Case study
The ChatGPT ads halo effect: clicks showed less than half of the orders
Dose, a liquid supplement brand, switched ChatGPT ads off in some US regions and left them on in others. The regions with ads sold far more than click tracking gave the ads credit for, on its own online store and on Amazon.
First published .
Bottom line
57%
of incremental online store orders were invisible to click attribution (it reported 1 in 2.3)
67%
of incremental orders came from customers who had never bought from Dose
34.5%
halo effect on Amazon orders, where an ad click cannot be followed
Source: a geo holdout case study on Dose published in October 2026 by a third-party measurement provider. Aiso did not run the test. The 57% is an Aiso calculation from the reported 2.3x: (2.3 − 1) ÷ 2.3 = 57%. It applies to online store orders only.
Clicks undercount ChatGPT ads
Click attribution only credits an ad when someone clicks it and buys in the same tracked session. ChatGPT does not work that way. People read an answer, close the tab and come back later through a search, a typed web address or Amazon. The ad did its job, but the click report never sees it.
Dose sells a doctor-formulated liquid supplement for heart and metabolic health, through its Shopify store, Amazon and some retailers. It was testing ChatGPT ads against a strict customer acquisition cost target. The team wanted to know whether the ads created new orders, or only claimed orders that would have happened anyway.
How the test worked
The test was a geo holdout. US regions (Nielsen designated market areas) were matched on seasonality and sales trends. Some regions had ChatGPT ads switched off as a control. The rest kept them on. The difference in sales between the two groups is the ads' real effect. Sales were measured on both the online store and Amazon.
What it found
- ChatGPT ads drove 2.3 times more incremental online store orders than click attribution reported. Put another way, clicks showed about 43% of the real effect and missed about 57%.
- 67% of the incremental orders came from net new customers.
- Amazon orders showed a 34.5% halo effect. No click links a ChatGPT ad to an Amazon order, so this sale is invisible to ad platform reporting.
- The case study says the Amazon halo lowered blended acquisition cost, but it does not give a figure.
What the numbers do not show
This is one brand, in one category, in one published case study. It does not give the test dates, ad spend, the number of regions, or confidence intervals. The provider that published it also sells the measurement. Treat the figures as a strong signal, not a benchmark for your brand.
Why this matters for organic AI visibility
The halo works because ChatGPT plants a brand name that people act on later, somewhere else. The same thing happens with organic answers. When ChatGPT recommends you without an ad, buyers also go on to search for you, or find you on Amazon, and no click report shows where they came from.
Paid and organic support each other. An ad puts you in front of the buyer once. If the organic answer to their follow-up question names a competitor, the ad has helped someone else. Before you scale ChatGPT ads, check how often the free answer already recommends you for the same questions.
What to do about it
- Judge ChatGPT ads with a holdout or other incrementality test, not last-click reports.
- Measure the effect on every place you sell, including Amazon and retailers, not only your own store.
- Track how often ChatGPT names your brand in organic answers for the questions your ads target.
- Fix the pages ChatGPT reads about you first, then work on relevant third-party sources. Our guide on why AI mentions your brand explains where to start.