In a non-peer-reviewed paper released in October 2026, European researchers compared sales performance between ChatGPT ads and Google search ads. For U.S. visits, revenue per visit was $0.89 for ChatGPT ads versus $2.45 for Google search ads, a gap of about 64%. Average order value was higher for ChatGPT ads, but a lower purchase rate left revenue per visit well below that of Google search ads.
OpenAI began introducing ads to ChatGPT in the U.S. in February 2026 and expanded them to Japan and South Korea in June. But advertisers deciding where to spend need to consider ad costs as well as revenue. Combining the researchers' analysis with results published by another ad management firm also reveals the revenue level ChatGPT ads would need to match the cost-effectiveness of Google search ads.
ChatGPT ad purchase rate is 0.37%, versus 1.17% for Google search ads
The study was conducted by Maximilian Kaiser of the Vienna University of Economics and Business and Christian Schulz of the Frankfurt School of Finance & Management. It analyzed 387 sites that received about 450,000 visits from ChatGPT ads over a 60-week period from June 29, 2025, to August 22, 2026. Data from 102 of those sites, where revenue could be fully tracked, was used to compare visits and purchases from ads.
For U.S. visits, the purchase rate was 0.37% for ChatGPT ads and 1.17% for Google search ads. Average order value, on the other hand, was higher for ChatGPT ads at $240, versus $209 for Google search ads. However, revenue per visit, which is the purchase rate multiplied by average order value, came to $0.89 for ChatGPT ads and $2.45 for Google search ads. Although ChatGPT ads won on value per order, a smaller share of visits led to purchases, so revenue per visit was about 64% lower.
The average order value comparison needs some caution, however. When the researchers used data from all markets and statistically adjusted for differences across sites, devices, and weeks, they estimated that average order value from ChatGPT ads was $74 lower than from Google search ads. That differs from the simple U.S. figures of $240 versus $209, because both the regions covered and the weighting used in aggregation are different. If orders are concentrated at a few sites selling high-priced items, the simple overall average and the site-adjusted estimate can point in opposite directions.
Among 87 U.S. sites that received visits from both types of ads, 97% had lower revenue per visit from ChatGPT ads. In addition, 84% of those 87 sites recorded no purchases at all from ChatGPT ads, and the 570 purchases observed were concentrated in the remaining 14 sites. Not just the overall average but also site-level results showed a large skew in sales.
That said, the study observes the results of placements that advertisers chose for themselves and does not directly measure the incremental sales caused by ad exposure. The analysis records the traffic source immediately before a site visit and the purchases made during that visit. It therefore cannot distinguish cases where someone saw an ad and later bought through another channel, or cases where someone would have bought without the ad.
The research data and computing resources were provided free of charge by Grips Intelligence, a web traffic analytics company. Kaiser is a former employee of the company and currently serves as a scientific advisor. According to the authors, however, the company was not involved in the study design, analysis, or interpretation of results, and had no authority to approve publication of the paper.
To match Google search ads' cost-effectiveness, ChatGPT ads need a cost per click of about $0.26
To compare the cost-effectiveness of ads, one has to consider cost per click in addition to revenue per visit. In this data, ad spending was recorded at the account level and not broken down by visitor country. The researchers therefore started from 31 sites where Google Ads spending records could be linked, and extracted 16 accounts that met the conditions for search ad spending and share of U.S. visits.
Dividing each account's ad spending over the period by its number of search ad clicks gave a median cost per click of $0.72. Assuming one click equals one visit, the researchers took the $2.45 revenue per visit observed for Google search ads in the U.S. as the benchmark, and calculated the cost per click at which ChatGPT ads would achieve equivalent return on ad spend.
The calculation is as follows.
$0.72 × ($0.89 ÷ $2.45) ≈ $0.26
In other words, if the observed revenue per visit stays the same, a ChatGPT ad cost per click of about $0.26 would yield revenue per dollar of ad spend comparable to Google search ads.
The researchers also assessed the uncertainty of this estimate by resampling site-level data. With the Google search ad cost per click fixed at $0.72, the central 95% of estimates for the equivalent ChatGPT ad cost per click ranged from $0.05 to $0.57. Details are given in Appendices J and K of the paper. This comparison treats clicks and visits as roughly the same thing, and assumes that the profit margin on sales from both types of ads is similar.
Meanwhile, in a case study published June 15, 2026, the ad management firm Opascope reported that a ChatGPT ad campaign it had chosen to scale up averaged a cost per click of $1.72 over 15 days. Ad spending was about $60,000 and measured revenue was about $89,000, for revenue of $1.49 per dollar of ad spend and a purchase rate of 2.35%. However, this is the result of a campaign scaled up in a specific account based on its performance, and does not represent the average for ChatGPT ads as a whole.
Fixing $1.72 as the ChatGPT ad cost per click and working backward to the revenue needed to match the revenue per ad dollar of Google search ads gives the following.
$1.72 × ($2.45 ÷ $0.72) ≈ $5.85
For ChatGPT ads to match the cost-effectiveness of Google search ads at a cost per click of $1.72, they would need about $5.85 in revenue per visit. That is about 6.6 times the $0.89 observed in the study.
This $5.85, however, is an estimate combining revenue observed in the study with a cost per click reported by a different ad management firm. Because it uses figures from different advertisers and campaigns, it does not indicate the actual break-even point for ChatGPT ads overall. The data underlying the calculation come from Table J.1 in Appendix J, Appendix L, and Opascope's published case study.
Did the introduction of ads reduce organic traffic from ChatGPT?
In addition to ad-driven sales, the researchers analyzed whether the introduction of ads in ChatGPT affected organic traffic, meaning visits to external sites from links within answers. This analysis covered 1,898 sites, separate from the ad sales comparison, and observed about 10 billion visits from all sources over 60 weeks. Of these, about 22.4 million were organic visits generated by clicking links in ChatGPT answers.
ChatGPT ads began in the U.S. on February 9, 2026, and were introduced in Japan and South Korea on June 19. In the EU, no ads were introduced during the observation period. The researchers used this difference in timing to compare changes in organic traffic between regions with ads and regions without.
The analysis also used visits from Google organic search to the same sites as a comparison. This is because if a site itself becomes more popular, visits from both Google search and ChatGPT could rise. The researchers compared the movement of the two channels week by week for each site, visitor country, and device, and, using the ad-free EU as the baseline, estimated how much organic traffic from ChatGPT changed in regions where ads were introduced. They also took care not to treat regions that already had ads as controls for other regions' pre-launch periods.
However, the later ads were introduced in a region, the shorter its post-introduction observation period, so simple averaging changes the weight of each region before and after introduction. Table F.1 in Appendix F, which examines this effect, shows estimates with altered regional weighting using the same data for up to 13 weeks after introduction.
| Estimation method | Change in ChatGPT organic traffic | 95% confidence interval |
|---|---|---|
| Original estimate | −5.9% | −12.6% to +1.2% |
| Regional weights equalized before and after introduction | −3.9% | −10.4% to +3.0% |
When regional weights are aligned, the estimated decline in organic traffic narrows by 2.0 percentage points, from 5.9% to 3.9%. However, because the 95% confidence interval includes zero in both estimates, it cannot be stated statistically that the introduction of ads reduced organic traffic. Nor can the table alone show whether the 2.0-point difference is itself statistically significant.
The analysis assumes that, had ads not been introduced, ChatGPT organic traffic relative to Google organic search would have moved similarly in the regions with ads and in the EU. If ChatGPT usage itself grew faster in the ad regions than in the EU, the decline in organic traffic caused by ads could be underestimated.
The researchers therefore also ran calculations under conditions where usage in ad regions grows 6% or 12% faster per year, drawing on differences in ChatGPT usage growth between North America and Europe. The estimated changes in organic traffic were −7.2% and −8.4%, respectively. In other words, results when evaluating the impact of ad introduction also depend on what one assumes about the pace of usage growth in each region.
Limited to the nine weeks after ad introduction in Japan and South Korea, the estimated change in organic traffic was +13.6%. However, the 95% confidence interval was wide, from −6.1% to +37.5%, so this is not a result that establishes an increase. Across all regions with ads, the gap between the purchase rate of ChatGPT organic traffic and that of Google organic search was estimated to have fallen by 0.024 percentage points after introduction. ChatGPT's pre-introduction purchase rate was 2.21%, and this change was not statistically clear either.
The analysis therefore does not support concluding that the introduction of ads clearly harmed ChatGPT organic traffic or purchase rates. What the researchers could capture were changes in visits and purchase rates in regions with ads; they did not track whether individual users saw ads, how they received the answers, or whether their trust in ChatGPT changed.
Funding free usage with ChatGPT ads will require better profitability
On August 31, 2026, OpenAI announced that ChatGPT ad revenue had reached an annualized pace of $1 billion and that ChatGPT's weekly users had topped 1 billion. The researchers also estimated how much of the inference cost of free users this ad revenue could cover.
The estimate assumes 1 billion weekly users, 95% of whom are free users. Dividing the annualized ad revenue of $1 billion by roughly 950 million free users gives about $1 per user per year. However, this is a figure obtained by annualizing the ad revenue pace at a particular point in time and dividing it by an assumed number of free users, and does not represent actual annual revenue.
In Appendix P, the researchers assumed, based on external reports, an annual inference cost of about $5 per free user. They then calculated whether showing more ads could make up the shortfall, under the following conditions.
- Each free user asks 511 questions per year that could lead to a purchase
- One ad slot is shown per eligible question
- The ad click-through rate is 0.50%
- The cost per click is $0.262, the equivalent level derived in the study
Under these conditions, annual ad revenue per free user works out as follows.
511 × 0.50% × $0.262 ≈ $0.67
To cover the assumed $5 annual inference cost with ad revenue alone, and with click-through rate and price unchanged, an average of about 7.5 ad impressions would be needed per question that could lead to a purchase. However, increasing ad slots does not necessarily preserve the click-through rate, and this figure is a simplified estimate for thinking about the revenue structure.
OpenAI is also exploring new formats and measurement to improve ad profitability. On October 5, 2026, it announced a test of showing ads during image generation. It also said it is exploring region-level experiments with measurement partners to gauge how much ads increase sales and other outcomes. On October 21, it explained that its system for selecting ads based on context drawn from multiple turns of a conversation is separate from the processing that generates answers.
If the places where ads appear and the targeting mechanisms change going forward, click-through rates and purchase rates could change as well. What matters to advertisers, however, is not just purchases immediately after a click but understanding how much ads actually lifted sales, including cases where people bought through another channel after seeing an ad.
The study shows that early ChatGPT ads trailed Google search ads in revenue per visit, but it cannot determine whether the gap is due to the ads themselves or to differences among advertisers, products, and users. If region-level comparative experiments and similar methods measure the pure sales lift from ads, advertisers will be able to judge cost-effectiveness more accurately. For OpenAI, comparing ad revenue per user with inference cost will be an important input for gauging how far advertising can support the operation of the free version of ChatGPT.