Most Shops Got No Orders From ChatGPT Ads. Is It the Channel or Your Setup?

Most shops that bought ChatGPT ads recorded no order from them. That is the finding of the first large study of the channel, a working paper by Maximilian Kaiser and Christian Schulze dated October 2026. It follows 1,898 online shops through OpenAI's country by country ad rollout, using the shops' own Google Analytics ecommerce data supplied by Grips Intelligence.

In US traffic, a visit from a paid ChatGPT ad produced $0.89 in revenue against $2.45 for a paid Google search visit. Paid ChatGPT visitors converted at 0.37 percent against 1.17 percent. Of the 87 advertisers buying both channels, 73 saw no paid ChatGPT purchase at all across 18,726 sessions. The remaining 14 carried every one of the 570 orders.

The same shops tell a second story that matters more for a lean ecommerce team. Unpaid ChatGPT referrals, the visits that arrive when the assistant recommends a product in its answer, earned $3.27 per session in US traffic. That was the highest of the five channels the authors compared, ahead of paid search, direct and organic search.

What the study does not say

The paid comparison is descriptive, not causal. Advertisers chose to buy ChatGPT ads, so they were not assigned at random. The data stops on August 22, 2026, which covers only the first 28 weeks of ads and predates the measurement options OpenAI added in October. It records the last click only, so a purchase made later through another route is not counted. It cannot see prompts, impressions or which products the ad showed.

One advertiser accounts for 66 percent of all paid ChatGPT purchases in the sample. Removing it makes the channel look worse, not better, so the shortfall is broad. Still, the panel's largest paid advertiser converted 0.76 percent of 52,232 sessions and earned $1.99 per visit. Some accounts clearly make the channel work. The paper cannot say what they did differently, and that gap is the useful part for a buyer.

The organic finding has its own limits. Ads arriving in a country cut unpaid ChatGPT referral volume by roughly 6 to 8 percent, a range across assumptions rather than a tight estimate, while the conversion rate of the visitors who still arrived did not move. Treat it as evidence that the answer and the ad are separate systems, not as a forecast for your store.

The decision this puts on your desk

If you run ChatGPT ads, the question for the holiday budget is whether your results are a setup problem or the channel norm. If you do not, the question is whether the organic answer already names you, because that is where the measured value sits. We covered the sequencing argument in whether to buy ChatGPT ads before you show up in answers. The new study gives that argument numbers.

Benchmark your paid numbers against the panel, not against hope

Pull revenue per paid ChatGPT session and the conversion rate for the campaign to date. The median advertiser in the study sat at zero orders and $0.00 per visit, and the 90th percentile reached 0.14 percent and $0.25. A few hundred sessions with no purchase is the panel norm, not proof that tracking is broken. Any order at all puts you in the top sixth of the sample.

Then compute your own parity price. Divide your revenue per paid ChatGPT session by your revenue per paid search session and multiply by your paid search cost per click. The authors did this on the panel and arrived at about $0.26 per ChatGPT click, against a known account paying $1.72 and OpenAI's suggested opening bid of $3 to $5. Your number decides whether the bid is defensible.

Rule out the setup causes the study could not see

Before cutting, check the three things the data was blind to. First, which products actually served. A campaign optimizing against a feed where half the catalog is blocked or stale sends the wrong products to the wrong prompts. Second, whether each landing page states the facts the prompt asked for, since the authors' own reading is that the sponsored card sits apart from the answer, so a click needs the page to finish the argument.

Third, the event path. If purchase events are missing or doubled, the comparison with paid search is meaningless. Our checklist for ChatGPT Ads conversion warnings walks through that. Since late October, ChatGPT can also rewrite headlines and change landing pages by default, so confirm the versions that ran are the ones you reviewed before judging the creative.

Put the organic answer first in the holiday plan

In the same shops, an unpaid ChatGPT referral was worth more than three paid ones. That visitor arrives because the assistant chose the product from its information: the feed, the product page and what third parties say. Write ten to twenty prompts the way a first time buyer in your category types them, run them in ChatGPT and Google AI Mode, and record who is named and which pages are cited.

Where your products are missing, trace the gap back to the product record. Missing attributes, a stale price, an availability value that disagrees with the page, or a description that never mentions the use case the prompt asked about. Shopify's new channel mapping page shows which feed fields are empty per channel, which makes this triage faster for Shopify stores.

What to leave for later

Do not rebuild attribution to prove the ads work. The study found no detectable lift in other channels after an advertiser's first paid click, within wide bounds, so a view through story needs its own holdout test, not a dashboard setting. Do not raise bids to buy scale before the per product numbers justify it. And do not read the drop in organic referral volume as a reason to advertise, because the shoppers who still arrived converted exactly as before.

A ChatGPT ads line is a reasonable small test when the organic answer already names you, the feed is clean and the event path is verified. Keep spend, clicks and attributed purchases separate from appearances in unpaid answers, so the two decisions stay separate.

This is the work Paz does for lean ecommerce teams. A Paz expert finds the product information gaps that keep you out of the answer and off the eligible product list, you approve the fixes, and the platform publishes them and tracks appearances and referrals against a baseline.

For ChatGPT Ads, the expert chooses the products, clears what blocks them and writes the ad copy from your sources, while your media buyer keeps the budget. If you want to know whether your ChatGPT results are the channel or the setup, talk to us.