Skip to content
OpenAIOpenAI

Episode 13 - The Thinking Behind Ads in ChatGPT

How should advertising work in an AI product? Asad Awan, one of the ad leads at OpenAI, walks through how the company is approaching this decision and why it’s testing ads in ChatGPT at all. He explains how ads are built to stay separate from the model response, keep conversations with ChatGPT private from advertisers, and give people control over their experience. Chapters 00:00:29 — Mission and principles 00:04:01 — Separation between ads and answers 00:07:31 — Who will see ads 00:08:52 — Internal input and decision-making process 00:11:06 — Controls and how ads will work 00:15:53 — Guardrails for sensitive conversations 00:17:33 — Skepticism about ads 00:20:26 — Helping small businesses 00:24:13 — Future of ads

Asad Awanguest
Feb 9, 202625mWatch on YouTube ↗

CHAPTERS

  1. Why OpenAI is adding ads to ChatGPT now: mission, access, and funding higher limits

    Andrew Main frames the episode around how ads will work in ChatGPT and how OpenAI plans to preserve user trust. Asad Awan explains that ads are a practical way to fund broad access to “the best version” of ChatGPT—especially higher usage limits—while aligning with OpenAI’s mission to benefit all of humanity.

    • Ads positioned as a proven model to scale consumer products to very large audiences
    • Goal: provide higher limits and better models to more people, not a crippled free tier
    • Ads should be helpful to both users and businesses, not just revenue-generating
    • Core premise: trust must remain central even with monetization
  2. Principles for ads in ChatGPT: independent answers, privacy, transparency, and incentives

    Asad outlines the principles OpenAI wants to publicly stand behind before rolling out ads. He emphasizes that answers must remain independent, sensitive conversations must not receive ads, and the company must avoid incentives that degrade the product into “empty calories.”

    • Answers must be independent from ads (visually and technically)
    • Private conversations aren’t shared with advertisers; matching happens internally
    • Sensitive conversations will never have ads and won’t be used for ad matching
    • Incentives should prioritize user value over engagement/time spent
  3. The “wall” between ads and answers: the model can’t see ads unless you opt in

    The discussion drills into the separation: the ChatGPT model does not know what ad is being displayed and will say it doesn’t know if asked. Users can explicitly choose to bring an ad into the conversation via a dedicated action, similar to pasting a link for analysis.

    • Model is unaware of the ad slot content by default
    • Clear visual separation (e.g., distinct banner area) to reduce confusion
    • Explicit user action: “Ask ChatGPT about this ad” to discuss it in-chat
    • Goal is to prevent perceptions of collusion between ads and responses
  4. Preventing long-term drift: why OpenAI says trust is the core business

    Andrew raises the concern that ad-revenue pressures could erode the wall over time. Asad argues that OpenAI’s long-term product strategy (including future devices and enterprise use) requires deep trust, making the separation non-negotiable rather than optional.

    • Trust framed as the company’s central asset across consumer and enterprise
    • Long-term assistant/agent relationship requires users to share sensitive context safely
    • Contrast with other models (search/content feeds) where trust may be less central
    • Claim: incentives are designed so the company cannot “drift” without harming its core
  5. Who will see ads: free and Go tiers; no ads for Plus, Pro, or Enterprise

    Asad clarifies where ads will appear across ChatGPT’s tiers. Ads are limited to Free and Go, while Plus, Pro, and Enterprise remain ad-free—positioning subscriptions and enterprise contracts as alternative funding models.

    • Ads shown on Free and Go tiers
    • No ads for Plus, Pro, and Enterprise users
    • Enterprise context explicitly kept ad-free due to data sensitivity and contract model
    • Ads framed as enabling generous free usage limits at scale
  6. How internal decisions get made: debates, roundtables, and a trust-first rubric

    Asad describes OpenAI’s internal process as research-driven, with extensive debate and broad input across the company. He introduces a hierarchy for decision-making: user trust first, then user value, then advertiser value, then revenue.

    • Company-wide input (hundreds of roundtables) informed ad principles
    • A simple but “deep” rubric guides trade-offs and governance
    • Priority order: user trust > user value > advertiser value > revenue
    • Example: even “effective” ads are unacceptable if they create ‘creepy’ mistrust (e.g., mic/listening concerns)
  7. User controls and personalization: what data is used, and how you can opt out

    The conversation shifts to what users will see and control. Asad argues personalization can make ads genuinely useful, but only if users can transparently understand what data is used and can restrict or disable it—including clearing data and turning off personalization entirely.

    • Transparency: users can see what data is used for ads
    • Controls over whether past chats can be used for personalization
    • Sensitive chats are excluded by default from use and ad matching
    • Options include clearing data, disabling personalization, and upgrading to remove ads entirely
  8. Ad frequency and placement: conservative rollout and “only show an ad if it’s useful”

    Asad explains that early tests will show few ads and that OpenAI won’t force ads where there’s no good match. The governing idea is usefulness and relevance rather than maximizing impressions or time spent.

    • Early rollout aims to be conservative with low ad volume
    • If no good match exists, OpenAI would rather show no ad
    • Focus on high quality/relevance of both the conversation and the ad content
    • Avoids optimizing for “time spent” or excessive ad load
  9. Guardrails for sensitive conversations: definitions, detection, and enforcement

    Asad details how OpenAI identifies and handles sensitive contexts (e.g., health, politics, violence). He describes a policy-driven approach with rigorous definitions, internal/external review, and model-based detection to prevent ads from appearing or being matched in those contexts.

    • Sensitive categories include health, politics, violence, and more
    • Policies are defined rigorously and reviewed with internal/external partners
    • Models classify conversations with high precision to enforce exclusions
    • Sensitive contexts are filtered so ads aren’t shown and aren’t used for matching
  10. What ads will look like: balancing native feel with unmistakable separation

    The episode returns to product design: ads should not feel jarring, but must remain clearly distinct from answers to preserve trust. Asad says OpenAI is starting with a conservative, clearly separated format and will evolve based on learning and data.

    • Design tension: native integration vs. clear separation
    • Initial approach favors conservative clarity and visual distinction
    • Principle remains constant: answers must not be confused with ads
    • Formats may evolve over time as OpenAI learns what works
  11. Addressing skepticism: rebuilding confidence in an industry with a ‘creepy’ legacy

    Asad responds directly to “no ads” commenters, acknowledging the ad industry’s history as a reason for distrust. He argues OpenAI must earn trust through stronger principles, transparency, and user choice, while keeping a paid option for those who prefer no ads.

    • Skepticism seen as valid due to broader industry practices and privacy concerns
    • OpenAI’s obligation: clearer rules, better transparency, and better controls
    • A paid upgrade path remains a legitimate choice for ad-averse users
    • AI can potentially make ads more useful and less spammy if governed well
  12. Helping small businesses: making ads simpler, more agent-like, and less wasteful

    The conversation pivots to advertiser value, especially for SMBs that lack performance marketing expertise. Asad describes a future where businesses can express goals in plain language and ChatGPT-like agents can run experiments, suggest bids, and optimize within constraints.

    • Today’s ad platforms are complex, requiring specialists and risking wasted spend
    • Vision: goal-driven setup (“sell more shoes in the Midwest”) with agentic optimization
    • Automation could reduce SMB costs and broaden access to effective advertising
    • Better matching could help niche products find the right audiences
  13. The future of ads in an agentic world: conversational discovery and behind-the-scenes deals

    Asad sketches longer-term possibilities: more conversational ad experiences and agentic discovery that proactively surfaces relevant products or discounts. He emphasizes that future evolution must still preserve control, understandability, and trust as systems become more capable.

    • Potential shift to conversational ads that users can explore and understand
    • Agentic systems could aggregate deals/discounts and surface discoveries proactively
    • Discovery is two-way: users search, and businesses want to be discovered
    • Future growth depends on maintaining relevance, control, transparency, and trust

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.