A new ad channel with hundreds of millions of users and almost no playbook. Here's what's real, what it costs, and how to test it without wasting budget.
Every few years a new ad channel shows up and the same thing happens. Half the industry rushes in with too much money, the other half waits too long, and a small group quietly figures out what works while the clicks are still cheap. ChatGPT ads are that channel right now.
I've had the same question from clients almost every week since OpenAI opened its self-serve Ads Manager: should we be testing this?
My honest answer is "probably, but not the way most people are doing it."
This guide covers how ChatGPT ads actually work, what they're costing advertisers so far, who should test them now and who should wait, and how to set up a first test that tells you something useful. A lot of what's been written about this platform is either hype or guesswork, so I'll flag clearly where the numbers are still early.
ChatGPT ads are sponsored placements that appear below ChatGPT's answers, clearly labelled and visually separate from the response itself. Each ad shows your brand name, a headline, short copy, an image and a link to your landing page.
A few ground rules shape everything else:
That last point matters more than people realise. On Google, you pick the words. With ChatGPT ads, the platform reads your page and decides which conversations you belong in. Your landing page is effectively your targeting.
The platform has moved fast this year. Here's the short version, based on OpenAI's announcements and industry reporting:
In Europe, ads are selected by conversation topic, general location and device type rather than a personal profile. If you sell into the UK and EU, plan around contextual relevance, not audience tricks.
This is where I'd urge some caution, because nobody has a large, reliable dataset yet. What's been reported so far:
Those numbers aren't cheap. For many ecommerce categories, they're higher than what you'd pay on Meta and comparable to competitive Google search terms. You're paying for an audience in the middle of researching a decision, and right now you're also paying a novelty premium.
For a meaningful first test, I'd budget at least $1,500 to $3,000 over three to four weeks. Spend much less and you won't get enough clicks to learn anything. Spend much more before you've seen real conversion data and you're gambling.
Not every business is a good fit yet. Here's how I'd split it.
If you're in a restricted niche, your energy is better spent on channels where approval is possible today. We've covered how we run profitable ads in restricted niches separately.
If you've decided to test, this is the setup I'd use. It's built to answer one question: do ChatGPT ads bring in customers at a cost you can live with?
Set up OpenAI's pixel and Conversions API, and add UTM parameters to every ad link so the traffic shows up separately in GA4 and your store analytics. Some agencies have already reported gaps between what the platform reports and what their own tracking shows. Treat your store's data as the source of truth.
Since the platform reads your page to decide where your ads fit, a thin product page with three bullet points won't get matched well. Write the page the way you'd answer a customer's questions: who it's for, how it compares, what it costs, what happens after purchase. Our funnel marketing team rebuilds pages this way for exactly this reason.
Someone reading a ChatGPT answer is in research mode. Ads that read like a useful next step ("Compare our three mattress firmness levels side by side") fit the moment far better than shouty discount banners. Keep the copy specific and calm.
Begin with the Clicks objective at the recommended bid range so the system can gather data. Once you've got a steady flow of tracked conversions, test conversion-based bidding against it. Don't switch objectives every few days; give each setup at least a week.
Clicks from ChatGPT ads can look great on paper. What matters is cost per acquisition and revenue, checked against your store. Compare the test period with the weeks before it. If total sales didn't move, the channel didn't earn its budget, whatever the dashboard says.
I see a lot of panic about AI search replacing Google. In practice, the two do different jobs.
Google captures people who already know roughly what they want and type it in. ChatGPT catches people earlier, while they're still working out which option suits them. That makes ChatGPT ads closer to a mid-funnel channel, somewhere between social discovery and search intent.
So don't pull budget from campaigns that work. Carve out a separate test budget, usually 5 to 10% of your paid media spend, and keep your ecommerce ads on Google and Meta running as normal.
ChatGPT ads are real, they're growing quickly, and they put your brand in front of people at the exact moment they're researching a decision. They're also expensive, early and lightly measured.
The smart move for most advertisers is a small, well-tracked test with a landing page built to answer questions, judged on actual revenue. If it works, scale it slowly. If it doesn't, you've spent a few thousand to learn something your competitors are still guessing about.
Either way, keep the channels that already pay your bills running while you find out.
We'll look at your product, margins and current tracking and tell you honestly whether ChatGPT ads are worth testing for your brand yet, and how we'd structure the test if they are.
Talk to Our Team