services they might find valuable, even if the user hasn’t explicitly searched for them. This future relies on a marketplace where both users and businesses can benefit from targeted discovery.

Addressing Skepticism & Maintaining Trust

Awan acknowledges the skepticism surrounding online advertising and emphasizes OpenAI’s commitment to earning user trust through transparency, control, and a focus on delivering genuine value. He recognizes that maintaining this trust requires ongoing effort and a willingness to adapt based on user feedback. As he states, “It’s easy to start now and say, ‘Oh, yeah. We're going to do the great thing. We'll do it right.’ But 10 years later when there's an entire division in charge of ad revenue might say like well do we need the wall between the model and the ads.” The established principles and internal rubric are designed to prevent this drift.

Platform-Wide Implementation

Ads will be shown to users on the free tier and the “Go tier.” Pro, Plus, and Enterprise subscribers will not see ads. This tiered approach allows OpenAI to balance accessibility with a premium, ad-free experience for paying customers. The company aims to avoid limiting the free tier to a severely restricted experience, instead using ads to support higher usage limits and access to more powerful models.

Notable Quote:

“Is creepy okay if it is good? It's not. We are in the business of trust.” – Assad Awan, emphasizing the paramount importance of user trust in OpenAI’s advertising strategy.

This approach represents a deliberate attempt to navigate the complexities of monetization in the AI space, prioritizing user experience and trust as foundational elements. OpenAI’s success will depend on its ability to deliver on these promises and maintain a transparent and user-centric approach to advertising.