ChatGPT ads are pricing at roughly $60 CPM: the same number Perplexity touched at launch, and the same number Netflix defended before its own ad business began sliding downward. Advertisers are pushing back, and analysts are debating whether chatbot advertising deserves its own sixth channel category, alongside search, social, retail media, the open web and streaming. That debate asks the wrong question. What chatbot advertising is missing is not a new category. It is an off-platform ad model, one that extends beyond a single chatbot's own inventory to the publisher pages the answer actually depends on.

  • ChatGPT's roughly $60 CPM mirrors Netflix's 2022 launch price, which fell to about $20 to $30 once buyers gained leverage.
  • Advertisers are pressing OpenAI for negative targeting controls, a sign the platform is still building discipline search and social already have.
  • Analysts led by Eric Seufert argue chatbot ads form a genuine sixth ad category, blending search-level intent with social-style targeting.
  • The entire pricing and category debate is confined to inventory ChatGPT itself owns, and says nothing about paying the publishers whose pages get cited.
  • The category the market actually needs is an off-platform ad model built on cited publisher pages, not a sixth slot inside one chatbot.

What an off-platform ad model actually means

An off-platform ad model places brand advertising on the pages an AI engine cites to build its answer, not inside the chatbot interface itself. ChatGPT's own ad unit, a sponsored chat card or a CPM-priced reach campaign, is on-platform inventory the platform controls end to end. Every dollar in this week's coverage, from the $60 pilot CPM to the CPC rollout, has stayed on that side of the line.

Netflix's $60 CPM and ChatGPT's $60 CPM

In the open-web economy, scarce new inventory tends to open high. Netflix launched its ad tier in 2022 at a $60 CPM, defended that price through 2023, then watched it slide to roughly $20 to $30, as reported by Digiday and Adweek, once buyers gained leverage and alternatives multiplied. ChatGPT's ad pilot opened in February 2026 at that same $60 figure, alongside a $200,000 minimum spend, and third-party desks such as StackAdapt were already discounting access to CPMs as low as $15 by April, months into the rollout.

The agentic-web version of this story runs faster, not different. A platform sells scarce attention while advertisers are curious, then price compresses once supply catches up. The open question is not whether ChatGPT's price will fall; history says it will. The real question is whether anyone builds the off-platform layer that pays publishers before the price does.

What AdExchanger reported, and what we verified

AdExchanger's August 28, 2026 roundup ties ChatGPT's CPM to Perplexity's early launch pricing and to Netflix's initial ad floor, and reports that advertisers are pushing OpenAI for negative contextual controls, citing Adweek's reporting on a small group of test advertisers. We traced the Netflix comparison independently: reporting from Digiday and Adweek spanning 2023 through 2026 confirms Netflix's ad CPM did fall from about $60 to roughly $20 to $30 as the business matured, supporting the parallel AdExchanger draws.

AdExchanger also notes ChatGPT held to CPM-only pricing until a cost-per-click option arrived in April 2026, matching separate reporting on that month's self-serve Ads Manager launch. That gap is the window in which OpenAI's ad product still looked more like a media buy than a search product.

The case for a genuine sixth ad category

Seufert's argument, in its strongest form

Eric Seufert of Mobile Dev Memo argues chatbot advertising sits in a privileged position: it carries search-level intent when a conversation turns commercial, and social-style targeting when it doesn't. On his count, that makes it a genuine sixth category, distinct from search, social feeds, retail media, the open web and mobile publisher inventory, and streaming or CTV. It's a credible argument. No existing channel buys on live conversational context the way ChatGPT's ad auction now does.

Why a new category still doesn't answer who gets paid

Even granting Seufert's category, categorization only settles how a platform's own inventory should be bought and sold. It says nothing about where the content came from. Search advertising eventually built an ecosystem that shared revenue with the sites its results depended on. Chatbot advertising has built a CPM auction, but not the equivalent payment rail for the publisher pages its answers depend on. A sixth category is an ad-buying convenience. It is not a revenue model for the supply side.

What this means for buyers and for publishers

For CMOs, media buyers and SEA managers: don't fund this from one line item

Treat ChatGPT ad spend as provisional, not fixed. Netflix's own CPM history says $60 rarely holds once buyers get better targeting tools, and the push for negative contextual controls is the first sign repricing has already started. Build measurement now, but hold back from committing a full year's budget to one platform's owned inventory before an off-platform, citation-based alternative exists.

For publishers: this week's CPM war isn't yours yet, and that's the problem

Every figure in this week's coverage, the $60 pilot CPM, the $200,000 minimum spend, the $15 discount pilots from third-party desks, stays inside OpenAI's own product. None of it reaches the sites supplying the facts inside the answer. Until an off-platform ad model exists, publishers remain unpaid infrastructure for a pricing war they can't see, let alone bill for.

Three things to watch as the CPM debate matures

Watch whether OpenAI's own CPM compresses the way Netflix's did as the self-serve, CPC option scales adoption beyond the original pilot advertisers. Watch whether more third-party desks follow StackAdapt in discounting access to chatbot inventory, which would accelerate commoditization on-platform. And watch for the first ad network or AI engine to pilot a citation-based revenue share for the publisher pages actually feeding the answers, since that is the moment an off-platform ad model stops being theoretical.

Conclusion

Hold on to this: chatbot advertising's pricing debate is not a categorization problem, it is a supply problem, and the open web already showed how that resolves when one platform controls all the inventory. Smalk AI is built for the off-platform layer this week's CPM story keeps skipping: Generative Engine Advertising that places native ads for AI agents on the publisher pages those agents actually cite, paying the sources that feed every answer. What to watch next: whether any platform starts pricing publisher-side citations before ChatGPT's own on-platform CPM finishes its Netflix-style slide.