OpenAI now sells ad placement inside ChatGPT's answers in 40 more countries

A free-tier user in India asks ChatGPT which laptop to buy, and inside that answer sits a labeled ad someone paid to place ^[7][8][13]. The placement matters because the tiers carrying ads are the ones most of OpenAI's roughly one billion weekly users sit on ^[9].

Now the arithmetic. One billion annualized against a stated path past $40 billion in total annualized revenue is about 2.5 percent of the company's own target ^[1][6][16]. Measured against second-quarter sales of $6.7 billion, which annualize to $26.8 billion, ads are about 3.7 percent of the current pace ^[19][22]. A full year of ads at this run rate is roughly 8 percent of the $12.3 billion operating loss reported alongside that quarter ^[21][23]. Anthropic, which spent Super Bowl money mocking the decision to run ads, reported $11.6 billion for the same period with growth above 50 percent and a small profit ^[20][15]. So the year-end goal is $2.5 billion, two and a half times where ads sit now ^[5][17]. That gap is what applies pressure to the answer pane.

Take the disclosure promise at face value: OpenAI says ads are labeled and do not influence outputs, and that advertisers cannot see private conversations ^[13]. The company also says it will test new formats and measurement tools so brands can reach people in more "native ways" within ChatGPT ^[14]. Native is the word doing work there, because it describes inventory built to read like an answer.

The two proof points OpenAI published are worth reading closely. One e-commerce brand hit a three-fold return on spend across multiple campaigns in 28 days ^[11], and one technology partner said more than 80 percent of its ChatGPT traffic came from new customers ^[12]. Both are partner-reported single cases. New-customer share is a mix statistic, and it rises both when a surface reaches people who never heard of you and when it fails to bring back people who already bought. Twenty-eight days is shorter than most renewal cycles. Neither number says anything about retention or time-to-first-value for that cohort.

Here is the grid for a team that gets pipeline from assistant answers. One axis: is that traffic material to your acquisition or not. Other axis: can you buy placement in your market yet, given the expansion covers more than 40 new countries including India and markets across Europe, the Middle East and North Africa ^[7]. Material and buyable means you have a new budget line priced off a measurement stack the vendor says it is still building ^[14]. Material and not yet buyable is the exposed quadrant, where competitors in adjacent markets are learning the auction and you are not. Immaterial and buyable is a capped test with a stop date. Immaterial and unbuyable still deserves instrumentation, so the day it changes shows up in your dashboard and not in a quarterly review.

The measurement that settles any of this is unglamorous: assistant-referred sessions tagged apart from search, then 90-day retention and time-to-first-value compared between the two cohorts, plus a count of how often your brand gets named when nobody is paying. Teams that already have those three numbers can price a ChatGPT ad line. Teams working from click volume will be buying on the vendor's framing, and the vendor has $2.5 billion to find by December ^[5].