ChatGPT ads don't work yet. Here's the proof

We spent $477 on ChatGPT ads over six weeks and got 370 clicks, zero conversions and no revenue at all. I've analyzed what went wrong, and why ChatGPT ads are unlikely to work for any company selling products to a specific audience.

OpenAI launched ChatGPT ads in February 2026, and tens of thousands of businesses already use them. My company Barn2 tested the ads for four of our products between 1 June and 14 July 2026 and then stopped, disappointed with the results.

I'll give you the full spend, the traffic Google Analytics recorded and the evidence that the clicks were worthless. I'll also explain why I think ChatGPT ads as they're currently offered are fundamentally flawed, and designed to generate revenue for OpenAI rather than give advertisers properly targeted exposure.

How many websites are running ChatGPT ads?

Store Leads has detected OpenAI's advertising pixel (used for ChatGPT ads) on 38,269 ecommerce stores. That sounds like a lot until you see what it's a share of. Among the stores big enough for Store Leads to track, fewer than one Shopify store in 100 is running ChatGPT ads, and fewer than one WooCommerce store in 1,000.

This tells me that while plenty of ecommerce websites have already tried ChatGPT ads, the vast majority haven't. Several fellow business owners have told me that they're waiting to see how other companies get on with it. That's why I'm sharing the results of my own (failed) experiment.

19,154 ChatGPT ad impressions. Zero sales

We ran two campaigns. One promoted our hosted Document Library Pro service, and the other held separate ad groups for three of our WordPress plugins: Document Library Pro, WooCommerce Product Options and WooCommerce Wholesale Pro. Every ad paid per click and targeted the countries where most of our customers are.

Our ad account bills in British pounds, and I've converted every amount to USD:

Admittedly, $477 is a small budget by any standard. Even so, 370 clicks should produce something (I'll explain why below), and we got nothing.

We used Google Analytics to confirm that the clicks OpenAI claimed were real. The ads sent 210 sessions to barn2.com and 191 to documentlibrary.pro, which is 401 sessions from 374 people, which very close to OpenAI's claim. However, despite the clicks being real, neither site recorded a single conversion.

The country targeting was definitely working, so that's not the problem.  Our 401 sessions broke down as:

• United States: 162
• Canada: 117
• United Kingdom: 52
• Australia: 41

So why didn't the ad clicks from ChatGPT convert?

ChatGPT ads refer worse traffic than organic ChatGPT

Over that six week period, barn2.com received traffic from ChatGPT in two ways: 552 sessions where someone followed a link inside a ChatGPT answer, and 210 sessions from our paid ads.

The two groups behaved nothing alike:

Visitors who arrived from a ChatGPT answer spent almost 12 times longer on our site, and five of them bought something. Almost all of our ad clicks briefly looked at one page and then left.

As a result, I don't think ChatGPT's audience is the problem. People do arrive from ChatGPT ready to buy from us, just not when we pay to put ourselves in front of them. That's the opposite of what I would expect - paying should help you to jump the queue and reach a more targeted audience 🙃

The conversion rate our ChatGPT ads should have hit

Our ads sent people to four destinations: the three plugin pages on barn2.com, plus our separate Document Library Pro site. Taking the three barn2.com product pages on their own, I compared our ad clicks against how everyone else who arrives there converts.

Over the same six weeks, organic search sent 946 sessions to those pages and produced 23 sales, a conversion rate of 2.43%. Direct traffic converted at 1.52% and referral traffic at 1.03%.

At the organic rate, our 210 ad sessions from the ChatGPT ads should have produced about five sales. Getting zero instead has a probability of roughly 1 in 176, so this wasn't ordinary bad luck. Something was fundamentally wrong.

And it wasn't the conversion tracking. Google Analytics recorded between 42 and 66 sales a week on barn2.com throughout the campaigns, so it was logging purchases normally. I also checked all 373 people whose first visit came from a ChatGPT ad. In the four months to October, one of them reached a cart and none loaded a checkout page.

The problem: ChatGPT ads reach the wrong people

19,154 impressions sounds healthy until you compare it with how many people are actually looking for what we sell. At Barn2, we sell niche products that attract a very specific and limited audience.

I pulled US monthly search volume for 14 keywords covering everything we were advertising, including "woocommerce product add ons", "wordpress document library plugin" and "woocommerce wholesale pricing". Across June and July 2026 those 14 keywords were searched 1,370 times in total.

Nobody has to search Google before they see an ad, and ChatGPT matches ads to a conversation rather than to a keyword. Even allowing for both of those, the gap is too wide to explain any other way: I have to conclude that the ads were being shown to people with no interest in document libraries, product options or wholesale pricing.

ChatGPT ads can't target an audience

ChatGPT currently gives you four ways to target your ads:

• LocationCountry, plus cities and postcodes in some countries/states. We only showed the ads in the countries where we get the most sales.
• PlatformHow the user accesses ChatGPT, i.e. the iOS app, the Android app or the web. OpenAI's documentation notes that someone using ChatGPT in a browser on an iPhone counts as web, so this isn't even a mobile-versus-desktop split. I think it's fairly irrelevant anyway.
• Context hintsFree text where you describe the conversations your ad should appear in.
• Custom audiencesLists of email addresses or phone numbers you already hold, which you can include or exclude.

At first glance, both context hints and custom audiences sound like an opportunity to do real targeting. However, in reality they're not this at all.

Context hints don't do what they look like they do

When I set our campaigns up, this field looked incredibly promising. You write a description of your product, and I expected ChatGPT to use AI to show the ads to people asking about that product, or about the problem it solves. That's the whole promise of advertising inside an AI assistant.

As a result, we wrote careful prompts about what each product does, the key use cases and verticals, and it appeals to. This should have told ChatGPT exactly who to show the ads for. For example, if someone asked "How do I create an online document library?" then I would expect our ads to show up. If they asked something unrelated to that, then I would expect the ad not to be shown. Simple, right?

However, the impression data from our ads shows that it doesn't work like that. Our ChatGPT ads served 19,154 impressions in six weeks. Across two months, people searched for every related term just 1,370 times. This tells me that OpenAI showed our ads to people who were never going to buy one of our products.

Custom audiences only work in reverse

Custom audiences let you upload a list of email addresses or phone numbers. You can then tell ChatGPT to show your ads only to those people, or never to show them to those people.

Neither option helps you find a new customer. Inclusion needs at least 25,000 matched users and OpenAI recommends 100,000, which rules out most businesses our size. A smaller list can only be used to exclude people, and excluding someone means already having their email address.

Power users don't see ChatGPT ads at all

OpenAI's help documentation says ads appear for people on the Free and Go plans, and that Plus, Pro, Business, Enterprise and Edu accounts never see them. Our customers are either developers or people running a website and paying for software. They're far more likely to be on one of the more expensive plans, so our ads were never shown to the people we most wanted to target.

ChatGPT's ad reporting tells you nothing

OpenAI tells advertisers they "receive aggregated reporting such as views and clicks", with no access to chats, history or personal details. I understand the privacy reasoning, but that's not an excuse for how useless the reports truly are.

As an advertiser, you're almost completely blind. Having spent the money, I know how many people clicked. But I have no way of finding out what sort of people they were, what sort of topics they were asking about, or what type of conversations my ads turned up in. Every other ad platform I've used tells you something about who you reached.

ChatGPT could easily provide more helpful data in a non-identifiable, aggregated way - so privacy isn't an excuse for such opaque reporting. The current report makes it absolutely impossible to see what's working or fine-tune your campaign like you can with other types of paid ads such as Facebook or Google AdWords. Instead, your only option is to disable your ads completely and stop paying for them - so I did.

Would I run ChatGPT ads again?

Our ChatGPT ads are now paused, and I wouldn't advertise with them again until they significantly improve their targeting and make their reporting more transparent. Here's what I posted when I paused them:

Decided to pause our OpenAI ads - the number of impressions was unrealistic, which suggests that they are showing the ads where it's not relevant, and we should have had several conversions based on the number of clicks. I might try again once they've improved things a bit. pic.twitter.com/DrKeHVdoDI

— Katie Keith (@KatieKeithBarn2) July 14, 2026

In many ways, it's not surprising that we didn't get results from our ChatGPT ads. After all, it's a brand new advertising platform, and I'm sure they need to do a lot more development work on it. However, the sheer amount of wastage and the overwhelming evidence that our ads were shown to people who couldn't possibly be interested in our products is pretty damning.

When you set up ChatGPT ads, you get the impression that they'll use AI to intelligently show your ads to the right people. I saw no evidence that that was happening. Instead, your budget gets eaten up by ads shown to a ridiculously wide audience. That's the only explanation for the data.

If you sell very broad products that almost everyone would be interested in, then maybe ChatGPT ads are a good idea. For a niche product that needs careful targeting, like Document Library Pro, WooCommerce Product Options or WooCommerce Wholesale Pro, they're completely unsuitable.

I accept that one advertiser, six weeks, four products and 370 clicks isn't a study. I'd love to know how other companies have got on with ChatGPT ads and whether their findings match mine. Please leave a comment below or come and find me on social media.