OpenAI 在 ChatGPT 推出全新视觉广告格式并扩展广告测量工具

Key Highlights

OpenAI has cracked the advertising door open a notch. The newly launched visual ad format begins US-only testing this month, and only within image-generation contexts — when you ask ChatGPT to draw, you may see clearly labeled promotional content. It is the company's most explicit step yet from "we sell subscriptions" toward "we also sell attention", and the careful framing shows it knows the line it is walking.

The move is overdue in business terms. OpenAI carries enormous inference costs and a valuation that demands multiple revenue streams; subscriptions and the enterprise tier alone are a thin base for that kind of burn. Advertising is the oldest scaling monetization on the web, and ChatGPT's front-page traffic makes the inventory tempting. The only question was always how to do it without poisoning the product people came for.

What Happened

Per the official description, the ads come with boundaries: first, they must be clearly labeled and will not masquerade as ordinary results; second, they do not alter ChatGPT's answer to your question — the promotion sits inside the image-generation experience rather than inside your conversational reply; third, the initial scope is narrow, limited to US image-generation scenarios, a cautious probe rather than a full rollout.

The framing is deliberately modest, which signals OpenAI knows how easily ads can sour a trusted product. By starting inside image generation — a creative, lower-stakes surface — it avoids the most damaging version of the experiment: a sponsored answer to a medical, legal or financial question. That restraint is probably the difference between a tolerable test and a backlash.

Alongside this, OpenAI expanded its ad-measurement tooling so advertisers can see campaign performance more clearly. That is the standard move for any monetization platform: measurable delivery comes before sustainable revenue, and you cannot sell a premium inventory you cannot report on. Expect the measurement layer to mature faster than the creative one, because metrics are what unlock bigger budgets.

Technical Detail

Anchoring ads to image generation rather than conversation replies is a deliberate product choice. Replies carry trust; mixing in promotions risks the "am I being sold?" backlash that has punished other assistants. Image generation is a creative context where brand visuals feel less intrusive and are easier to render as native formats — a sponsored style suggestion is far less jarring than a sponsored answer to your tax question.

There is also a natural fit between the medium and the message. When a user is already generating images, a tasteful brand placement or a relevant product visual reads as part of the creative act rather than an interruption. The hardest problem in ad tech is contextual relevance without intrusion; image generation may be one of the few surfaces where those two stop conflicting.

Versus Competitors

Versus Google embedding ads in search results or Meta inserting promotions in feeds, ChatGPT's move looks restrained: visuals before text, labeling before precision targeting. It is clearly trying to avoid the early social-platform trap of ads ruining the experience, a lesson written in the churn of a decade of feeds. The bet is that a slower, cleaner entry earns longer-term tolerance.

The contrast is instructive. Search and social built ad businesses by gradually pushing the boundary until users accepted it; the risk for OpenAI is that AI assistants are held to a higher trust bar than a search results page ever was. Users granted ChatGPT a kind of advisory role, and polluting that role with commerce is more dangerous than it would be for a neutral utility.

Industry Impact and Use Cases

For OpenAI: a third growth curve after subscriptions and the enterprise tier, but paced carefully so the core product stays credible. For advertisers: a new high-intent, strongly visual inventory suited to brand and creative campaigns, especially for design, fashion and consumer products. For users: largely invisible in the short term; the key is whether the platform holds the line on clear labels and no effect on answers once the revenue pressure to expand inevitably arrives.

The test worth watching is what happens at the next inflection point — when the narrow US image-generation trial "succeeds" and someone proposes extending it to text answers or other regions. Every ad product in history has followed that exact path. OpenAI's stated restraint will be measured not by this launch but by the one it refuses to make.