ChatGPT is selling ads after all: the unsettling part is how close they are getting to answers

The bottom line

ChatGPT finally started selling ads, but what’s really uncomfortable is not that the ads appear, but that the ads are getting closer to the “answer”.

OpenAI's advertising policy has been clear: ads can appear near qualified chats, but will not be placed in sensitive user situations and brand-unsafe situations, such as suicide and self-harm, mental health, child safety, politics, privacy, online abuse, etc. In the initial stage, advertising will mainly cover consumer products, local services, travel experiences, digital products and education.

Meanwhile, a study filed on August 5 analyzed 91 simulated accounts, 335 real queries, more than 3,000 ads and 186 advertisers. Ads began to appear 14 days after account creation, and lower-income accounts were more likely to see ads across simulated income tiers, the study said.

Don’t rewrite this sentence as “ChatGPT will discriminate against you based on your true income.” The study used simulated accounts with income signals and observed differences in ad display. It is not that OpenAI has confirmed that it has done an income portrait for each real user.

Advertisements and answers are still separate, but the boundaries are changing

OpenAI's policy requires that advertisements and model responses be clearly distinguished. Advertisers cannot make unfounded claims about functionality, price, performance, or comparative relationships, nor can they pretend to be a ChatGPT interface.

The study observed that current advertising is clearly separate from LLM responses and is primarily directed to a certain advertiser rather than a specific product. Researchers also expect that this separation may continue to change as advertising becomes more integrated into chat interfaces.

This is the difference between search advertising and AI advertising: the search page gives you a row of links, and the chat interface gives you an answer like a suggestion. When an advertisement is close to the answer, what the user really wants to ask is not just "is this an advertisement?" but also "does it affect the order of the answer?"

The most important thing to be wary of is not the advertiser, but the context

OpenAI policy prohibits advertising next to sensitive and vulnerable conversations, as well as categories such as politics, weapons, gambling, adult, and unsubstantiated health claims. These restrictions indicate that the platform itself knows that the chat context is more private than the ordinary web page location and is more easily regarded as a trust relationship by users.

Once the advertising business expands, the difficulty of review will also increase: advertising copy, landing page, user questions and model answers must be viewed together, and it is not enough to review one of them alone.

Our judgment: ChatGPT is changing from an answering tool to a distribution portal

Ads will not immediately replace searches, nor does it mean that every answer will be rewritten by an ad. But once a user asks "what to buy, where to go, which tool to choose" in ChatGPT, the platform moves from answering questions to decision distribution.

For those who work on independent websites, foreign trade, and products, this means that in the future, you can’t just do SEO, but also pay attention to whether AI includes your brand in comparison, recommendation, and purchase paths. For ordinary users, you need to develop a simple habit: when you see a recommendation, first confirm whether it is model judgment, advertising space, or the two are adjacent.

What to watch

• Whether ads will go from "near the answer" to inside the answer.
• Display differences in different regions, account ages and user portraits.
• Whether OpenAI exposes more ad attribution, ranking and user control options.
• Whether the research results can be replicated in a larger sample and real user environment.