ChatGPT ads hit $100 million in annualized revenue in under seven weeks. For context, it took Google search ads about five years to reach that number.

And that $100M came from less than 20% of eligible US users seeing ads on any given day. The full eligible audience is around 80 million people. Do the math on full deployment and you’re north of $500M annualized before international expansion even starts.

This could quickly become a channel and I’ve already had a few brands ask me whether they should be buying ChatGPT ads.

On April 14, Pacvue (the company that manages about 12% of global retail media spend across Amazon, Walmart, and Target) is presenting at the IAB Connected Commerce Summit. Their CEO has said publicly they want to be the first company to connect to OpenAI’s ad infrastructure. If that’s what April 14 is about, it means brands running retail media campaigns on Amazon could soon manage ChatGPT ads from the same dashboard.

Criteo signed on as OpenAI’s first official ad tech partner in early March. They’re connecting 17,000 advertisers and $4 billion in annual media spend to ChatGPT inventory. Smartly is building conversational ad units, little mini-chatbots that appear after the AI’s answer and can actually respond to you. They ran a similar format for Boots on Meta and saw 5x better sales than standard ads.

The ad tech stack for AI chat is being assembled right now, in real time. Construction, not a roadmap.

Here’s what the current experience looks like: you ask ChatGPT for a product recommendation. It gives you an answer. Below that answer, a sponsored product carousel appears — images, descriptions, prices, labeled as sponsored. Click through and you land on the merchant’s site for checkout. OpenAI tried doing checkout inside ChatGPT directly. Walmart measured it. Conversion was 3x lower than sending people to Walmart.com. They killed it in late March.

So the model now is the same as Google’s: ChatGPT owns discovery. The merchant owns the transaction. Standard retail media playbook.

Except there’s a problem that makes this different from every other retail media channel brands have dealt with.

The click-through rate on ChatGPT ads is 0.91%. Google search ads get 6.4%. Agency execs at DEPT have gone on record saying they “have not noted measurable business outcomes” from ChatGPT ads. The attribution infrastructure doesn’t exist yet. You’re spending $60 CPMs — roughly 3x what you’d pay on Meta — and you can’t close the loop on whether it drove a sale.

There’s one counter-data point worth knowing. Criteo reports that users who arrive at a product page via an LLM recommendation convert at 1.5x the rate of other referral channels. That’s across all LLMs, not just ChatGPT. But it suggests something real: when someone has a conversation with an AI about what they need and then clicks through to buy, the intent is higher than a search click or a social scroll.

The intent signal is richer. The measurement is broken.

Perplexity, 780 million monthly queries and growing fast, pulled all advertising in February. Gone. Their reasoning was blunt: “The challenge with ads is that a user would just start doubting everything.” They’re betting that the entire value of an AI recommendation depends on it being unsponsored. The moment you put ads next to it, you’ve contaminated the trust that makes it work.

ChatGPT is betting the opposite. That users will accept ads in AI chat the way they accepted them in Google search — annoying at first, then invisible, then just how things work.

We don’t know who’s right. But the stakes are real because AI recommendations carry more persuasive weight than a list of search results. When Google shows you ten blue links, you know you’re choosing. When ChatGPT gives you a recommendation in a conversation, it feels like advice. Mixing sponsored content into that is a fundamentally different proposition than mixing it into a search results page.

And the ads are coming fast. OpenAI is dropping the $200K minimum spend this month with self-serve access. That opens the door to every mid-market brand with a retail media budget and a fear of missing the next channel.

Meanwhile, look at what Amazon and Google are doing. Rufus ads are still free — Amazon’s collecting data before it starts charging. One seller tracked 88 clicks from Rufus ads versus 500,000 from traditional Amazon ads in the same period. The volume isn’t there. Google is further along: AI Mode has 75 million daily users, shopping ads appear alongside 25% of AI Overview results, and brands running Performance Max campaigns already qualify with no new setup. Google has the attribution wired up. ChatGPT doesn’t.

So what should you actually do with this?

  1. Don’t let FOMO drive a $200K bet. Self-serve access is coming this month. Wait for it. Test at a spend level where broken attribution doesn’t matter.
  2. If you’re already running retail media on Amazon or Walmart through a platform like Pacvue, pay attention to what they announce April 14. Managing ChatGPT from the same dashboard you use for Amazon changes the operational cost of testing dramatically.
  3. Google AI Mode is the more measurable bet right now for most brands. If you’re running Performance Max, you’re already there. Check your AI Mode placement data before you chase a new channel.
  4. Watch Perplexity. If the ad-free model wins consumer trust and starts pulling market share, that changes the math for every AI platform running ads. The “Google search model applied to AI” assumption isn’t guaranteed.
  5. Regardless of ads, get your product data agent-ready. Every one of these channels — ChatGPT, Gemini, Rufus, Perplexity’s Merchant Program — runs on structured product data. The brands that show up cleanly in AI recommendations, sponsored or not, are the ones with complete, consistent, real-time catalog feeds. That’s the work that compounds no matter which channel wins.

The AI recommendation layer is becoming a retail media channel whether anyone’s ready or not. The question isn’t whether to pay attention. It’s whether to pay for placement in a channel that can’t yet prove it works — while the alternative is an ad-free AI that might win on trust alone.

That tension is going to define the next 12 months of this space.