ChatGPT Ads 2026: The Self-Serve Test Playbook

A message lands in your Slack. “Are we on ChatGPT ads yet?”

You’ve dodged that one for months. And you had a good reason.

The pilot launched in February behind a gate: a $200,000 to $250,000 commitment just to walk in. Easy to say “not this quarter.”

That gate is gone.

In four months the entry cost fell from a quarter of a million dollars to nothing. Self-serve. No agency. No waitlist. Clicks are landing around three dollars.

So the excuse died. The decision did not.

Here’s what you’ll get: whether ChatGPT ads deserve a slice of your Q3 budget, how to track them when the platform hides half the numbers, and how to call the test before the auction fills up.

Not vibes. A framework.

What actually changed with ChatGPT ads in 2026?

ChatGPT ads went self-serve. OpenAI removed its platform minimum on May 5, opened cost-per-click bidding with a $0 floor, and added conversion tracking. Criteo cut its managed retail minimum to $10,000. OpenAI then published terms for first-party audiences and AI-built creative, and the pilot spread to nine countries.

The speed is the story. Here is the compression, one lever at a time.

Date	What moved	Entry cost
Feb 9	Pilot launches at a $60 CPM	$200K to $250K
May 5	OpenAI opens self-serve Ads Manager, adds CPC	$0
Mid-May	StackAdapt drops its minimum for ~1,000 advertisers	$0
Jun 4	Criteo cuts its managed retail minimum	$10K
Jun 17	Ad Tools Terms define audience upload + AI creative	(policy)

That is a drop from a quarter of a million dollars to zero across the self-serve routes, with Criteo holding a $10,000 managed tier for retailers in between, after it cut its threshold from $50,000 on June 4.

The reach behind it is not in dispute. When TechCrunch covered the February launch, ads were a curiosity. By an eMarketer forecast in June 2026, ChatGPT reaches nearly three-quarters of the 152.9 million Americans who use generative AI.

A quarter-million-dollar door became a free one in a single quarter.

Why the cheap window is genuinely closing

The auction is cheap because almost nobody is bidding in it yet. That does not last.

Look at the price. The pilot opened at a $60 CPM. By mid-April, buyers were reporting it as low as $25. One media buyer put it plainly: the cheap early-access window is closing fast.

Now look at the cadence. New features are landing most weeks. MediaPost reported on July 7 that OpenAI is shipping ad and targeting updates faster than most marketers can track: custom audience uploads, AI that writes the ad for you, advanced bid adjustments.

Then look at the map. OpenAI’s VP of Monetization Benji Shomair announced the pilot would push beyond the US into the UK, Japan, South Korea, Brazil and Mexico, bringing the footprint to nine countries. Ads stay in the free and Go tiers only. Hold that fact. It decides who should test at all.

Every one of those moves pulls the channel closer to Google and Meta parity.

The cheap auction is a moment, not a channel.

The enemy here is your own FOMO, working both ways. Rush in on hype and you’ll waste budget on a channel you can’t measure. Wait for it to feel safe and you’ll pay Meta prices for what costs a third of that today. The move is neither. It’s a small, instrumented test, run now.

Your clicks are lying to you

Here’s the part the demo won’t show you.

The clicks the platform reports run hot against what GA4 counts. Buyers testing the beta say the reported number inflates versus their own analytics. Some see roughly half the clicks register downstream at all.

And the visibility is thin. One advertiser three-plus weeks in said it straight: the biggest issue is not knowing what to change, because there is no reporting on how targeting affects match quality. Another called tracking performance a headache compared to a normal paid channel.

There is a deeper problem than reporting, though.

Intent on ChatGPT is mid-funnel, not bottom.

People are asking questions, not pulling out a card. The traffic is real, but early testers describe “curiosity clicks for days.” One ran a campaign that drove plenty of clicks and a single conversion, while organic ChatGPT referrals converted far better. That is the signature of a discovery surface, not a closing one.

So last-click attribution will undersell this channel every time. A curious click today becomes a branded search or a direct visit next week, and your last-click report hands the credit to Google.

Judge ChatGPT ads the way you judge bottom-funnel search and you’ll kill a channel that was working upstream. It’s the same trap we broke down in our guide to beating the ROAS mirage: the platform number is not the business number.

Should you test ChatGPT ads right now?

Not everyone should. The channel is genuinely wrong for some brands right now, and knowing which side you’re on saves you a wasted month.

Test now if you are one of these.

• A retailer with a live product feed. Criteo lets you connect an existing catalogue straight into the auction, no manual ad build. Criteo reports its referred traffic converting around 1.5 times other channels, with some retail categories near twice traditional search. Treat those as vendor-stated figures from its own clients, not audited numbers, but the feed shortcut is real.
• A considered, mid-funnel purchase. Software, services, anything a buyer researches before choosing. That is exactly the intent ChatGPT surfaces. One agency reported closing new clients inside the first few weeks, with lead quality mid-to-bottom funnel at a better cost than Google.
• Sitting on a clean first-party list. The new audience tools reward brands whose customer data has a real consent trail. Craig Graham, CEO of Grayvault Consulting, told MediaPost the upload feature “moves the platform a step closer to the kind of first-party data targeting that serious advertisers already rely on in Google and Meta.”

Wait if you are one of these.

• High-ticket, and your buyers pay for Plus or Pro. Ads only serve to the free and Go tiers. If your customers are the ones paying to remove ads, they will never see yours. You will fund free-tier curiosity clicks from people who will not buy. One high-ticket ecommerce buyer called it exactly that: wasted money.
• A pure last-click, bottom-funnel shop. If your whole model is capturing existing demand at the moment of purchase, a discovery channel you can barely attribute is not your first dollar this quarter.

How to run a disciplined test without overcommitting

You’re testing a channel you cannot fully see. So the discipline goes in before the spend, not after.

Step 1: Instrument tracking first. Tag every campaign with strict UTMs. ChatGPT strips referrer data, so your analytics will not name the source on its own.

Wire the Conversions API and the pixel. Build a custom channel in GA4 so this traffic never hides inside “direct” or “referral.” Testers confirm the fix: with a standard UTM and some backend tooling, the gap closes.

Step 2: Cap the budget and the clock. A small daily budget and a two-to-four-week window is plenty to read a signal. This is a read, not a scaling line. Run it the way we lay out in our guide to structured ad testing: one hypothesis, a fixed window, a decision at the end.

Step 3: Judge on blended MER, not platform ROAS. Watch your whole-account marketing efficiency ratio across the test window. If total revenue lifts against total spend while ChatGPT is live, the channel is contributing, whatever the last-click report says.

Set your kill line and your scale line before you launch. That way a curiosity-click week does not spook you into cutting a mid-funnel winner early.

Step 4: Answer the question they’re already typing. This is where a conversational channel is won or lost. Your ad shows up inside someone’s question. If your headline speaks to the exact problem, objection and language they use, it lands. If it’s generic, it’s scenery.

And that is a customer-knowledge job, not a copywriting one.

You cannot guess the words. You get them by talking to the people who buy.

hollie has real conversations with your customers and brings back their exact questions and objections, ranked. So you build ads on what they actually say, not what you hope they think.

It’s the same discipline behind asking the questions that get honest answers out of customers, done at scale. Start from real customer voice before you write a line.

Frequently asked questions

Is there a minimum spend for ChatGPT ads?

No. OpenAI removed its platform minimum on May 5, 2026, and the self-serve Ads Manager runs with a $0 floor and roughly a $200 daily cap. StackAdapt also dropped its minimum to zero. Criteo keeps a $10,000 managed tier aimed at retailers with a product feed.

What do ChatGPT ads cost right now?

Cost-per-click is landing near $3, with some buyers reporting $4 to $5. CPMs fell from about $60 at launch to as low as $25 by mid-April. Early click-through rates are low and spread wide, from near zero to about 2.4 percent, under a typical Google Search benchmark.

Who actually sees my ChatGPT ad?

Only the free and Go subscription tiers. Plus, Pro and Enterprise subscribers see no ads at all. As OpenAI’s Benji Shomair put it, ads do not influence ChatGPT’s answers and are always clearly labelled. That tier limit is the single most important factor in whether the channel fits your buyer.

Can I measure conversions from ChatGPT ads?

Partly. You can set up conversion tracking through the Conversions API and pixel, but the in-platform numbers run hot and referrer data gets stripped. Lean on disciplined UTMs, a GA4 custom channel, and blended marketing efficiency rather than trusting the platform’s reported clicks.

The window is open. The excuse is gone.

ChatGPT ads are cheap because you’re early. That is the whole opportunity, and it is also the whole risk. Move on hype and you burn budget you can’t measure. Freeze and you pay double next year.

Instrument. Test. Judge.