$100 million in six weeks. Now ChatGPT runs two ads per answer.

OpenAI just started showing two ads under a single ChatGPT answer. Search marketer Brodie Clark spotted the double placement around July 20, the first time the chatbot has stacked two competing advertisers in one response. Six months ago ChatGPT had no ads at all. The pace here is the story.

That pace matters to anyone whose traffic depends on being the answer. OpenAI Ads launched as a US beta on February 9, hit a reported $100 million in annualized revenue within six weeks, and now runs a Google-style auction with product feeds and a self-serve dashboard. The company is telling investors it’ll pull $2.5 billion this year and $100 billion by 2030. Plenty of analysts think that math is fantasy. Either way, the ad layer is being welded into the tool that roughly 900 million people open every week.

How the ads actually work

If you pay for ChatGPT, you won’t see any of this. Ads show up only for Free and Go tier users, and OpenAI says it won’t serve them to anyone it believes is under 18. Pro, Business, Enterprise, and Education stay clean.

The targeting is the part worth understanding. Instead of buying exact-match keywords the way you would on Google, advertisers write plain-language “context hints” that describe the conversations where their product should surface. Ask ChatGPT for “a good carry-on backpack under $100” and a luggage brand whose hint matches that scenario becomes eligible. Retailers can also upload product feeds of up to two million items over SFTP, and those products only appear in ads, never in the organic answer.

The pricing looks familiar to anyone who’s run a Google campaign. OpenAI started with a $60 default cap on cost per thousand impressions, added cost-per-click bids in the $3 to $5 range, and is testing cost-per-action pricing where the advertiser pays only when someone converts. The mechanics are Google’s playbook. The raw material isn’t. A keyword tells Google what someone typed into a box; a ChatGPT conversation tells an advertiser what that person is actually trying to get done, budget and constraints included.

OpenAI is emphatic on one point: the ad sits in a labeled sponsored unit below the response, and “ads do not influence the answer ChatGPT gives”. Whether a distracted user actually reads the two as separate is a different question, and one OpenAI can’t answer for them. The company also walls off sensitive categories: no ads run against dating, personal health, mental health, financial services, or political queries, which are exactly the conversations where a sponsored nudge would be most toxic.

Two slots, one auction

The two-ad expansion runs on a relevance-weighted, second-price auction, the same mechanism Google and Meta built their empires on: the winner pays one cent above the next-highest bid. OpenAI told advertisers about the multi-advertiser format on June 9, and Clark’s July sighting is the first sign it’s live for real users. During the test, a single unit below a response can carry items from one advertiser or several.

The rest of the machine filled in fast. The entry minimum was $200,000 when the beta opened in February, dropped to $50,000 in April when OpenAI added cost-per-click pricing, and hit zero in May when the self-serve Ads Manager opened to any US advertiser. More than 600 brands were on the platform before self-serve even launched. The pilot has since pushed past the US and Canada into the UK, Japan, South Korea, Brazil, and Mexico.

Doubling the slots is the quietest big move in that timeline. One ad per answer is a placement. Two ads competing in a live auction is an ad market, because now bidders are pushing each other’s prices up instead of buying a fixed spot. It’s the same lever Google pulled when it went from a single sponsored result to a stacked block of them at the top of the page. More inventory, more competition, higher clearing prices. For OpenAI, the auction does the revenue work automatically as demand grows.

The money and the skeptics

Here’s the number OpenAI wants you to see: about $100 million in annualized ad revenue within six weeks of launch. What makes that figure loud is the base it came off. Fewer than 20% of eligible users were seeing ads on a given day, and roughly 85% of Free and Go users are eligible in the first place. OpenAI’s pitch to investors is that today’s revenue is a sliver of a barely-switched-on system. From there the company projects roughly $2.5 billion in 2026 and $100 billion by 2030.

Now here’s the number the skeptics point at. eMarketer, in an analysis flagged by Adweek, puts the entire US standalone-chatbot ad market, meaning ChatGPT plus Copilot plus Google’s AI Mode plus Alexa, at under $1 billion in 2026 and just $5.41 billion by 2030. If that ceiling holds, OpenAI’s own 2030 target misses by about 90%. The firm’s read is blunt: the forecast “assumes OpenAI captures search ad budgets en masse from traditional search ad sellers, dominates a fully mature chatbot ad market, and outperforms every ad format in history, all at once.”

The gap is a scope fight as much as a growth fight. eMarketer’s ceiling is US-only and covers standalone chatbots; OpenAI’s $100 billion isn’t framed that tightly, so part of the distance is apples against a slightly bigger orchard. But even generously, three things all have to break OpenAI’s way at once for the target to land: advertisers have to move real search budget out of Google, the chatbot ad format has to mature into something buyers trust, and ChatGPT has to hold its user lead against Gemini and everyone else. Miss any one and the number collapses toward eMarketer’s.

There’s a real signal under the hype, though. Measure Protocol, sampling almost 143,000 conversations, found that more than one in five ChatGPT chats show commercial intent. People don’t type keywords at a chatbot. They describe a budget, a deadline, a use case. For an advertiser that’s a richer buying signal than most display inventory, which is exactly why the money is chasing it even if the $100 billion is a stretch.

What this means for you

If your traffic depends on being the answer, this is the shift to watch. For twenty years the game was ranking in Google’s blue links. ChatGPT is now doing to AI answers what Google did to search results: dropping paid placements next to the organic response, in a surface where there’s often one answer instead of ten links to choose from.

That surface is unforgiving in a way the old search page never was. On Google you could lose the top ad slot and still catch clicks from the organic listings below it. In a chatbot, the model hands back one recommendation, and if a paid unit sits above it, the free citation you worked to earn is now the second thing a user sees, if they scroll at all. Optimizing to be mentioned by ChatGPT was already the new SEO. Now there’s a paid lane running right alongside it, and the two are going to compete for the same eyeballs.

Three things to do now. Get your product data into a clean feed, because feed quality decides whether you’re even eligible when a relevant question shows up. Watch whether the sponsored unit starts eating the attention that used to go to cited sources, the way ads slowly pushed organic results down the Google page. And don’t build a budget around OpenAI’s $100 billion; build it around eMarketer’s sub-$6 billion and treat anything above that as upside.

My honest read: the ad business is real, the growth is real, and the 2030 number is marketing. Test it while inventory is cheap and the auction is thin. That window closes the moment two slots become four and everyone piles in.