Logo
FrontierNews.ai

ChatGPT Ads Are Getting Cheaper, But Advertisers Still Can't See Inside the Black Box

OpenAI's advertising partners are making it cheaper and easier to buy ads in ChatGPT, but a fundamental problem remains: advertisers still can't see how their money is actually being spent. Criteo, which became OpenAI's first ad tech partner in March, recently dropped its minimum investment requirement from $50,000 to $10,000 and now matches every dollar clients spend on ChatGPT ads. These incentives arrive as ChatGPT surpasses 1 billion monthly active users, making it an increasingly valuable advertising channel.

The moves signal confidence in ChatGPT's growing appeal as an ad platform. Some advertisers have already started seeing positive results, and the reduced barrier to entry could unlock spending from smaller brands and agencies that previously couldn't afford to test the channel. Yet the underlying tension remains unresolved: even as adoption barriers fall, transparency barriers stay firmly in place.

Why Are Advertisers Frustrated With ChatGPT Ads?

Ben Kahan, head of programmatic at agency Brainlabs, describes the experience of buying ChatGPT ads as "putting money into a black box." Brainlabs is one of the agencies using Criteo's platform, and while Kahan acknowledges the offering is "compelling" because it lets brands feed existing product catalogs directly into OpenAI's system, the lack of control is troubling.

"I don't have control over the volume or the saturation until the end of the campaign," said Ben Kahan.

Ben Kahan, Head of Programmatic at Brainlabs

This constraint highlights a critical gap in how ChatGPT's ad network operates. Unlike traditional search or social advertising platforms, where advertisers can adjust bids, pause underperforming creatives, and monitor impressions in real time, ChatGPT advertising remains largely opaque. Advertisers don't know how often their ads appear, which users see them, or how saturation might affect performance until campaigns conclude.

How Are Advertisers Adapting to Limited Visibility?

Despite the transparency challenges, several factors are pushing brands to experiment with ChatGPT ads:

  • Lower Entry Costs: Criteo's $10,000 minimum investment, down from $50,000, allows mid-market brands and smaller agencies to test the channel without major budget commitments.
  • Dollar Matching Incentives: Criteo's matching program effectively doubles the value of initial ad spend, reducing the financial risk of early adoption.
  • Feed Integration Simplicity: Brands can funnel existing product feeds directly into OpenAI's system, reducing manual setup work compared to building campaigns from scratch.
  • Growing User Base: With ChatGPT now exceeding 1 billion monthly active users, the potential audience for ads has become too large for advertisers to ignore.

The appeal is practical: advertisers can leverage infrastructure they already have (product feeds) and test a new channel at lower risk. But the lack of mid-campaign visibility means brands are essentially making a leap of faith that their ads are reaching the right people and generating value.

Interestingly, this transparency gap is not unique to ChatGPT. Google recently introduced its first generative AI performance reports within Search Console, allowing advertisers to isolate how their content performs in AI-driven search results like AI Overviews and AI Mode. These reports track impressions, specific pages cited in generative responses, geographic breakdowns, device types, and performance over time. The contrast is striking: Google is moving toward greater transparency in AI-driven advertising, while ChatGPT's ad network remains largely a black box.

For advertisers, the question is whether lower costs and simpler setup can compensate for the lack of control and visibility. As more brands test ChatGPT ads and Criteo's incentives drive adoption, the pressure on OpenAI to offer better reporting and campaign controls will likely intensify. Until then, advertisers buying ChatGPT ads are betting on trust rather than data.