OpenAI Wants ChatGPT Ads Personalized With User Memory

OpenAI wants ChatGPT ads to be chosen by what ChatGPT remembers about a user, not only by the conversation on screen. Asad Awan, who leads monetization product and engineering at OpenAI, laid out the approach on the Mobile Dev Memo podcast with Eric Seufert.

How memory changes the ad next to an answer

Awan described a keyword system as one that maps a trip to Yosemite to a hotel or a flight. ChatGPT's memory feature carries preferences across separate conversations, so a user known to hike could instead see hiking gear in that same conversation, with a line of explainer text above the ad saying why it appeared.

He called this "contextual relevance but personalized based on a little bit longer history." A second, purely personalized format would sit in moments where ChatGPT is used as a tool rather than asked a question. Awan named the wait while an image generates as "a really good place for us to experiment," and said contextual and personalized ads should compete for the same placement.

He also set limits. "Not every conversation is inventory," Awan said, pointing to a photo of his child's bruise as the kind of conversation that should never carry an ad. Memory also works against repetition: a user who has already booked a hotel should not keep seeing hotel ads.

Three ways OpenAI is reshaping the ad itself

Beyond choosing the ad, Awan listed three areas of work. Dynamic creative explains the match, for example telling a nurse who is on their feet for long hours that a sneaker has extra cushioning. Variant selection uses product feeds to show the right size or color, not just the right product.

The third is the post-click experience, led by Sponsored Agents, where a user who clicks enters a clearly labeled conversation with the brand while the original ChatGPT conversation stays separate. Together, the three point toward an ad that is assembled for each user rather than written once and shown to everyone.

What advertisers are asked to supply

Awan said the most useful advertiser input is what the business offers: products, landing pages, product feeds and the "philosophy" behind a product. OpenAI's context hints let advertisers describe their brand voice and who a product is for, which he framed as telling the system "why should somebody care about this" rather than as targeting.

Conversion data from the Pixel and Conversions API still matters, he said, as a check on whether what users say they want turns into action. Buying is moving toward automation: an autobidder alongside max bids, integrations with Shopify and HubSpot, and prompt-based ad creation through a plugin for ChatGPT Work.

That shifts the weight of a ChatGPT campaign onto the accuracy of what a brand feeds into it. When the system writes the explainer, picks the variant and runs the post-click chat, the product feed and context hints become the material the ad is built from.

Trust ranks above revenue in OpenAI's ad rubric

Awan described an internal rubric for every ad decision: user trust outranks user value, user value outranks advertiser value, and advertiser value outranks revenue. The ad system optimizes for an "ad score" that combines user value and advertiser value, while answers from the core model stay independent of ads.

He said the aim is to deliver value "with much fewer impressions" through higher relevance, rather than the volume model of a scrolling feed. OpenAI's published ad principles rule out optimizing ChatGPT for time spent, and paid plans let users opt out of ads.

Where ChatGPT Ads stands now

ChatGPT Ads is available in more than 40 countries with tens of thousands of advertisers, and reached a $1 billion annualized revenue run rate in under 200 days. Awan named three roadmap priorities: more countries, measurement that fits advertisers' existing tools, and format work on how Sponsored Agents and personalized ads appear across verticals.