OpenAI Added 3 ChatGPT Ad Markets Because Self-Serve Is the Real Launch
OpenAI began rolling out ChatGPT ads to Canada, Australia, and New Zealand in the weeks following a March 26 announcement, expanding beyond the US pilot that launched February 9. Ads appear only on Free and Go tiers for logged-in users over 18. Self-serve advertising tools launch in April 2026, dropping the agency-only $200,000 floor to roughly $50,000 for direct advertisers.
That is three markets added and a 75% cut to the entry ticket inside a five-week window. Search Engine Land confirmed the geographic rollout on April 17. Earlier, CNBC reported on March 26 that the US pilot had crossed $100 million in annualized revenue before any of this expansion happened.
The three markets are not random
Canada, Australia, and New Zealand share a cluster of traits that made them obvious first picks, and it is not size. Population-wise, the three together barely move the needle against India, Germany, or Brazil. What they share: English-language primacy, privacy regimes that are GDPR-adjacent but workable, and high ChatGPT penetration relative to Google Search share.
OpenAI picked the markets where the product is already overperforming the ad inventory that currently exists. That is the map to read. These are not the biggest markets. They are the ones where OpenAI can measure cleanly without fighting translation infrastructure or a regulatory wall. Germany and France are still blocked internally by privacy teams. Japan needs Japanese-language ad infrastructure. India does not have the paying user base to matter for the Free-tier economy.
For advertisers running campaigns in Canada, Australia, or NZ specifically, that is the signal. You now have a new ad channel most of your competitors do not know opened yet, sitting on top of an audience that has already shifted search behavior.
The CPM math was already compressed before any of this
New ad markets usually launch with a scarcity premium. Few advertisers, lots of inventory, CPMs sit high for the first month until demand catches up. Google's early YouTube rollouts in new countries looked like this. Meta's WhatsApp ads are doing it now in Brazil.
ChatGPT ads are not doing that. We reported earlier this month that US ChatGPT ad CPMs fell from $60 to $25 in nine weeks, a 58% drop. That was under the US-only pilot and the $200K agency floor. Now three more markets open, and the self-serve floor drops to roughly $50K, per The Keyword's reporting on the pilot extension.
What that means concretely: Canadian and Australian CPMs are probably launching already compressed. You do not get a one-month premium window to exploit. You get a low-floor channel that seems likely to stay low through most of 2026, because inventory expansion keeps outrunning advertiser onboarding.
That changes the arbitrage. Early movers used to chase premium pricing in a thin market. The actual value now sits in the attribution data you can build before 30x more advertisers enter the auction six months from here.
Self-serve is the bigger story than three new countries
If you read this week's Search Engine Land piece as "OpenAI expanded to three markets," you missed the actual move. The three markets are scaffolding for self-serve. OpenAI has been operating through agency partners only: Omnicom, WPP, Dentsu, DEPT, Jellyfish. The $200,000 minimum priced out every direct advertiser not running a full agency relationship.
Self-serve, launching in April, drops that floor to roughly $50,000. OpenAI's ads manager went into private beta right before the geographic expansion got announced. That is the sequencing. First, expand to markets where measurement works cleanly. Second, open the floodgates to direct advertisers.
Three more markets plus self-serve probably pushes the advertiser count on ChatGPT up 10-20x inside Q2. Inventory will absorb some of that. CPMs will absorb the rest.
From what I have seen with Meta's equivalent self-serve pushes, the direct-advertiser pool bids more aggressively than agency accounts because it does not have pacing teams enforcing discipline. Expect noisier auctions, worse average creative, and CPM spikes on high-intent keywords even as the overall floor stays low. On paper, that sounds like a problem. For a disciplined buyer sitting on clean attribution, it is mostly noise you can ignore.
The buy-side move to make before April ends
Concrete action for the next three weeks, if you run paid for clients or brands with Canadian, Australian, or New Zealand presence:
Get into the pilot now while it is agency-led. The $200K floor is still in effect until self-serve launches. If you can hit it with a two-month insertion order, the data you pull before self-serve crashes the auction becomes your attribution baseline. You cannot rebuild that baseline after thousands of new advertisers flood in.
Benchmark on query intent, not keyword bids. OpenAI's system blends user query with advertiser keywords in a way that is not fully transparent to participants. Keyword match rates are not meaningful the way they are in Google Ads. Track actual conversion per query-intent cluster, then back into keyword ROI from there.
Treat the $50K self-serve floor as a staging threshold, not a ceiling. If you are planning to spend anywhere close to $50K, go to $75K instead. The direct advertisers flooding in at exactly the minimum will have the worst creative and the weakest attribution setup. Sitting 50% above the floor gets you cleaner auction positioning for very little incremental spend.
Build for dismissal rate, not CTR. OpenAI keeps flagging "low ad dismissal rates" as a success metric. That is a user signal, not a click signal. Creative that reads like useful information in a chat thread outperforms creative that reads like a banner. Manscaped already pulled this off by matching AI's register instead of writing ad copy on top of it.
What OpenAI is not saying out loud
Two absences in the public announcement feel louder than the announcement itself.
First, OpenAI has not released conversion-tracking specifications for advertisers outside the US pilot. We covered recently that OpenAI is building ChatGPT conversion tracking only OpenAI can read. That system appears to still be US-only for now. Canadian, Australian, and NZ advertisers are flying with OpenAI-reported metrics and nothing else. That is a black box worse than PMax.
Second, there is no disclosure on what percentage of Free-tier users actually see ads. OpenAI's public stance on its approach to advertising says ads are restricted near health, mental-health, and political topics, and that under-18 users do not see them. Those are three large carve-outs. The real served-impression rate is probably well below the registered user count. From what I have seen in other ad platforms handling equivalent carve-outs, the range tends to be roughly 40-55% of active users, but OpenAI has not confirmed anything here.
Neither gap is disqualifying. Both are reasons to run any ChatGPT ads budget with a measurement partner that can triangulate against server-side tracking. If OpenAI is your only source of truth, you are in roughly the same position Meta advertisers occupied in 2014, trusting them until the incentive structure breaks.
Where the money gets made in the next eight weeks
The self-serve launch is the real inflection point. Three new markets is the setup move. If you can run the agency-floor pilot in Canada, Australia, or NZ before April ends, the attribution data you pull outlives the advertiser flood that comes right after.
Most teams will not make that call because $200K looks expensive for an experiment. The alternative, though, is being one of 30,000 $50K advertisers in June, all pattern-matching the same winning creative, all bidding against each other, and all reading the same OpenAI-reported metrics as truth. Personally, I would rather pay $200K for two months of clean signal than $50K for eight weeks of noise.
Notice Me Senpai Editorial
