A working paper examining ChatGPT’s international advertising rollout suggests that paid traffic from the platform has, at least in the observed data, generated substantially less revenue per U.S. visit than paid Google Search. The research also found that unpaid ChatGPT referrals declined during the initial rollout period, raising an important measurement question for publishers and marketers: whether AI advertising changes the volume of organic discovery even when the visitors who still arrive convert at similar rates.
The findings are not a universal verdict on ChatGPT advertising or AI referral traffic. They are based on outside-the-platform data, last-click attribution and a defined collection window. Still, they provide an early, unusually specific comparison of monetized paid visits and organic referrals as ads enter an AI discovery environment.
What the paper measured
Maximilian Kaiser and Christian Schulze studied ChatGPT’s international ad rollout using data from 1,898 websites in an organic-traffic panel and 387 websites receiving paid referrals. The observation period covered 60 weeks, from June 29, 2025 through August 22, 2026.
The researchers used the European Union as an untreated control and compared it with markets where the advertising treatment occurred. Their approach relies on Grips Intelligence data and last-click attribution rather than prompt-level data from ChatGPT. That distinction matters: the paper can observe referred visits and reported outcomes, but not the prompts, user intent or complete path that preceded every visit.
In the U.S. data cited in the study, paid ChatGPT visits produced $0.89 in revenue per session, compared with $2.45 for paid Google Search visits. The reported conversion rates were 0.37% for paid ChatGPT and 1.17% for paid Google Search.
| Observed U.S. paid-traffic metric | ChatGPT ads | Google Search ads |
|---|---|---|
| Revenue per session | $0.89 | $2.45 |
| Conversion rate | 0.37% | 1.17% |
The authors estimate a parity price of about $0.26 per ChatGPT click, meaning a price at which the channel would match Google’s revenue-per-click performance in their analysis. They also report examples of accounts paying materially more, including $1.72 per click, while still seeing weaker revenue outcomes.
These are study findings, not published platform benchmarks. Performance can vary by advertiser, offer, landing page, country and attribution model. However, the comparison is a useful reminder that a click from an AI interface should not automatically be priced or valued as if it carries the same commercial intent as a Google Search ad click.
For the full methodology and results, see Kaiser and Schulze’s working paper.
Organic ChatGPT referrals fell, but conversion did not materially change
The paid-traffic gap is only part of the story for site owners. The paper estimates that organic referrals from ChatGPT fell by about 5.9% over the first 13 weeks of the advertising rollout. The reported range was roughly -12.6% to +1.2%.
When the comparison was adjusted to North America versus the EU, the estimated decline was larger, at about 7.2% to 8.4%. Yet conversion on organic ChatGPT referrals was essentially unchanged: the researchers report a -0.02 percentage-point movement from a 2.21% baseline.
That pattern is significant because it separates traffic volume from visitor quality. In this dataset, the rollout was associated with fewer unpaid referrals, but it was not associated with a meaningful deterioration in the conversion rate of the unpaid visitors who continued to arrive.
It does not prove that ads alone caused every referral change, nor does it establish that the same effect will occur for every website or future rollout. The EU control design strengthens the comparison, but the study still uses external traffic data and cannot account for every platform, user-behavior or market change. OpenAI also expanded ChatGPT Ads to additional countries after the data window, so the results should be treated as early evidence rather than a final picture of a broader expansion.
Why the results matter for search visibility measurement
AI referral reporting can become difficult to interpret when organic recommendations and paid placements coexist in the same product. A rise or fall in visits alone does not reveal whether the change came from ranking, citation visibility, ad exposure, changing user behavior or differences in how the platform sends referral data.
For publishers and SEO teams, the study supports a more disciplined view of ChatGPT and other AI referrals:
- Track organic AI referrals separately from paid AI traffic. Do not combine them into one channel simply because both originate from the same platform.
- Compare sessions, conversions, revenue and revenue per session. A referral decline can have a different business impact when conversion efficiency remains stable.
- Preserve a baseline before advertising availability changes in a target market. Weekly trend lines are more useful than a single before-and-after snapshot.
- Evaluate paid campaigns against a defined economics threshold, rather than assuming search-ad CPC expectations transfer directly to AI ads.
- Review attribution assumptions. Last-click reporting is useful, but it may understate assist value or obscure journeys that include multiple discovery surfaces.
For GEO and answer engine optimization work, the practical priority remains earning eligible, useful organic discovery. Clear content, accessible pages, accurate entity information and pages that directly satisfy relevant questions help create the conditions for an AI system to reference a site. But teams should measure that visibility separately from advertising performance. A paid placement can produce a visit without demonstrating organic citation strength, while an organic referral may reflect visibility that ads do not capture.
The study also reports a highly concentrated paid-advertising picture. One advertiser accounted for about 66% of paid ChatGPT revenue in the sample. Among 87 advertisers that used both ChatGPT ads and paid search, 73 recorded zero sales from ChatGPT ads across 18,726 visits. This concentration makes broad averages especially important to interpret carefully: a small number of accounts can strongly influence early-channel results.
If AI referrals are becoming an important part of your acquisition mix, create a GEO Search Leads account to assess your website’s GEO readiness and identify content and technical optimization opportunities. That gives your team a structured starting point for improving the foundations of unpaid AI discovery while you keep paid AI traffic, organic referrals and conversion outcomes separate in reporting.
Frequently Asked Questions
Did the study find that ChatGPT ads do not work?
No. The study found lower reported revenue per U.S. paid ChatGPT session and a lower conversion rate than paid Google Search in its dataset. It does not establish that every advertiser or campaign will perform the same way.
How large was the reported revenue-per-session gap?
The paper reported $0.89 in revenue per paid ChatGPT session and $2.45 per paid Google Search session in the U.S. data it analyzed.
Did organic ChatGPT referrals stop converting after the ad rollout?
No. Organic ChatGPT referrals declined by an estimated 5.9% over the first 13 weeks, but their conversion rate was essentially unchanged, moving -0.02 percentage points from a 2.21% baseline.
What are the main limitations of the research?
The analysis uses outside-the-platform Grips Intelligence data and last-click attribution, not prompt-level ChatGPT data. Its results cover a specific panel, markets and time period, so they should not be generalized to all websites or all future ad rollouts.
Conclusion
The early evidence suggests that AI advertising and unpaid AI discovery need separate commercial and visibility measurements. For the websites in this study, paid ChatGPT visits were less valuable per session than paid Google Search, while unpaid ChatGPT traffic declined without a meaningful conversion-rate shift. As AI ad availability expands, teams should watch referral volume, conversion efficiency and attribution together rather than treating any AI-originated click as equivalent.
Top comments (0)