OpenAI’s ads signal desperation

OpenAI’s decision to introduce advertising inside ChatGPT isn’t a bold new growth chapter.

It’s a pressure move.

The structure of the rollout — CPM pricing, no self-service console, vague targeting, small trial commitments — looks less like the launch of a serious ad platform and more like a hurried attempt to unlock revenue where other levers aren’t scaling fast enough.

Have a look:

Here’s what the move actually signals.

• CPM over CPC or CPA puts all the risk on advertisers
  In modern digital advertising, CPM-only buying is a relic. Search runs on CPC and CPA because intent is explicit and outcomes are measurable. Performance social evolved the same way. CPM survives mostly in brand advertising where attribution is secondary.
  By leading with CPM, OpenAI avoids accountability for results. Advertisers pay for “views” without clear intent, clicks, or conversions. That’s a hard sell in 2026 — especially for performance budgets.
• Sub-$1M trial commitments reveal low confidence
  OpenAI is reportedly asking early advertisers for less than $1 million each over several weeks. For serious brand advertisers, that’s a rounding error. If OpenAI believed ads were a durable revenue engine, these commitments would be larger, longer, and tied to clearer performance guarantees.
  Small tests suggest OpenAI itself isn’t confident enough to anchor ads as a core business line.
• No self-service console means this isn’t a real ad platform
  Every modern ad platform — Google, Meta, LinkedIn, Amazon — launched with self-serve at its core. Campaign creation, budgets, pacing, targeting, reporting, iteration. That is the platform.
  OpenAI is starting with manual sales. That’s not strategy. That’s triage. It buys time, avoids transparency, and delays hard performance questions.
• Targeting is structurally incompatible with ChatGPT’s promise
  Advertising depends on audience definition. ChatGPT’s value proposition is privacy, confidentiality, and trust. Those two goals collide.
  OpenAI cannot share user profiles, enable retargeting, allow lookalike audiences, or expose prompt histories without breaking trust. That leaves only weak substitutes:
  • Prompt-based contextual guesses
  • Model-inferred “intent” buckets
  • Coarse plan or geography filters
    None of these resemble the targeting advertisers are used to — or willing to scale.
• LLM intent is ambiguous, not transactional
  A prompt like “How does amortization work?” doesn’t mean someone is shopping, evaluating vendors, or open to persuasion. Search ads work because intent is explicit. LLM usage is exploratory.
  CPM ads dropped into that ambiguity feel more like interruption than utility — and advertisers hate paying into uncertainty.
• Attribution is weak by design
  Without clicks, conversions, or persistent identity, attribution collapses into “trust us, it worked.” That’s pre-programmatic logic from the mid-2000s, not how modern budgets are allocated.
• CPM forces disclosure OpenAI has avoided
  Selling impressions invites uncomfortable questions:
  • How long is attention actually held?
  • How often are ads noticed?
  • How frequently do users return?
  • How repetitive is exposure?
    CPM makes thin engagement visible — something OpenAI has largely sidestepped by leaning on broad usage claims.
• This looks like a research experiment, not a business system
  OpenAI is run by research scientists. They experiment. They demo. They ship features. Advertising inside ChatGPT appears to be treated the same way — as a feature to “show,” not an economic engine to operate at scale.
  The absence of industry-standard pricing, tooling, and infrastructure suggests ads are being tested because they can be, not because they fit.
• Pressure, not panic — but still a tell
  This isn’t chaos. It’s pressure from GPU burn, flat subscription ARPU, and investor expectations. Advertising is being pulled forward because other revenue paths aren’t compounding fast enough.
  Strong platforms delay monetization until fit is obvious. This is monetization before fit.
• The deeper mismatch
  Ads work when behavior is predictable and attention units are standardized. ChatGPT sessions are variable, cognitively demanding, and utility-driven. LLMs are tools, not media.
  Monetizing surface area instead of value is a familiar early-internet mistake — and one that rarely holds.

Bottom line

This ad rollout doesn’t signal confidence. It signals uncertainty. CPM avoids accountability. Manual sales avoid transparency. The lack of targeting exposes a fundamental mismatch between advertising economics and conversational AI.

Advertising may buy OpenAI time. It won’t solve LLM economics.

Metered, value-based usage will.

– Published Saturday, January 24, 2026