$477 on ChatGPT ads, 370 clicks, zero sales: what it means for your budget
Somebody spent $477 on ChatGPT ads over six weeks, got 370 clicks, and made zero sales.
That somebody is Katie Keith, who runs Barn2, a company that sells WordPress plugins. She didn’t just quietly eat the loss. She published the whole thing: ChatGPT Ads Don’t Work Yet. Here’s the Proof. (October 6, 2026). I’d read it before you put a dollar into AI advertising. Credit to her for sharing numbers most people would bury.
I run a small agency, not an ad lab, so I’m not going to pretend I have a better test. What I can do is tell you what her numbers mean if you’re an owner deciding where your next $500 goes.
The numbers, from her post
Everything below comes from Katie’s write-up. The test ran for six weeks, June 1 to July 14, 2026, across two campaigns.
- Spend: $477
- Impressions: 19,154
- Clicks: 370 (a 1.93% click-through rate)
- Sessions: 401, from 374 people
- Conversions: zero
Here’s the part that stuck with me. People who arrived from the ads spent about 37 seconds on the site. People who arrived by clicking a link inside a ChatGPT answer spent about 443 seconds. Same chatbot, completely different visitors. In her breakdown, the organic ChatGPT referrals produced 5 sales. The paid ones produced none.
She also noted that her ads got far more impressions than there were relevant searches to match, around 14 times as many by her count, and concluded the ads were being shown to people with no real interest in the products. Her advice for now is to pause until targeting and reporting get better.
What this does and doesn’t prove
It’s one company, one niche product category, one six-week window. Barn2 sells plugins to a specific audience. A plumber, a law firm or a local restaurant is a different animal, and I wouldn’t claim her result predicts yours.
Still, it’s real money, real tracking, and a published result. That’s more than most hot takes on this topic have. And it lines up with something I keep coming back to: the people who find you inside an AI answer behave very differently from people who get shoved an ad.
Buying a slot vs. earning a mention
Think of two ways to show up when someone asks a chatbot for a recommendation.
One is paying to appear next to the answer. It’s early, the targeting is thin, and as Katie’s test suggests, the reporting is still catching up.
The other is being the source the answer is built from. Nobody’s selling that slot. You earn it with clear pages, direct answers, real facts and a site that machines can read. Her own data hints at why it’s worth chasing: visitors who came that way stayed longer and bought.
This could change fast, and when it does I’ll happily eat these words. Until then, the cheap, boring work of being the clearest answer on the internet pays off no matter which way the ad products go.
Where I’d put the first $500
If you asked me over coffee, and you had $500 earmarked for “AI stuff,” this is the order I’d work in. It’s my opinion, not a guarantee.
- Find out where you stand. Ask the major assistants the questions your customers would ask. Does your business show up? Our AI visibility self-check walks through it, and it costs you an afternoon, not a budget.
- Fix what keeps you out of the answer. Thin service pages, vague headlines, no direct answer near the top, key facts buried in a PDF. I covered the mechanics in how to get cited by AI Overviews and ChatGPT.
- Check that AI crawlers can actually reach you. A surprising number of sites block them by accident. The Cloudflare AI bot deadline post explains one common way that happens.
- Publish something worth citing. One page with a real number, a real process or a real answer beats ten generic posts.
- Measure it. Watch referral traffic from AI tools and compare it to everything else, the way Katie did. If you can’t tell where visitors came from, fix that before you spend anything.
None of that needs an ad account. Most of it needs a few hours and somebody willing to rewrite a page or two.
If you do want to test ads anyway
Fair. Curiosity is a good reason to run an experiment. Just treat it as one.
- Set a hard cap you can lose without flinching, and a stop date.
- Decide in advance what counts as a win: a sale, a booked call, a form fill.
- Track time on site and engagement, not just clicks. Katie’s 370 clicks looked fine until she looked at what those visitors did.
- Send them to a page built for one job, not your homepage.
I’m not anti-ads. I’m anti-spending before you know what you’re buying, which is how I’d describe a lot of early-stage ad products.
Want a starting point?
If you’d like help with the earn-it side, that’s what our AI SEO work is for. The SEO Snapshot is a quick way to see where your site stands before you decide on anything bigger.
Have you tried any AI advertising yet? I’d like to know what you saw, especially if it went better than Katie’s.
