Nobody can tell you from benchmarks whether ChatGPT Ads will pay off, because there aren’t any yet. OpenAI says so itself. What you can do is judge the fit before you spend, and run a test that gives you a real answer. This guide covers what OpenAI has published, what it hasn’t, and how to find out for your business.
All figures are OpenAI’s, selected and measured by OpenAI. Where we give our own view, we say so.
The short answer
- OpenAI publishes no performance benchmarks for ChatGPT Ads yet.
- The figures it does publish show spending and a few success stories, not typical results.
- Whether ads pay off depends on your margins, your category and your tracking.
- The only way to know is a test with a target written down first.
What OpenAI says about performance
OpenAI’s advertiser FAQ is direct. As an early platform, ChatGPT Ads doesn’t yet have performance benchmarks across advertisers, industries or campaign types. It says results depend on the campaign objective, creative, landing page, relevance to the user and overall setup.
It also describes the platform as a beta in the early stages of scaling. That follows a focused pilot from February to April 2026. Delivery and spend can fluctuate as OpenAI improves its systems. For conversion-optimized campaigns, the goal is better conversion outcomes, not spending the full budget.
The numbers OpenAI has published
OpenAI has shared a handful of figures. Each tells you something, and each leaves something out.
| Figure | What it says |
|---|---|
| $1B run rate | ChatGPT Ads reached $1 billion in annualized revenue in under 200 days. |
| Advertisers | Tens of thousands, with small and medium-sized businesses a material share since Ads Manager opened in May. |
| Buying models | Cost-per-click and outcome-optimized bidding now account for the majority of campaigns. |
| One store | An ecommerce advertiser reached a 3x return on ad spend across its campaigns over 28 days. |
| New customers | A technology partner reported that more than 80% of ad-driven ChatGPT traffic is from new customers. |
| User response | During the pilot, OpenAI saw no impact on consumer trust metrics, low ad dismissal rates and improving relevance. |
| Against paid search | DV Rockerbox measured WeightWatchers’ attributed cost per acquisition at 15.3% below its blended paid-search benchmark. |
| Incremental orders | WorkMagic’s geo-lift study estimated 2.3 times more incremental orders for Dose than last-click attribution captured. |
| View-throughs | In an early analysis, 52.7% of eligible one-day view-through conversions came within an hour of the ad impression. |
How to read those numbers
None of these figures is an independent benchmark. The table shows what each can and can’t support.
| Figure | Useful for | Not useful for |
|---|---|---|
| $1B run rate | Knowing advertisers are spending | Knowing whether any one advertiser profits |
| Tens of thousands of advertisers | Seeing the channel is real | Estimating your odds |
| CPC and outcome bidding as the majority | Seeing advertisers buy performance, not only reach | Choosing your own objective |
| 3x return for one store | A sense of what a good outcome can look like | Predicting yours. It’s one example, picked by OpenAI |
| 80% new customers | A hint that ads reach new people | Any claim about your traffic. A partner reported it |
| Low dismissal and no trust impact | Confidence the format is accepted | Anything about conversions |
| 15.3% below paid search | Seeing the channel compete with search for one brand | Your own costs. A partner measured one advertiser, featured by OpenAI |
| 2.3 times the orders last-click showed | Knowing click-only reports can undercount | Assuming yours undercount by the same amount |
| 52.7% of view-throughs within an hour | Judging whether a one-day view window is reasonable | Proving the ad caused the purchase |
OpenAI shows demand, and a few wins. It doesn’t show the distribution of results, which is what you’d need to set a target.
What’s missing
Three gaps matter when you decide.
Independent benchmarks
We haven’t found a methodologically sound benchmark for click-through rate, cost per click or conversion rate. Figures circulating online tend to come from small individual tests or from vendors. Treat them as anecdotes.
OpenAI’s measurement partners have started to publish results, but each covers one advertiser that OpenAI chose to feature. Its incrementality work with Haus, Measured and WorkMagic is, in OpenAI’s words, still in its early stages.
Current prices from the early days
Reports of a $60 CPM and a $200,000 minimum described the pilot, before self-serve buying. They aren’t today’s prices. Our cost guide covers how pricing works now.
Everyone as an audience
Only people on the Free and Go plans, or using ChatGPT logged out, see ads. Plus, Pro, Business, Enterprise and Edu accounts never do. See who sees ChatGPT ads.
When a test is likely worth it
These criteria are ours, drawn from how the platform works.
- People research and compare options before buying what you sell. OpenAI describes ChatGPT as the place where people explain goals and weigh decisions.
- Your category is allowed in the markets you’d target. Check the policy tracker.
- Your buyers are likely to use ChatGPT on the Free or Go plans rather than through a company plan.
- You can track conversions on your own site, so you judge results on outcomes.
- You can fund a test for several weeks. In the US, each campaign needs at least $25 a day.
When to wait
Waiting is the right call in four cases, and each has a different fix.
- Your category is disallowed, or restricted to approved advertisers in your market.
- Your buyers mostly use ChatGPT through work Business or Enterprise plans, which never show ads.
- You can’t track conversions yet.
- Your business isn’t based in one of the 63 countries where self-serve buying is open.
What a test costs at the minimum
OpenAI sets a minimum daily budget per campaign that depends on the billing currency, $25 a day in US dollars. One campaign at that minimum for 30 days is $750.
That’s only the floor. A test needs enough budget to reach the number of conversions you’ll judge by. Our 30-day test plan shows how to work that out from your own cost per click and conversion rate.
How to find out whether ChatGPT ads are worth it for you
Six steps give you an answer you can act on. Do them in order.
- Work out your break-even point. Value per conversion times conversion rate gives the most you can pay per click. The cost calculator does the arithmetic.
- Set up tracking first, and test that events arrive before any money is spent. See our tracking guide.
- Agree success metrics and stop rules before launch. Decide the cost per conversion you need, and when you’ll stop if you’re not getting it.
- Start focused. Use one or two campaigns with specific landing pages, several distinct ads and well-written context hints.
- Allow for the delays. Attributed conversions take 24 to 48 hours to report, so give a test several weeks before you judge it.
- Check against your own data as well as Ads Manager. Then decide to scale, reshape or stop.
A scorecard for your fit
Score your business on five questions. Each “no” is a reason to wait or to fix something first.
| Question | If the answer is no |
|---|---|
| Do buyers research before they buy? | A cold click is less likely to convert. |
| Is your category allowed where you’d advertise? | You can’t run the ads at all. |
| Do your buyers use ChatGPT on a free plan? | Many may never see your ad. |
| Can you track conversions? | You can’t judge the result. |
| Can you fund several weeks of testing? | The test will end before it answers. |
Four or five yes answers make a test worth running. Two or fewer mean fixing the gaps first.
What “worth it” means in numbers
Return on ad spend only means something against your margins. To break even on gross profit, you need a return of at least one divided by your gross margin. At a 40% margin that’s 2.5x: every $1 of ad spend has to bring in $2.50 of sales.
| Gross margin | Break-even return on ad spend |
|---|---|
| 20% | 5.0x |
| 25% | 4.0x |
| 40% | 2.5x |
| 50% | 2.0x |
| 60% | About 1.7x |
By that measure, OpenAI’s 3x example would be profitable at a 40% margin, but only just. $3 of sales at 40% leaves $1.20 of gross profit for each $1 spent. At a 25% margin, the same 3x would lose money. This is arithmetic, not a forecast, and it’s why we set targets from your margins rather than from anyone’s headline.
A second worked example, for leads
Not every business sells online. The same arithmetic works for a business that sells through leads. The numbers are illustrative.
- An average deal is worth $4,000 in sales, at a 50% gross margin. That is $2,000 of gross profit.
- One lead in ten becomes a customer, so a lead is worth $200 of gross profit.
- If one click in twenty becomes a lead, a click is worth $10.
- A click costing more than $10 loses money at that close rate.
OpenAI suggests starting click bids at $3 to $5, so this business has room. Change the close rate to 3% and the break-even click falls to $3. The answer swings on numbers only you have.
What would change the answer
The platform is moving. OpenAI says its next phase will bring more markets, formats, objectives, buying options and measurement. A channel that isn’t worth testing today may be worth a look in six months, and the reverse is possible too. Revisit the question when your numbers or the platform change. Keep your break-even figures somewhere you can find them, because each new format or objective is a reason to rerun the arithmetic. The same arithmetic works for every new option OpenAI adds.
Questions
What results should I expect?
OpenAI doesn’t publish benchmarks, and neither do we yet. Expect results to depend on your category, offer, landing pages and tracking. Set your targets from your own margins, then run a test to see how far from them you are.
Are ChatGPT Ads worth it for small businesses?
They can be tested cheaply. US campaigns start at $25 a day, and OpenAI says small and medium-sized businesses are a material share of advertisers. Whether it pays depends on the same fit questions as for anyone else, and on whether you can track conversions.
How long before I know?
Plan for several weeks. Conversions take a day or two to report. Results swing from day to day. You also need enough conversions to tell a trend from noise. Agree the point at which you’ll decide before you start.
Should I test ChatGPT Ads before Google Ads?
Usually not if you haven’t covered the searches people already make for what you sell. Our comparison of ChatGPT Ads and Google Ads goes through when each makes sense.
Is a 3x return realistic for me?
It’s possible, and it’s one example. OpenAI doesn’t say how typical it is, or what the advertiser’s margin was. Compare it with your own break-even return from the table above before you treat it as a target. A 3x return that is profitable for one business loses money for another.
What if my test fails?
Stop at your stop rule, and write down what you learned. A failed test with a stop rule costs less than a campaign that drifts for months. You can change one variable, such as the offer or the landing page, and test again later.
Want to know whether your business fits before you spend? Tell us about your business.
