Mark Chen, Chief Research Officer at OpenAI, publicly apologized on the X platform regarding the issue of "paid users seeing Target and Peloton shopping links," stating, "any experience that resembles advertising must be handled with extreme caution, and we did not meet the standards this time," and announced that the relevant recommendation logic has been temporarily disabled until model accuracy and user control options are upgraded.

Event Recap:

- Starting December 3, multiple ChatGPT Plus subscribers received "shopping recommendations" cards when asking technical questions, redirecting to Target's online store. Users took screenshots and questioned, "Why are there ads on a paid interface?"

- On the same day, Nick Turley, product lead for ChatGPT, responded that "there is no ad testing," explaining it was merely "App Platform app recommendations" and "no monetary component." This explanation faced user backlash: "Bruhhh… Don't insult your paying users."

Company Internal Priorities Shift:

- On December 4, CEO Sam Altman issued a "code red" memo, requiring all employees to prioritize improving ChatGPT's core quality, with advertising products and other projects explicitly postponed.

- Fidji Sumo, former executive from Instacart who was in charge of the advertising business, joined in early 2025, but according to sources cited by The Wall Street Journal, her team resources have been reallocated to the model accuracy task force.

Technical Details and Follow-up Measures:

- Chen revealed that the recommendation algorithm was originally intended to "display third-party plugins on the App Platform," but the model mistakenly identified e-commerce websites as "available tools," resulting in the shopping card popping up.

- Repair Plan: (1) Optimize the intent classifier, adding "commercial/non-commercial" confidence thresholds; (2) Launch a user control panel, allowing users to choose between "completely turning off shopping prompts" or "only displaying developer plugins"; (3) Re-launch after internal evaluation, with an unspecified timeline.

Market and Regulatory Impact:

- Paid users showed significant resistance, with the #CancelChatGPT topic reaching 120 million views within 24 hours, and some users stated, "If ads appear again, I will switch to Claude."

- Lawyers pointed out that if platforms insert commercial links without clear notification, it may violate the FTC regulations in the U.S. requiring "paid content to be clearly marked," and OpenAI needs to avoid the suspicion of "stealth advertising."

AIbase Commentary

This "recommendation error" incident highlights the fragility of large models in the "intention understanding → commercial recommendation" chain. Between the subscription model and advertising model, OpenAI has yet to find a balance that doesn't harm user experience. As GPT-5.2 approaches its release, the company has chosen to prioritize core performance before commercialization—this is undoubtedly an urgent correction toward "user-first" for a Silicon Valley unicorn eager to monetize. We will continue to track the roadmap on the control panel and regulatory developments.