OpenAI 已在美国开始为 ChatGPT 免费方案投放广告，Reece Rogers 在行动版中以500个问题做测试。依其非正式抽样，约每五个问题就会在回应下方出现一次广告，即约20%，而且每则广告都以网站连结按钮呈现并多半依据提问主题。广告覆盖范围很广，从Uber、Booking.com到狗粮、印表机、旅馆预订、软体与旅游服务都有出现。旅游相关提问似乎特别容易触发广告，例如询问规划Palm Springs行程时即推荐订房。OpenAI目前仍以试点方式推进，并表示广告放置、格式与密度会依回馈持续调整。
这段叙事与OpenAI领导层早前的立场形成对比，尤其是Sam Altman在2024年公开称在聊天机器人中加入广告『令人不安』并视为最后手段的获利方式。然而OpenAI已转向广告支援模型，作为维持产品广泛可及性、并精简非核心计划（例如停运Sora及取消情色版ChatGPT）的一环。公司于2月开始测试，作者3月初即察觉，并声称广告不会改写回答内容，广告主也不会看到完整对话纪录。广告投放据称会根据当前问题、近期聊天脉络与系统记忆的使用者讯息进行匹配。行销学者Olivier Toubia与Stefano Puntoni都认为，这代表传统搜寻广告数十亿美元正在转向一个新的数十亿级AI广告市场。
此轮推广的商业回报必须与信任风险同时衡量。OpenAI承认免费用户成本高昂，若广告过度激进，可能推动使用者转向Gemini、Claude或其他模型；目前主要竞争者尚未在输出中内建赞助按钮，但Google未排除未来采用。当使用者在提示中提及DoorDash或Netflix时，系统也曾回传其直接竞争者广告，这是被称为poaching的既有数位广告手法。OpenAI在3月26日更新中指出信任指标未受负面冲击且广告关闭率低，并计划扩展至加拿大、澳洲及纽西兰。但Puntoni警告，如果精准度或信任防护失效，结果可能是使用者外流，因为使用聊天机器人最终依赖的是信任。
OpenAI has begun serving ads to ChatGPT free-tier users in the US, and Reece Rogers tested the new format by asking 500 questions on the mobile app. In his informal sampling, an ad appeared at the bottom of responses in roughly one out of every five questions, about 20%. Ads were always link-style buttons and were generally aligned with the prompt topic, spanning a broad set of categories from Uber and Booking.com to dog food, printers, hotel reservations, software, and travel services. Travel queries seemed especially likely to trigger ads, such as a Palm Springs trip request that surfaced booking options. The rollout is still in pilot mode, and OpenAI says ad placement, format, and density may continue to change as the system iterates.
The rollout contrasts with earlier leadership rhetoric, especially Sam Altman’s 2024 view that chatbot ads were “unsettling” and should be a last-resort business model. Nevertheless, OpenAI is now adopting an ad-supported model as part of a broader effort to keep services broadly accessible while shedding non-core initiatives, including discontinuing Sora and shelving a sexual-content version of ChatGPT. The company began testing in February and the author noticed it in early March; OpenAI says ads do not alter answer content and advertisers do not receive full chat logs. Targeting is said to use the current query, recent conversational context, and stored memory. Marketing scholars Olivier Toubia and Stefano Puntoni describe this as a shift of search-ad spending from billions toward an emerging multi-billion-dollar AI advertising market.
The financial upside is balanced against a trust risk. OpenAI says free users are costly, and aggressive ad design could drive users to Gemini, Claude, or other chat models. Competitors currently do not include sponsored buttons in outputs, though Google has not ruled that option out. Some ads even showed a brand competitor when prompts mentioned DoorDash or Netflix, a practice known as poaching in digital ads. In a March 26 update, OpenAI reported no consumer trust decline and low ad-dismissal rates, and said it would expand to Canada, Australia, and New Zealand. Puntoni warns that if relevance and trust safeguards fail, users may migrate away, because trust is the core condition for chatbot adoption.