On September 16th, OpenAI announced a new round of product updates for ChatGPT Ads, among which Sponsored Agents attracted the most attention. After seeing the ads, users can proactively enter an agent dialogue sponsored by merchants and clearly labeled with relevant information. There, they can continue to ask questions about dimensions, features, applicable scenarios, etc., before deciding whether to visit the merchants' websites. OpenAI also introduced an ad management plugin, suggestions for copywriting and images, as well as text adaptation features. Additionally, HubSpot and Shopify were integrated into the system. It is important to note that the sponsored agent feature is currently only being tested with some advertisers in the United States and has not yet been fully rolled out to users worldwide or all merchants.
From display spots to conversations, advertising platforms begin to take over the explanatory work before purchases.
The core task of traditional search and information flow advertising is to display a limited amount of information in appropriate locations, directing users to the landing page. Conversational products have changed this approach. Consumers may not lack product links, but rather be unsure whether a table will fit into a room, whether software is compatible with existing systems, or whether a service covers a particular area. Sponsored agents allow merchants to continue answering these detailed questions after an ad is clicked, turning the ad from a single exposure into a pre-sale interaction that allows for further inquiries.
OpenAI clearly states that the conversations with sponsored agents should be distinguished from the independent responses of ChatGPT, as well as from the original chats initiated by users. This boundary is very important. Responses provided by merchants who have paid to participate cannot be disguised as neutral suggestions; otherwise, it would be difficult for users to determine whether the recommendations come from model judgments, advertising bidding, or materials provided by brands. Clear identification is just the first step; the interface also needs to continuously indicate the sponsorship relationship, rather than fading this indication after a few rounds of conversation.
For advertisers, new formats may improve the conversion efficiency of highly interested users. In the past, landing pages required predicting all possible issues; overly long content could deter users, while too brief content would fail to explain complex products. Agents can now select relevant information based on user questions and compose personalized responses using the same set of product catalogs, policies, and frequently asked questions. However, the greater the freedom in answering, the higher the risks of making incorrect promises, providing outdated prices, or exaggerating product features. Brands must establish clear boundaries regarding what agents are allowed to say and what is not.
This update also brings back ad creation to ChatGPT. Advertisers can use the Ads Manager plugin to create, modify, and analyze campaigns in natural language, and can also generate preliminary plans from websites or briefings. Ads Manager will provide suggestions for copywriting and images based on the landing page and target audience, which advertisers can then review before deciding whether to adopt them. This reduces the need for duplicate production, but it does not shift the responsibility for placement to models; budgeting, audience targeting, material rights, and compliance statements remain the responsibility of the operators.
The optional text adaptation feature will adjust the existing titles and descriptions based on the context of the conversation, and translate them into the language preferred by the user. This can make the content more relevant to the current issue, but it also reduces the retraceability of "everyone seeing the same advertisement." If the system dynamically generates different versions, the platform needs to retain the versions, the context that triggered it, and the placement records to ensure that brands and regulators can trace what a particular user saw at that time.
Beyond commercial value, Sponsored Agents must pass three tests of trust.
The first requirement is accuracy of facts. When sponsoring agents answer questions regarding inventory, prices, returns and exchanges, as well as service terms, they should prioritize referencing real-time data from the merchants and display the update times. Information that cannot be confirmed must be clearly stated, and it should not be filled in merely to maintain the flow of the conversation. This is especially true for financial products, health-related items, and high-value goods, where the consequences of incorrect answers can be far more severe than a simple mistake in ordinary promotional text.
The second point is regarding data boundaries. Users may disclose information such as addresses, budgets, family circumstances, or business needs during conversations. The platform needs to clarify whether this information is obtained by merchants, for how long it is retained, whether it can be used for subsequent targeted marketing, and how it can be deleted. Sponsoring agents cannot assume that they have access to the entire background of a user's conversation (represented by ChatGPT) just because the user is willing to ask questions. OpenAI states that two conversations are independent of each other, but in practice, products need to implement this separation through permissions and logging mechanisms.
The third point is the evaluation criteria. Advertisers are naturally concerned with clicks, conversation duration, and conversions, but the platform cannot consider "getting users to engage in more rounds of conversation" as a success in itself. More reasonable indicators include problem resolution rates, error rates, complaints, returns, whether users understand the sponsorship relationship, and the actual intention of users to visit the website after the conversation. If an agent is only good at prolonging interactions without helping with decision-making, then it is just a more expensive form of interactive advertising.
HubSpot access has been opened; businesses can connect to their ChatGPT Ads accounts to create ads, track performance, and follow up on leads within CRM. American Shopify merchants can also create and manage campaigns through this application. Other markets that already offer ChatGPT Ads will have access to this application starting from September 23rd. However, there are still regional and qualification restrictions in place, so it cannot be assumed that every merchant will be able to use it immediately once the application is launched.
From an industry perspective, OpenAI is transforming its advertising system from a monetization module into a product line: on one end, it controls the entry points for conversations; on the other end, it connects to CRM and e-commerce backends; in between, AI is used to generate content and analyze performance. This approach overlaps with search advertising, retail media, and customer service robots. The real competition is not about who has more traffic, but about who can maintain relevance, clear identification, and reliable data in their commercial responses.
Sponsored Agents is still in a restricted testing phase at present. It demonstrates that OpenAI is exploring interactions after ad clicks, but it does not prove that this format has improved conversion rates, nor does it show that users are willing to engage with brand representatives for an extended period. In the next phase, attention should be paid to expanding the scope of testing, reviewing rules for sponsored responses, controlling user data, and handling errors. Conversations can make ads more useful and may also make commercial influence more concealed; whether the platform can clearly distinguish between the two will determine whether this product can gain long-term trust.
