CPMs that dropped from $60 to $25 in three months. A minimum spend that fell from $200,000 to $50,000. Cost-per-click bids running $3 to $5. If you've been in B2B marketing long enough, you recognize this pattern: a platform scrambling to fill inventory before the auction gets crowded.
INMA's reporting on OpenAI's ad acceleration tells the story of a company building its ad stack at breakneck speed. The pilot hit $100 million in annualized revenue in under six weeks. Self-serve access opened this summer. DSP deals with Criteo, Smartly, and StackAdapt are live. This isn't a beta test anymore; it's a land grab.
The question for your 2027 media plan isn't whether ChatGPT Ads will matter. It's whether the arbitrage window will still be open by the time you get budget approval.
The Distribution Problem That Creates Your Opportunity
Here's the tension that makes this interesting. Search Engine Land's analysis overlays browser market share data from 2008 to 2026 against monthly active user data for AI assistants, and the pattern is uncomfortable for OpenAI bulls. ChatGPT sits in the Internet Explorer seat: the incumbent that defined its category and now has to fight for the next one. Gemini takes the Chrome seat, not because it's the better product, but because it's plugged into an ecosystem that never has to ask permission to show up.
ChatGPT's audience flatlined in mid-2025. Worse, only 5% of its users don't overlap with Google's. Every single ChatGPT user has to be earned through downloads, habit formation, and retention loops. Gemini inherits Android, Search, Workspace, and YouTube distribution for free.
This is precisely why the ad inventory is cheap. OpenAI needs revenue diversification, and they need it fast. The Information's investor deck projects ChatGPT Plus subscribers dropping 80% in 2026, from 44 million to 9 million. The replacement strategy is ChatGPT Go, an ad-supported tier at $5 to $8 per month, expected to grow from 3 million to 112 million subscribers. The consumer business is going ad-supported whether the product wins the platform war or not.
For B2B marketers, this creates a window. The platform needs advertisers more than advertisers need the platform. That imbalance won't last.
What the Targeting Model Actually Looks Like
Forget keyword lists. Carnegie's breakdown of ChatGPT Ads for higher education explains the mechanic: advertisers provide contextual signals, or "context hints," to help the platform understand which conversations may be relevant. Ads appear when a user's conversation signals relevance, clearly labeled and visually separated from ChatGPT's organic response.
This is closer to Meta's interest-based targeting than Google's keyword auction, but with a critical difference. The user is in an extended dialogue, sometimes 10, 20, or 30 minutes at a stretch. AdVenture Media's timing analysis makes the point that matters: these are high-cognition, high-intent moments. Users aren't scrolling mindlessly. They're researching a purchase, drafting a comparison of service providers, or evaluating a software solution.
For B2B, that's the dream state. You're reaching someone actively working through a decision, not interrupting their feed.
OpenAI's stated principles commit to answer independence (ads don't influence ChatGPT's responses), conversation privacy (no selling data to advertisers), and user control (personalization can be turned off). Whether those principles hold under revenue pressure is a governance question your legal team should track, but the initial framework is cleaner than most platforms launched with.
The Math on Testing
Let's model a pilot. Assume a $50,000 minimum spend over a quarter. At current CPMs of $25, that's 2 million impressions. At a 0.5% click-through rate (conservative for a new format with novelty attention), you're looking at 10,000 clicks. At $3 to $5 CPC on the click-based model, the same budget gets you 10,000 to 16,000 clicks directly.
The question is conversion rate. If your landing page converts at 2% to demo requests, you're looking at 200 to 320 demos from a $50,000 test. If your demo-to-opportunity rate is 30% and your average contract value is $50,000, the math works out to $3 million to $4.8 million in pipeline from a $50,000 investment.
Those are illustrative numbers, not guarantees. The point is that the unit economics are testable at a budget level that doesn't require board approval. You can run a 90-day pilot, measure cost-per-demo against your Google and LinkedIn benchmarks, and make a data-driven call before committing 2027 budget.
Where the Money Comes From
INMA's analysis projects OpenAI's ad trajectory at $2.5 billion this year, $11 billion in 2027, $25 billion in 2028, and $100 billion by 2030. That last figure is roughly 8% of the global digital ad market, which analyst Brian Wieser called "extremely aggressive."
The natural assumption is that money comes from Google Search, but Google's Q1 numbers don't support that read yet. Search ad revenue was up 19% year over year. What's eroding is Google's network business, the off-Google publisher inventory, which fell 4%.
For B2B budgets, the more likely reallocation source is Meta and LinkedIn performance spend. The David Dugan hire from Meta, the DSP deals, and the conversion event types (leads, orders, registrations, subscriptions, trials) point to a direct response platform built for performance advertisers. If your LinkedIn CPCs have been climbing while conversion rates flatten, ChatGPT Ads offer a testable alternative.
The Pilot Framework
Run a 90-day test with these parameters:
- Allocate $50,000 to $75,000, enough to generate statistical significance on conversion metrics.
- Pick two to three high-intent use cases where your buyers are likely researching: software comparisons, implementation planning, or vendor evaluation.
- Build landing pages that match conversational intent, not your standard demand gen templates.
- Instrument everything: UTMs, conversion pixels, CRM source tracking.
- Define success criteria before launch: cost-per-demo, cost-per-opportunity, and pipeline generated versus your Google and LinkedIn benchmarks.
The risk is that the platform doesn't scale, the targeting stays crude, or Google's distribution advantage makes ChatGPT's audience share irrelevant by 2028. The risk of not testing is that your competitors lock in learnings and audience data while you're still waiting for the channel to "mature."
Arbitrage windows close. The 2027 planning cycle is the time to find out whether this one is real.
