ChatGPT Ad CPMs Fell 58% in Nine Weeks. The Arbitrage Window Is Closing.
OpenAI launched ChatGPT ads on February 9 with a $60 CPM and a $200,000 minimum commitment. Nine weeks later, PPC Land reports CPMs are trading at $25 to $35 per thousand impressions, with some placements as low as $15. The pricing floor is collapsing in preparation for a global rollout, which means brand teams have roughly one auction cycle to buy benchmark data before density pushes prices back up.
The $60 number was always a placeholder
OpenAI's initial ChatGPT ad pricing got framed as premium positioning. It wasn't. A $60 CPM with a $200K minimum does one thing: it gates out small advertisers so OpenAI can operate the pilot with a handful of holdcos. Call it a managed buy at scale, because that is what it was. Even the $60 base didn't hold consistently. The Jellyfish team told PPC Land that effective CPMs were averaging closer to $45 depending on inventory composition, and that was before the recent drop.
Now the effective range sits between $25 and $35, which is roughly where ad-supported streaming TV landed when it moved from first-party commitments to open auctions. LinkedIn's CPM of around $39, cited via Gupta Media's benchmarks, used to be the closest analog people pointed to. For the moment, ChatGPT is priced below LinkedIn, which is a weird sentence to write and probably won't last.
The other Gupta Media benchmarks: Facebook at $4.82, TikTok at $3.02, Instagram at $7.63, Google Display at $10.33. ChatGPT is still 3 to 5x more expensive than the open social platforms. But it's no longer priced like a private salon, and that's the signal that actually matters.
The global rollout is what's driving this
Ashley Fletcher, Adthena's CMO, summarized the thesis bluntly in the PPC Land piece: "It feels like everything is coming down, preparing for a wider auction accessibility." OpenAI doesn't telegraph global rollout dates the way Meta does. It runs on a quiet cadence and ships. If CPMs are dropping this fast this quickly, the auction is being primed for more advertisers, which means the invite-only phase is ending soon.
Current placements are still narrow. Confirmed participants include Target, Ford, Mrs. Meyer's, Adobe, Expedia, Qualcomm, Best Buy, Enterprise Mobility, and The Knot Worldwide. An OpenAI spokesperson told PPC Land that "a subset of advertisers are beginning to test an early version of an ads manager platform." Translation: the ads manager is under construction. When it ships, the auction gets real.
Separately, The Information reported on April 15 that OpenAI is building cost-per-click and cost-per-action pricing to sit alongside the current reach-CPM model. WinBuzzer covered the CPC plan and flagged the obvious point: CPC and CPA put ChatGPT directly on Google and Meta's performance turf, not just Netflix's brand turf.
The arbitrage is real but narrow
Here is what is actually on the table right now. Between $25 and $35, you can buy brand-safe impressions on a surface that OpenAI's own documentation says is available to roughly 85% of ChatGPT's user base, where fewer than 20% of users encounter ads in a given day. Low density, limited advertiser count, premium context, measurable CPMs. That is the textbook setup for an arbitrage window.
It is also narrow. Three reasons.
First, when OpenAI opens the auction globally, density rises and clearing prices rise with it. Brands that already have creative in-market and a baseline for what works will bid more aggressively, because they have data. Brands that wait will bid blind.
Second, the Criteo data is starting to circulate. Their February 2026 study of roughly 500 US retailers found that LLM-referred users converted at about 1.5x the rate of other referral channels. That is the kind of number that moves CFO conversations. When it shows up in the April and May earnings commentary, budget committees will start asking why the brand team hasn't tested.
Third, the attribution infrastructure is still broken. We covered last month how OpenAI is building ChatGPT conversion tracking that only OpenAI can read. Today, ChatGPT Ads Manager tracks impressions and clicks, nothing else. No pixel, no standard post-view attribution, no cross-platform view. If you wait for attribution to land, you are waiting for the exact moment every brand floods the auction at once.
What a test buy actually looks like today
If you manage a brand budget that can justify a $200K minimum against brand-building line items, the test is straightforward. Run a reach campaign targeting adults on ChatGPT Free and Go tiers in the US. Optimize for impression share, not clicks, because clicks are not the conversion vector this quarter anyway. Use creative designed for a conversational context: brand recall, entity association, category-level messaging.
Do not treat this as a direct response channel. The Ads Manager gives you impressions and clicks, and that is it. What you are actually buying is incremental brand search lift and a surface-area bet on citation share inside ChatGPT. Measure it with pre-post brand search lift in Google Search Console and, if you can fund the study, with category-level citation monitoring inside ChatGPT itself.
Adthena's February analysis found ads appearing in just 0.8% of roughly 500 ChatGPT prompts they sampled. That percentage will rise. Brands that test at 0.8% density are buying clean benchmarks. Brands that test at 10% density are measuring a different product.
The counterargument worth taking seriously
Robert Webster, founder of TAU, threw the one cold-water quote in the PPC Land piece: "Google and Meta own industry measurement, and that's the real moat, not the UI." He is not wrong. OpenAI can ship a beautiful Ads Manager and still lose the attribution war, because the attribution war is not about UI. It is about MMPs, GA4 integrations, and reporting marketers already trust.
From what I've seen, though, that argument holds only if OpenAI plays the Google game. OpenAI has a different option available. Control measurement inside the chat surface entirely. Closed-loop, chat-native, no pixel required. If the CPC and CPA rollouts go that way, OpenAI doesn't need to fight for MMP integration. It becomes its own MMP.
The signal under the noise
The CPM drop is not weakness. It is a tell. OpenAI crossed $100M in annualized ad revenue six weeks after launch, confirmed publicly on March 26, and now it is cutting prices by more than half. Those numbers do not reconcile if OpenAI is optimizing for near-term revenue. They reconcile if OpenAI is optimizing for auction depth before the global expansion, which is roughly the move Google made in 2002 and Meta made in 2013, and both of those went reasonably well for them.
If your brand budget is above $500K this year and you haven't run a ChatGPT reach test, the cheapest period to do that is probably the next 30 to 60 days. After that you are buying into the same auction as everyone else, paying to learn what works instead of already knowing.
Honestly, I'd bet against OpenAI hitting its $100B-by-2030 target. I wouldn't bet against them hitting scale faster than the attribution industry can keep up with. That gap, between OpenAI's ad surface maturing and the measurement ecosystem catching up, is the part that actually matters when you're deciding whether to test now or later.
Notice Me Senpai Editorial
