TL;DR
- Kaiser (WU Vienna) and Schulze (Frankfurt School) used 1,898 online shops and the staggered ChatGPT ads rollout, with the EU as untreated control, in a working paper dated October 2026.
- Paid ChatGPT visitors spent $74 less per order, and 84% of the 87 advertisers buying both paid channels in US traffic recorded no paid ChatGPT sale despite 18,726 paid sessions.
- At paid search's rate of return a ChatGPT click is worth about $0.26; the largest public account paid $1.72, more than six times that.
- After ads arrived, organic ChatGPT referrals fell about 6 to 8% while their conversion rate held at about 2.2%.
- Caveats: listed as under review, revenue figures are US traffic, and co-author Kaiser is scientific advisor to data supplier Grips Intelligence.
A working paper by Maximilian Kaiser of the Vienna University of Economics and Business and Christian Schulze of the Frankfurt School of Finance & Management, version 1 dated October 2026, reports that in US traffic a paid ChatGPT ad visit generated $0.89 in revenue per session, against $2.45 for a paid Google Search visit to the same kind of online shops. Paid ChatGPT visitors converted at 0.37%, paid Google Search visitors at 1.17%. The paper, "Can Advertising Fund Free AI? Evidence from ChatGPT's International Advertising Rollout", draws on data from 1,898 online shops that together sell $16 billion a year. It is a working paper listed as under review, its revenue comparison rests on US traffic, and one author advises the company that supplied the data. All three points matter when you quote it.
What the ChatGPT ads study found for advertisers
The headline comparison sits in the section the authors call Condition B, which asks whether ChatGPT ads are worth their price to the businesses buying them. The authors benchmark paid ChatGPT against paid Google Search, which they describe as the channel these same websites already buy for similar commercial intent.
In US traffic, where the paper says ad delivery concentrated, a paid ChatGPT visit generated $0.89 in revenue per session and a paid Google Search visit $2.45. The authors call that a 64% shortfall. Two things drive it, according to the paper. Paid ChatGPT visitors bought less often (0.37% conversion against 1.17%), and when they did buy, their orders were smaller, by $74, after adjusting for website, device and week.
The gap is not driven by a handful of accounts, the paper says. Among the 87 advertisers that bought both paid channels in US traffic, 97% earned less per ChatGPT visit and 98% converted worse. Most of those 87 websites made no sale from ChatGPT ads at all: 73 of them (84%) recorded no paid ChatGPT transaction despite 18,726 paid ChatGPT sessions, and the remaining 14 carried all 570 transactions.
The paper also places paid ChatGPT among the shops' other channels, each measured against direct traffic. It reports that paid ChatGPT is the only channel that falls short of direct traffic on all six metrics it tracks: conversion rate, average order value, revenue per session, bounce rate, pageviews and time on site. Paid ChatGPT visitors stayed 104 seconds less and viewed 1.4 fewer pages than direct visitors. Google's paid search, by contrast, did not differ detectably from direct traffic on conversion, order value or revenue per session.
How the study was built
The research uses OpenAI's country-by-country ad rollout as a natural experiment. According to the paper, ChatGPT ads launched in the United States on 9 February 2026; in Australia, Canada and New Zealand on 16 April; in the United Kingdom on 4 June; in Japan and Korea on 19 June; and in Brazil and Mexico on 4 and 6 August. The European Union stayed without ads for the whole observation window, which ran from 29 June 2025 to 22 August 2026, so it serves as the untreated comparison group. The paper notes that EU ads went live on 24 August, after the window closed.
The data comes from Grips Intelligence and covers nearly 10 billion visitor sessions over 60 weeks. There are two panels. The organic panel of 1,898 websites is used to measure what happened to free ChatGPT referrals. The paid panel holds 387 websites that received visits from paid ChatGPT ads; 102 of them have complete financial tracking, and 31 of those have linked Google Ads cost records. The data captures the last click that brought a visitor to a site. It does not show prompts, ad impressions or which subscription tier a user was on.
The table below lists the main figures as the paper and its companion page state them.
| Measure (paper's own figures) | Value |
|---|---|
| Revenue per session, paid ChatGPT visit (US traffic) | $0.89 |
| Revenue per session, paid Google Search visit (US traffic) | $2.45 |
| Conversion rate, paid ChatGPT vs paid Google Search | 0.37% vs 1.17% |
| Parity price of a ChatGPT click at paid search's rate of return | about $0.26 |
| Price paid by the largest publicly known account | $1.72 per click |
| Organic ChatGPT referral volume after ads (first 13 weeks) | -5.9% (95% interval -12.6% to +1.2%) |
| Organic ChatGPT conversion change | -0.02 percentage points from 2.21% |
| Samples | 1,898 shops (organic), 387 (paid), 102 (financial), 31 (cost) |
The parity price: what a ChatGPT click is worth
The most practical number in the paper for anyone setting bids is the parity price, the price at which a ChatGPT click would return as much per dollar as a search click. The cost sample paid a median of $0.72 for a paid-search click. Because a ChatGPT visit returned $0.89 against $2.45, the authors calculate that a ChatGPT click is worth about $0.26 at the same rate of return. When they resample advertisers, the range runs from $0.05 to $0.57.
The paper compares that with the largest publicly known ChatGPT ads account, which paid $1.72 per click, more than six times the parity price. It adds that this figure sits toward the low end of what practitioners report and below OpenAI's recommended starting bid of $3 to $5 per click, which the paper cites. In the authors' words, the paid ChatGPT channel offers "less than a sixth of the value for money of paid Google Search."
Two assumptions sit under the parity price, and the paper states both. It assumes similar profit margins in the two channels, and it counts only the tracked visit. A ChatGPT ad could still push people to buy later through another channel. The authors tested this with matched comparisons around each advertiser's first paid week and found no detectable lift in the channels the campaigns did not buy, while noting that this check rules out only large indirect effects.
What happened to organic ChatGPT referrals
The other half of the paper, Condition A, asks whether ads hurt ChatGPT's value to users. The authors measure that from the outside, through the free ChatGPT referrals that reach online shops.
Over the first 13 weeks after launch, organic ChatGPT referral volume changed by -5.9%, with a 95% interval from -12.6% to +1.2%. The authors say the interval and a placebo test both point to a decline but neither establishes it with confidence. Allowing for ChatGPT's faster usage growth in North America than in Europe moves the estimate to between -7.2% and -8.4%. The companion page summarises this as organic referrals falling by about 6 to 8%.
Conversion held. Organic ChatGPT conversion changed by -0.02 percentage points from a 2.21% pre-launch baseline (p = .648). The paper also reports that, before the rollout, organic ChatGPT's conversion rate roughly matched organic search's. For brands that care about appearing in ChatGPT answers, that is the part of the paper that bears on AI search: the free recommendations kept converting at the same rate after ads arrived, even if somewhat fewer visits came through.
Caveats to read before quoting the numbers
It is a working paper, not a published journal article
The companion page lists the status as "Working paper · under review", version 1, October 2026. It has not yet been published by a journal, so its numbers may change in later versions. The authors say they will update the study as new data comes in.
The revenue comparison is US traffic
The $0.89 versus $2.45 figures are stated for US traffic, where the paper says ad delivery concentrated. The organic panel is international, but the advertiser value comparison should not be read as a global average, and nothing in the paper covers Thailand.
One author advises the data provider
The paper's competing-interests section states that Maximilian Kaiser was previously employed by Grips Intelligence and now serves as its scientific advisor, and that, beyond his salary and virtual company shares, the authors received no payment, equity or other consideration from Grips. Grips provided the data free of charge. The paper says the agreement gives Grips no right to review, approve or delay publication, that Grips has no commercial relationship with OpenAI, and that Christian Schulze declares no competing interests.
The advertiser comparison is descriptive
The authors say advertiser entry was self-chosen, not randomised, so the paid-channel evidence is descriptive rather than causal. They also list three limits: campaigns that never produced a click are invisible, first-click dates can lag true campaign starts, and the cross-channel checks rule out only large indirect effects. Paid ChatGPT traffic is also concentrated: one advertiser carries 66% of all paid ChatGPT purchases. The paper reports that dropping it deepens both deficits.
What the paper does not say
The paper does not measure ad impressions or click-through rates inside ChatGPT; it sees only the visits that reached shops. It does not compare ad formats, since the sponsored cards it observed sit below the answer, separate from the organic response. It does not say whether ChatGPT ads are available in Thailand, and it says nothing about future pricing from OpenAI. Its financing section, which estimates how many ads per commercial prompt would be needed to cover free users' costs, is described by the authors as illustrative arithmetic, not a prediction.
What to check before you test ChatGPT ads
The following is analysis built on the paper's figures, not findings of the paper itself.
- Know your own parity price. Take your paid search revenue per click and your average CPC, then work out what a ChatGPT click would need to return to match it. The paper's $0.26 comes from its US-focused samples and will differ from your own figure.
- Separate paid and organic ChatGPT traffic in analytics. The study could only compare channels because referral sources were tagged. If ChatGPT ad clicks land in the same bucket as organic ChatGPT referrals, you cannot judge either one.
- Judge on revenue per visit, not clicks. The paper's gap appears in conversion and order value at the same time, so click volume alone would hide it.
- Keep Google Search as the benchmark. The authors chose paid search as the comparison because it is what the same shops already buy for similar intent. Any ChatGPT test budget is easier to judge against that baseline.
- Give a test a defined stop rule. With 84% of the paper's 87 dual-channel US advertisers recording no paid ChatGPT sale, decide in advance what revenue per session would justify continuing.
If you run Search campaigns, our Google Ads management team can help you set that benchmark from your own account data, and our ChatGPT Ads service page explains how a structured test can be set up when the format reaches your market.
What this means for Thai marketers
The study does not include Thailand, and the paper does not state whether or when ChatGPT ads will run here. For Thai advertisers the value is a benchmark: when the format becomes available, the first question is how its revenue per visit compares with what Google Search already delivers in the same account. The organic finding also matters for brands investing in visibility inside ChatGPT answers, since organic ChatGPT referrals kept converting at the same rate after ads arrived in the markets studied. Work on generative engine optimisation targets exactly that organic channel.
FAQ: ChatGPT ads vs Google Search ads
How much less do ChatGPT ad visits earn than Google Search ad visits?
In US traffic, a paid ChatGPT visit generated $0.89 in revenue per session against $2.45 for a paid Google Search visit, a 64% shortfall, according to the Kaiser and Schulze working paper. Conversion was 0.37% against 1.17%.
Is the ChatGPT ads study peer-reviewed?
No, not yet: it is a working paper, version 1 from October 2026, listed on its companion page as under review. The figures could change in a later version.
Who supplied the data, and is there a conflict of interest?
Grips Intelligence supplied the data free of charge, and co-author Maximilian Kaiser serves as its scientific advisor, a tie the paper discloses. The paper states Grips had no role in the research design or writing and no right to review publication.
Are ChatGPT ads available in Thailand?
That is not stated by the paper, which covers the US, Australia, Canada, New Zealand, the UK, Japan, Korea, Brazil and Mexico, with the EU as the untreated control. Check OpenAI's own advertiser pages for current country availability.
What is a fair price for a ChatGPT ad click, according to the study?
About $0.26, the paper's parity price at paid search's rate of return, with a resampled range of $0.05 to $0.57. The largest publicly known account paid $1.72, which the authors put at more than six times that level.
The full paper, methods appendix and fact sheet are on the authors' companion page, ecommerceresearch.org. If you want to compare a ChatGPT ads test against your current Google Search returns, Relevant Audience can help you set up the measurement before you spend.
