Advertisers Will Judge ChatGPT Ads Half Blind

Picture the moment. Someone asks ChatGPT to generate a mood board for a living room refresh, and an ad appears beside the finished image. The advertiser will be able to see what that ad cost and what it drove. What neither the advertiser nor its verification vendor will see is the conversation it sat next to. 

OpenAI announced on October 5 that it will begin testing this format later in the month, during image generation, with an initial group of US advertisers. Ads will be labeled and kept separate from the image itself, and OpenAI says advertising does not influence the answers ChatGPT gives. The announcement spends far more of its length on measurement than on the format, and that proportion is the story. 

To understand how much weight the early results can carry, start with 15.3%. That is how much lower WeightWatchers’ attributed cost per acquisition ran on ChatGPT ads than its blended paid search benchmark, according to DV Rockerbox as cited by OpenAI. It is one advertiser, reported by the partner doing the attributing, over a period the post never defines. The two other results OpenAI shared follow the same shape, one brand each, one measurement partner each. 

The detail that reframes the announcement sits lower down, in the brand suitability section. OpenAI is developing pilots with DoubleVerify and Integral Ad Science so independent partners can evaluate how its safety standards operate in a controlled test setting, with no access to private user conversations. Privacy is a legitimate reason for that limit. It still leaves a gap. DoubleVerify’s CEO said in the same post that context in this channel is defined by the conversation itself rather than a page or video. The conversation is exactly what the verifiers cannot read, and a controlled test setting, however rigorous, is a different exercise from inspecting the live conversations ads actually run beside. 

Read as a sequence, the announcement front-loads whatever can be counted. OpenAI names ten attribution partners, three full-funnel measurement partners and data integrations with three platforms, and it describes its own incrementality work as still in its early stages. The outcome side of the ledger is crowded. The context side is two pilots. 

This is not a one-off format. OpenAI says ChatGPT reaches 1.2 billion people each week, frames advertising as a way to serve a larger market than subscriptions alone, and describes the visual format as part of a broader ambition to help people discover, evaluate and decide on products inside ChatGPT. It also says more formats are coming. The verification question advertisers face now will travel with every format that follows, because the privacy that protects users is the same thing that limits the auditor. 

What does this mean for a B2B marketer with no plans to buy a ChatGPT ad? The same trade is coming to every AI-native surface your buyers use, strong evidence about what happened after the click and thin evidence about what surrounded the impression. If your demand programs lean on AI discovery, the context question belongs in every vendor conversation, not just this one. 

There is a second signal in the one control advertisers hold directly. OpenAI says Negative Phrases are now available to qualifying advertisers for narrow placement requirements tied to a brand’s own policies. That is a phrase-level lever, gated by a qualification the post does not define, applied to conversations the advertiser cannot read. 

One caution before reading this as a warning label. OpenAI says its placement guardrails assess whether a conversation is appropriate for advertising and keep ads out of emotionally vulnerable or sensitive contexts, and the Integral Ad Science CEO said its evaluation will cover a broad and evolving range of scenarios. The pilots could mature, and three early results could hold across more advertisers. Today they are pilots and early results. 

The practical move is to put the unknowns in writing before the budget. Ask what placement evidence you will receive, whether suitability findings come from live placements or test environments, and how incrementality will be designed beyond single-advertiser benchmarks. Treat initial spend as a structured test with a holdout group, not a channel commitment. 

OpenAI will show advertisers precisely what its ads did. It has not yet shown them what the ads sat beside.