Generated by Codex with GPT 5.6 Sol High

Techmeme surfaced OpenAI’s October 5 announcement, “Building advertising for the way people use AI.” The immediate change is a visual ad format that OpenAI plans to test in the United States later this month while people wait for ChatGPT to generate images. The larger story is that ChatGPT advertising is becoming a full commercial system: creative formats, conversion tracking, bidding signals, brand-safety controls, and outside measurement partners are being assembled around conversations in which users are already deciding what to do or buy.

That is a more consequential shift than adding pictures to sponsored messages. Search advertising typically starts with a query and sends the user elsewhere. ChatGPT can remain involved while a person defines a need, compares options, creates an image, and refines a decision. OpenAI is trying to turn that unusually rich context into an advertising advantage without making the assistant’s answers feel purchased.

The new ad occupies a moment of attention

OpenAI says the visual format can show product inspiration, use cases, or experiences connected to a product or service. It will initially appear during image generation, clearly labeled and separate from the image being created. That placement is commercially clever: image generation creates a natural waiting period, and the user’s prompt may reveal the aesthetic, occasion, or object they have in mind.

It also makes the boundary between assistance and persuasion harder to judge. OpenAI says advertising does not affect ChatGPT’s answers, and the sponsored unit is visually separate. But independence is not only a question of whether an advertiser can edit model output. The ad-selection system still uses the context and intent of the current conversation; with personalization enabled, OpenAI’s help documentation says it may also use selected signals from past chats, memory, and ad interactions. Advertisers do not receive those conversations, but the platform can use them internally to decide which advertiser appears.

That distinction is important. “Conversations stay private from advertisers” does not mean conversational context is irrelevant to advertising. It means OpenAI acts as the intermediary: it keeps the raw context, performs the matching, and reports aggregated results. This resembles the role large search and social platforms already play, except the input can be more detailed and personal than a keyword or feed interaction.

Measurement turns the format into a platform

The announcement devotes as much attention to measurement as to the visual unit. Integrations with Hightouch, Tealium, and LiveRamp let businesses send conversion data from their existing systems. A longer list of attribution vendors can support reporting, click attribution, and OpenAI’s Conversions API. OpenAI is also exploring geographically controlled experiments intended to distinguish purchases caused by ads from purchases that would have happened anyway.

Those plumbing details show that ChatGPT Ads is moving beyond an experiment in placing sponsored links. Advertisers need to optimize campaigns against outcomes, compare the channel with paid search and social media, and connect an impression inside ChatGPT to activity on their own sites or apps. OpenAI is building the feedback loop required to compete for established marketing budgets.

The company cites favorable early partner results: DV Rockerbox reported that WeightWatchers’ attributed acquisition cost was 15.3% below its blended paid-search benchmark; WorkMagic reported a statistically significant purchase lift for Dose; and Triple Whale said 93% of Portland Leather’s ChatGPT-referred visitors were new. These are partner-reported examples selected for a launch announcement, not a neutral benchmark across advertisers. The methodology, campaign sizes, confidence intervals, and durability of the effects are not provided. They demonstrate what OpenAI wants marketers to measure, not yet how the channel performs in general.

Brand safety and user trust are different tests

Conversational advertising creates an unusual placement problem because there is no stable page surrounding the ad. The relevant environment is a live exchange that can move from ordinary shopping to health, grief, politics, self-harm, or another sensitive subject in a few turns. OpenAI says automated review, human oversight, and placement guardrails are intended to keep ads out of emotionally vulnerable or otherwise unsuitable contexts. Qualifying advertisers can also supply negative phrases, and DoubleVerify and Integral Ad Science are developing controlled evaluations of OpenAI’s brand-suitability safeguards without receiving private conversations.

These measures primarily answer an advertiser’s question: “Will my brand appear beside something damaging?” The user’s question is broader: “Will commercial incentives change what this assistant helps me notice, consider, or choose?” Clear labels and answer independence are necessary, but they do not fully resolve that concern. Even if the answer itself is untouched, an ad placed at the moment of deliberation can redirect attention and shape the set of options a user explores.

OpenAI’s scale makes the tradeoff material. The company says ChatGPT reaches 1.2 billion people each week and argues that advertising supports access beyond what subscriptions alone can fund. That is a plausible economic rationale, but the announcement does not disclose ad load, expected revenue, error rates for sensitive-context detection, or how often commercial relevance conflicts with user relevance. It also describes the external brand-safety work as pilots, not completed public audits.

The takeaway

The visual format matters because it reveals what ChatGPT Ads is becoming. OpenAI is not merely putting banners next to a chatbot. It is building an advertising market around intent expressed through conversation, then connecting that market to the same conversion, attribution, and brand-safety infrastructure used by mature digital platforms.

The durable question is whether OpenAI can preserve a legible boundary between the assistant and the marketplace around it. Keeping ads labeled, keeping answers independent, and withholding raw conversations from advertisers are meaningful safeguards. Yet the more precisely the ad system understands a user’s goals, history, and moment of decision, the more valuable—and potentially influential—it becomes. OpenAI’s commercial opportunity and its trust problem are two sides of the same context.