In 2002, Google did something that sounded insane to anyone who had ever bought an ad. It let you only pay when someone actually clicked. No upfront spend, no media buyer, no insertion order signed in triplicate. You typed in a credit card, wrote one line of text, and you paid a few cents only when a stranger who was already searching for the exact thing you sold raised their hand. The people who understood that early built whole companies on traffic that cost almost nothing, because the auction was empty and nobody had bid the prices up yet.
That window stayed open for maybe three or four years. Then everyone noticed, the clicks got expensive, and “cheap Google traffic” became a story founders tell the way grandparents talk about penny candy.
Last week, a window like that opened again. Just not where you were looking.
What actually happened
On June 5, OpenAI turned on cost-per-action ads inside ChatGPT. In plain English: you can now run an ad in ChatGPT and only pay when a person does the thing you actually care about. Not when they see it. Not even when they click. When they sign up, or buy, or complete whatever you’ve defined as a conversion. The whole thing runs on OpenAI’s own tracking pixel, which ties an ad someone saw inside a conversation to an action they took later on your site.
To see why this is a bigger deal than it sounds, you have to back up four months. ChatGPT only started showing ads at all on February 9 this year, back when ChatGPT first switched ads on and we walked founders through what the ad machine was going to mean for them. The ads sit below the AI’s answer, they’re labeled as sponsored, and they don’t change what the model tells you. That launch hit roughly $100 million in annualized ad revenue inside about six weeks. Six weeks. On a product that, per Reuters reporting this month, is now serving more than 900 million people a week.
And here’s the part that tells you it’s early. Those ads are still only showing to under 20% of eligible users, only in the US, only on the free and $8-a-month tiers. The CPA option is a limited pilot. The on-ramp to get in was almost comically low: any advertiser who had conversion tracking set up by June 1 got early access. That’s it. That was the velvet rope.
Why this should make a founder sit up
Think about what you’re actually buying here. Not attention. Intent.
Paid social sells you impressions. You pay to interrupt people who weren’t asking for anything, in the hope that a fraction of them happen to want what you’ve got. Old-school search ads were better, you at least caught people mid-search, but you still paid per click whether or not the click ever turned into a dollar. Pay-per-sale flips the risk. You don’t pay for the maybe. You pay for the yes.
Now layer in where this is happening. The person seeing your ad isn’t scrolling a feed killing time. They typed “what’s the best tool for tracking freelance invoices” or “compare Notion and Coda for a small team” into the assistant they now trust more than Google. They are standing at the cash register with their wallet half out, asking a machine for a recommendation. For a pre-seed founder with three hundred dollars and no growth team, getting in front of that exact moment, and only paying when it converts, is the friendliest acquisition math the internet has handed out in a long time.
That matters more now than it would have five years ago, because building stopped being the hard part the moment your next competitor became one person with a laptop and a weekend. When everyone can ship the product, the entire game moves to distribution. Getting seen by the right person at the right second is the whole job. OpenAI just put that second up for auction, and the auction is mostly empty.
The take, and the catch
Most of the noise about this is the wrong noise. “Ads are ruining ChatGPT, the assistant is getting polluted, this is the beginning of the end.” Fine. Maybe. But that’s the consumer’s complaint, not the founder’s opportunity.
Here’s the founder’s version. Intent moved. For twenty years the single most valuable piece of real estate in marketing was the Google search box, because that’s where a human went the instant they wanted something. That box is draining. People ask the assistant now. And the same shift that’s quietly strangling your SEO and your referral traffic, the zero-click thing everyone’s panicking about, is the exact thing that makes a ChatGPT ad worth buying. The assistant ate the search box. OpenAI just bolted a cash register onto the side of it. You can be annoyed about that, or you can be early. Those are the two options.
But I’d be lying if I sold you the upside without the catch, and there are two.
First, the window closes. It always closes. AdWords was a money printer in 2002 and a knife fight by 2006. Right now ChatGPT’s ad auction is cheap, the targeting is crude, and barely anyone is bidding. In eighteen months that will not be true. So treat this like the 2002 arbitrage it is, a temporary edge you exploit while it lasts, not a channel you build your whole company on top of.
Second, and this one’s uncomfortable, you’re renting paid distribution from the same company that’s destroying your free distribution. The thing erasing your organic traffic is now selling you the replacement. That’s not a partnership. That’s a toll booth on a road somebody built by ripping up the old road you used to drive for free. We’ve seen this movie in other corners of AI, where the platform that owns the bottleneck quietly sets the terms. Go in clear-eyed about who owns the scoreboard. The “conversions” you’re paying for are counted by OpenAI’s pixel, in OpenAI’s black box. You’re trusting the house to tell you how many times you won.
What to actually do about it
If you sell something with a clean online conversion, a sign-up, a purchase, a booked demo, set up a conversion event and get into the pilot now, while it’s cheap and empty. The entry requirement was literally “have conversions configured.” Do that this week.
Then don’t be a hero with the budget. Run it as an experiment with money you can afford to set on fire. Five hundred, a thousand dollars. Measure your real cost to acquire a paying customer against your other channels, not clicks, not impressions, not “engagement.” If a customer from ChatGPT costs less than one from Meta, scale it. If not, you learned something for the price of a nice dinner.
But honestly, the more important move has nothing to do with ads. Before you pay to show up, make sure the assistant can see you at all. Right now most brands are completely invisible to ChatGPT, and one company raised $14 million just to fix that problem for everyone else. If the model doesn’t know your product exists when someone asks it for a recommendation, no ad budget saves you. Get into the answer organically first. Buy the ad second.
And the biggest move is the one that isn’t a tactic. Rebuild your mental model of where buyers actually start. The blog post you ranked number one for is now a three-sentence AI summary nobody clicks. The buyer’s first question doesn’t go into a search bar anymore. It goes into a conversation. Plan your whole acquisition strategy around that fact instead of pretending it isn’t happening.
Closing
The founders who win the next few years won’t be the ones with the smartest AI product. Everybody has a good-enough AI product now. The model is nearly free and the code mostly writes itself. The winners will be the ones who figured out where the buyers quietly went, and got there first, while it was still cheap to show up.
In 2002, that place was a search box. In 2026, it’s a conversation. The cash register is already on. The only real question is whether you’ll be early, or whether you’ll be the founder in 2030 telling some kid about the time ChatGPT ads were dirt cheap and you didn’t bother.
