A working paper from Maximilian Kaiser and Christian Schulze, dated October 2026 and reported on 2 November, gives paid teams the first outside measure of ChatGPT ads. In US traffic a paid ChatGPT visit earned $0.89 against $2.45 for a paid Google Search visit, so a ChatGPT click is worth about $0.26 at search’s rate of return.
Key takeaways
- Kaiser and Schulze found a paid ChatGPT visit generated $0.89 in revenue in US traffic, against $2.45 for a paid Google Search visit, a 64% shortfall.
- Paid ChatGPT visitors converted at 0.37% against 1.17% for paid search, and their orders were $74 smaller after adjustment.
- Of 87 advertisers buying both channels, 73 recorded no paid ChatGPT sale from 18,726 paid ChatGPT sessions.
- The paper puts the parity price of a ChatGPT click at about $0.26, while the largest publicly known account paid $1.72.
- Organic ChatGPT referrals fell by an estimated 6% to 8% after ads launched in a market, and their conversion rate held steady.
Who measured ChatGPT ads, and how?
Kaiser is at the Vienna University of Economics and Business and Grips Intelligence, and Schulze is at the Frankfurt School of Finance and Management. Their paper, Can Advertising Fund Free AI?, is a working paper under review. Grips Intelligence supplied Google Analytics ecommerce data free of charge, and the authors say it had no role in the research design or writing.
The data covers 60 weeks, from 29 June 2025 to 22 August 2026, and nearly 10 billion sessions on 1,898 online shops that sell $16 billion a year. The authors used OpenAI’s staggered rollout as a natural experiment. Ads went live in the US on 9 February 2026, in the UK on 4 June and in the EU on 24 August, after the window closed, so Europe served as the comparison group. The paid analysis uses 387 advertiser websites, of which 102 have full sales tracking and 31 have linked Google Ads cost records.
Search Engine Journal covered the paper on 2 November.
What a ChatGPT click returned
The headline numbers come from US traffic, where ads ran longest. The full paper compares each channel against direct traffic on the same sites.
| Measure, US traffic | Paid ChatGPT | Paid Google Search |
|---|---|---|
| Revenue per visit | $0.89 | $2.45 |
| Conversion rate | 0.37% | 1.17% |
| Price per click in the study | $1.72, largest public account | $0.72, median of 16 US accounts |
| Price that matches search’s return | about $0.26 | $0.72 |
| Gap to direct traffic | Short on all six measures tested | No detectable gap on conversion, order value or revenue |
The authors checked that one big buyer wasn’t skewing the result. The largest advertiser carried 66% of paid ChatGPT purchases, and dropping it made the gap wider. Among the 87 advertisers who bought both channels, 97% earned less per ChatGPT visit. Most sites made no sale at all, and 14 websites carried all 570 paid ChatGPT transactions.
The paper also compared each advertiser’s sales in the channels they didn’t buy, before and after their first paid ChatGPT week, and found no detectable lift there. That cuts against the claim that ChatGPT ads pay back somewhere the last click can’t see, though the authors say their test rules out only large indirect effects.
What does this mean for a paid budget?
For a performance team, ChatGPT ads are a test channel that has to earn a bid. If your paid search clicks cost £0.90 and return £3, the paper’s ratio suggests a ChatGPT click returns about £1.09, so paying more than about £0.33 loses money against search on the tracked sale. The paper cites OpenAI’s recommended starting bid as $3 to $5 per click. Your own ratio will differ, so work it out before the next budget meeting.
OpenAI added a scale figure this week. Its ChatGPT consumer product lead Adam Fry told a New York dinner, as Big Technology reported on 3 November, that ChatGPT’s volume of search-like queries is roughly one fifth of Google’s, and that weekly users rose from 1 billion to 1.2 billion in two months. Audience size makes ChatGPT hard to ignore, but it doesn’t change what each click returns, so it shouldn’t set the bid.
Our view at Manual Focus is that UK advertisers should cap ChatGPT bids at their own parity price and hold spend at test level until a holdout on their own account shows incremental sales, because the first outside data shows ChatGPT clicks returning about a third of a search click. We said the same on OpenAI’s measurement update in October, and this paper is the first outside evidence for it.
Brand and SEO teams have a smaller signal to watch. Organic ChatGPT referrals fell by an estimated 6% to 8% once ads arrived, and visitors who still came converted at 2.21% before and almost exactly the same after. Organic ChatGPT already converts about as well as organic search, according to the paper, so being cited in the answer remains worth more than paying for the card beneath it.
Four moves before Black Friday
- Calculate your parity price. Take your median paid search cost per click and multiply it by your ChatGPT revenue per visit divided by your paid search revenue per visit. Set ChatGPT bid caps at or below that figure.
- Split ChatGPT out of blended reports. Make paid ChatGPT its own channel in GA4 and in any attribution tool, so a small number of sessions can’t hide inside a healthy total. The attribution tear-down playbook sets out how.
- Run a holdout before scaling. Pause ChatGPT ads in a set of regions for four weeks and compare total sales against regions that kept running. The paper found no lift in channels advertisers didn’t buy, so a short pause should cost little. Agree the threshold with finance first.
- Check the creative setting. OpenAI switched on creative enhancements by default at the end of October, so a weak test may also be a rewritten ad you never approved.
If you sell through Shopify, our post on ChatGPT Ads for UK ecommerce covers the setup, and the paid search bidding agent playbook shows how to run margin-aware bid caps across channels.
Where the study runs out
The data stops on 22 August 2026, before OpenAI added ten attribution partners in October, before the visual ad test and before creative enhancements became the default, so performance may have changed since. The paper counts the last click only and can’t see prompts, impressions or which tier a user was on, and ads appear only on the free and Go tiers. Advertisers chose to buy ChatGPT ads, so the results describe who bought rather than proving cause. The UK had under three months of ads in the window, and the headline figures are from US traffic. The $1.72 price comes from one public account, and the parity price assumes similar margins in both channels. The paper is also still under review, and peer review may change the numbers.
Further reading
- What ChatGPT’s new ad measurement tools mean for paid teams, on the attribution partners and geo-lift work OpenAI announced after this data ends
- What ChatGPT Ads’ automatic rewrites mean for UK advertisers, on the default setting to check before any test
- What Google’s AI Mode ads mean for UK paid search, on the other AI ad placement arriving in the UK this quarter
- Attribution tear-down using GA4 and incrementality signals, for testing what a new channel adds
If you want help setting bid caps and a holdout for ChatGPT ads, our services page explains how we work.
Frequently asked questions
Do ChatGPT ads perform as well as Google Search ads?
Not on current evidence. A working paper by Maximilian Kaiser and Christian Schulze, dated October 2026, found that in US traffic a paid ChatGPT visit generated $0.89 in revenue against $2.45 for a paid Google Search visit, a 64% shortfall. Paid ChatGPT visitors converted at 0.37% against 1.17% for paid search, and their orders were smaller.
What should I bid on ChatGPT ads?
The Kaiser and Schulze working paper puts the parity price at about $0.26 per click, the price at which a ChatGPT click returns as much per dollar as a Google Search click for the advertisers it studied. Resampling gave a range of $0.05 to $0.57. The largest publicly known account paid $1.72, more than six times that figure. Work out your own parity price from your own search data.
Did ads make ChatGPT send less organic traffic to websites?
Probably a little. The October 2026 Kaiser and Schulze paper estimated a 5.9% fall in organic ChatGPT referrals over the first 13 weeks after ads launched in a market, or about 6% to 8% once faster ChatGPT growth in North America is allowed for. Visitors who still arrived from organic ChatGPT answers converted at the same rate as before.
How big is ChatGPT search compared with Google?
OpenAI's ChatGPT consumer product lead Adam Fry said at a New York dinner reported by Big Technology on 3 November 2026 that ChatGPT's volume of search-like queries is now roughly one fifth of Google's. He also said ChatGPT went from 1 billion to 1.2 billion weekly users in the last two months. OpenAI didn't say which Google figure it compared against.
Sources
- Can Advertising Fund Free AI? Evidence from ChatGPT's International Advertising Rollout, Maximilian Kaiser and Christian Schulze, working paper
- Can Advertising Fund Free AI? (full paper, PDF), Maximilian Kaiser and Christian Schulze
- Google Search Ad Visits Out-Earned ChatGPT Ads In U.S., Search Engine Journal
- OpenAI says ChatGPT Search is about 20% of Google's, and chat is merging with work, Big Technology
From the podcast
James and Hannah go through the first independent study of what ChatGPT ads return. Maximilian Kaiser and Christian Schulze’s working paper, dated October 2026, finds that in US traffic a paid ChatGPT visit earned $0.89 against $2.45 for a paid Google Search visit, which puts a ChatGPT click’s parity price at about $0.26. They cover how the study used OpenAI’s country-by-country rollout, why the largest advertiser makes the result look better rather than worse, what it means for bids and budgets, why organic ChatGPT referrals still matter, and three jobs to hand to AI before Black Friday.
Get AI on this
Paste these into whichever AI assistant you use. Swap the bracketed parts for your own details. The first and third need file uploads. The second needs an assistant that can read files you share each week and run on a schedule.
1. Work out your ChatGPT parity price
I run paid media for [brand], [one line on what you sell] at [website],
selling mainly in [market].
A working paper by Kaiser and Schulze (October 2026) defines the parity price
of a ChatGPT ads click as the price at which it returns as much revenue per
pound spent as a paid search click. The formula is: parity price = paid
search cost per click x (ChatGPT revenue per visit / paid search revenue per
visit).
Attached are:
- a paid search export by campaign for the last [8] weeks, with clicks, cost
and conversion value
- a GA4 report for the same period with sessions and revenue for paid search
and for paid ChatGPT traffic, broken down by landing page or campaign where
possible
Work out revenue per visit for each channel, then the parity price overall
and for each campaign or product line where there is enough data.
Give me a table with the columns: Campaign or product line | Paid search
cost per click | Paid search revenue per visit | ChatGPT revenue per visit |
Parity price for a ChatGPT click | ChatGPT sessions in the period | Enough
data to trust? (yes or no, with the reason).
Below the table, give me one recommended bid ceiling for ChatGPT ads and say
how you got it. Flag any figure you had to estimate, any channel you couldn't
find in the data, and anything that looks mislabelled. Don't guess missing
numbers.2. Keep paid ChatGPT out of the blended total
Each [Monday] I'll share our GA4 channel report and our ChatGPT ads spend for
the previous week for [brand] at [website]. Run this as a weekly routine.
1. Pull paid ChatGPT traffic out as its own line, separate from organic
ChatGPT referrals and from other paid channels. If you can't tell paid
from organic ChatGPT traffic in the data, say so and stop.
2. Report sessions, purchases, revenue, spend and cost per purchase for paid
ChatGPT, with the same figures for paid search alongside.
3. Compare this week's actual cost per click with our parity price of
[parity price from job 1].
Output: a short table with the columns Channel | Sessions | Purchases |
Revenue | Spend | Cost per purchase, then three bullet points on what moved
since last week.
Add a clear warning at the top if ChatGPT spend rose while purchases stayed
at zero, or if the cost per click is above the parity price. Say which
numbers you couldn't check.3. Draft a holdout test for ChatGPT ads
We buy ChatGPT ads for [brand] in [market] and want to know whether they add
sales we wouldn't get anyway, before we spend more.
Attached are:
- weekly sales by region for the last 6 months
- our ChatGPT ads spend by region for the same period, if we have it
Propose a four-week geo holdout. Pick a set of regions where we pause
ChatGPT ads and a matched set where they keep running, matched on past sales
level and trend.
Give me:
- a table with the columns Region | Group (pause or keep) | Average weekly
sales | Trend over 6 months
- the success threshold you'd suggest (the difference in total sales between
the groups that would justify keeping or scaling spend), with your reasoning
- the risks to the test, such as seasonality, promotions or too few regions
Don't make the decision to pause. Write it as a proposal for me to agree with
finance. Tell me where the data is too thin for a clean test, and don't fill
gaps with assumptions.Read the episode transcript
James: What's a click in ChatGPT actually worth to the business paying for it? Until this week the only answers came from OpenAI and the case studies it picked. Now two academics have measured it from the outside. This is Manual Focus, I'm James Vickers, and Hannah's read the whole paper so I don't have to.
Hannah: Hi James. Appendices included.
James: Brave. Who wrote it, and what did they look at?
Hannah: Maximilian Kaiser, at the Vienna University of Economics and Business, and Christian Schulze, at the Frankfurt School of Finance and Management. It's a working paper dated October, still under review, and Search Engine Journal covered it on the second of November. The data comes from Grips Intelligence and covers around nineteen hundred online shops. Grips gave it free, and the authors say it had no say in the research.
James: And the clever part is how they used the rollout.
Hannah: Right. OpenAI switched ads on country by country. The US on the ninth of February, the UK on the fourth of June, and the EU only on the twenty-fourth of August, after their data window closed. So Europe becomes the comparison group, and they can see what changed when ads arrived somewhere.
James: So give me the number a head of paid search will care about.
Hannah: In US traffic, a paid ChatGPT visit brought in eighty-nine cents of revenue. A paid Google Search visit on the same kinds of sites brought in two dollars forty-five. That's a sixty-four per cent shortfall. Conversion rate was nought point three seven per cent for ChatGPT against one point one seven for paid search, and the orders were smaller too.
James: That's about a third of the value. Is one big advertiser dragging it around?
Hannah: They checked. The biggest one had sixty-six per cent of all the paid ChatGPT purchases, and taking it out made the gap wider, not narrower. Of eighty-seven advertisers buying both channels, seventy-three had no ChatGPT sale at all, from more than eighteen thousand paid visits.
James: Which brings us to price. What should a click cost?
Hannah: They work out what they call a parity price, the price at which a ChatGPT click pays back as well as a search click. Their advertisers paid a median of seventy-two cents for a search click, so a ChatGPT click is worth about twenty-six cents on the same terms. The largest public account they could find paid one dollar seventy-two. More than six times parity.
James: So here's what I'd do with that on Monday. Don't use their twenty-six cents, work out your own. Take what your search clicks cost, multiply by what a ChatGPT visit earns you divided by what a search visit earns you, and that's your ceiling.
Hannah: And on the post's example, if search costs ninety pence a click and returns three pounds, a ChatGPT click is worth about thirty-three pence. Anything above that is losing money against search on the tracked sale.
James: The obvious pushback is that last click misses things. Someone sees the ad, then buys through Google a week later.
Hannah: They tested that. They compared each advertiser's sales in the channels they didn't buy, before and after the first paid ChatGPT week, and found no detectable lift. The authors say that rules out only large indirect effects, not small ones.
James: When we covered OpenAI's measurement update in early October, our line was keep it at test budget until a holdout on your own account proves it. This is the first outside evidence for that.
Hannah: And OpenAI also gave a scale number this week. Adam Fry, who leads ChatGPT's consumer product, told a dinner in New York that ChatGPT's search-like queries are about a fifth of Google's. Big Technology reported it. He didn't say which Google figure he meant.
James: A fifth of Google is a lot of people. But a big audience doesn't change what each click returns, so it shouldn't set the bid. What about the brand and SEO side?
Hannah: Smaller effect. Organic referrals from ChatGPT, the free links in the answer, fell an estimated six to eight per cent once ads launched in a market. But the people who still clicked converted just as before, about two point two per cent. And organic ChatGPT now converts roughly as well as organic search.
James: So being in the answer is still worth more than paying for the card underneath it.
Hannah: On this data, yes.
James: Right, so what do you hand to AI? First one. Get AI on this, working out your parity price.
Hannah: Give it your paid search export and your analytics channel report, revenue and sessions for both channels. It works out revenue per visit for each and gives you a bid ceiling per campaign, flagging any where the sample's too small to trust.
James: Second, a weekly check so the new channel can't hide inside a blended total.
Hannah: Set it as a weekly routine. It reads the channel report, pulls those ads out on their own line, and gives you sessions, sales and cost per sale against paid search, with a warning if spend rises while sales stay at zero.
James: And third, the holdout plan.
Hannah: Give it your regional sales for the last six months and your ad spend by region. It proposes which regions to pause for four weeks and which to keep running, and drafts the success threshold for you to agree with finance. The decision to pause stays with a person. The prompts are in the show notes.
James: What can't this paper tell us?
Hannah: The data stops on the twenty-second of August. So it's before OpenAI's new attribution partners, before the visual ad test, and before creative enhancements went on by default. It's last click only, with no view of prompts or impressions. Advertisers chose to buy, so it describes them rather than proving cause. The UK had under three months in the window, and the headline numbers are American. And it's still going through peer review.
James: So the takeaway. The first independent look at ChatGPT ads says a click returns about a third of a search click, so cap your bids at your own parity price and prove the lift before you scale.
Hannah: The full article, with the table, is on the website.
James: Follow Manual Focus in Apple Podcasts, Spotify or wherever you listen and the next episode will turn up on its own. And if you'd like help setting bid caps and a holdout for your own account, book a free working session at manual-focus dot co dot uk. Thanks, Hannah.
Hannah: Thanks, James. Do the sum before the auction does it for you.