ChatGPT started showing ads in the United States on February 9, 2026. A new working paper by Maximilian Kaiser (WU Vienna) and Christian Schulze (Frankfurt School of Finance & Management) is the first large look at what those ads return. It uses first-party analytics from 1,898 online shops with about $16 billion in annual sales, over 60 weeks from June 2025 to August 2026.
The headline result: at today’s prices, ChatGPT ads return far less than Google search ads. But one table further into the paper matters more for anyone deciding where to spend. For the same advertisers, a visit from ChatGPT recommending them for free was worth more than three times a visit from a ChatGPT ad they paid for.
What the study measured
OpenAI rolled ads out country by country: the US on February 9, Australia, Canada and New Zealand on April 16, the UK on June 4, Japan and Korea on June 19, and Brazil and Mexico in early August. The EU got no ads until August 24, after the study window closed, so it works as the comparison group.
That setup let the authors ask two questions:
- Did ads make the free product worse? They measured this through organic ChatGPT referrals, the visitors who click an unpaid link in a ChatGPT answer, across all 1,898 shops.
- Are the ads worth what advertisers pay? For this they used 387 advertiser websites, 102 of which had full revenue tracking.
Ads appear only on ChatGPT’s Free and Go plans. People on Plus, Pro and business plans don’t see them.
Paid ChatGPT clicks against Google search ads
In US traffic, a paid ChatGPT visit generated $0.89 in revenue against $2.45 for a paid Google search visit, 64% less. Paid ChatGPT visitors converted at 0.37%, against 1.17% for paid search, and their orders were $74 smaller.
It wasn’t one bad advertiser dragging the average down. Among the 87 advertisers buying both channels, 97% earned less per ChatGPT visit. And 84% of advertisers recorded no sale from ChatGPT ads at all, across 18,726 paid ChatGPT visits.
The price makes it worse. At paid search’s rate of return, the authors put a ChatGPT click’s worth at about $0.26. The largest publicly known account paid $1.72 per click, and OpenAI recommends starting bids of $3 to $5. Put together, the paper says ChatGPT ads currently offer more than six times worse value for money than Google search ads.
Paid ChatGPT was also the only channel that fell short of direct traffic on all six measures the authors tracked: conversion, order value, revenue per visit, bounce rate, pages per visit and time on site.
The comparison that matters more: paid against free
The paper’s descriptive table puts both kinds of ChatGPT traffic side by side for the same 102 advertisers, over the whole study window:
| Same 102 shops, Jun 2025 to Aug 2026 | Free ChatGPT recommendation | Paid ChatGPT click |
|---|---|---|
| Conversion rate | 1.31% | 0.31% |
| Revenue per visit | $2.48 | $0.73 |
A visitor ChatGPT sent for free converted about four times as often and was worth more than three times as much as a visitor from an ad the same shop paid for.
These are raw averages, not the adjusted estimates from the paper’s models, and the two kinds of traffic come from a different mix of countries, since paid traffic was concentrated in the US. The adjusted results point the same way on engagement: organic ChatGPT visitors bounced less, viewed more pages and stayed longer than direct visitors, while paid ChatGPT visitors did worse than direct on every measure.
Three more findings round out the picture:
- After ads arrived, organic ChatGPT referrals fell by about 6 to 8%. The visitors who still came converted as before, at around 2.2%. So the ads cost ChatGPT some free clicks, but not their quality.
- Across the 1,898 shops, organic ChatGPT visits went from 0.23% of all visits over the whole window to 0.40% in the final four weeks. Still small, but close to doubling.
- In the authors’ earlier peer-reviewed study of 973 shops (Marketing Science, 2026), ChatGPT referrals converted below most established channels, but their conversion rate rose through the year while the other channels stayed flat. For complex products, they already out-converted five of eight channels.
Why the free answer wins
The authors think the ads are worth little because the sponsored cards sit apart from the answer, and the answer is what people came for. The person asked a question and got a recommendation. The ad is the thing next to it.
That recommendation isn’t bought. ChatGPT assembles it from the web. In our AI Search Index, which runs the same buyer questions for a set of tracked brands every day, ChatGPT searched the web on 98.5% of those questions over the last 90 days and cited about six sources per answer. Over the past week, around four in ten of its citations went to company websites. These are pooled rates across the questions we track, so they describe how ChatGPT sources answers, not how it treats one business.
So the free placement in this study comes from pages ChatGPT can find and read: your own pages, and the reviews, comparisons and articles it pulls in alongside them. We’ve written about what ChatGPT reads before it recommends you, and about how AI can use your content and still name a competitor.
What to do with your budget
- Don’t move proven paid search money into ChatGPT ads yet. If you test them, cap the spend, compare revenue per visit against your own paid search, and bid close to what a click is worth to you. In this data that was about $0.26, and your margins will differ.
- Know who you’re reaching. Ads only show to Free and Go users. If your buyers are professionals on paid ChatGPT plans, your ads never reach them.
- Work on being in the answer. Find the questions your buyers ask ChatGPT, see which brands it recommends and which pages it cites, and work on getting your brand onto those pages. Our guide to running a GEO audit shows how.
- Make sure ChatGPT can reach your site. If your robots.txt or firewall blocks OpenAI’s crawlers, your pages can’t make it into the answer at all. The robots.txt checker shows what you’re allowing.
- Track ChatGPT as its own channel. The authors treat it that way in both papers. Folded into “referral”, the free channel that outperformed the paid one in this study disappears from your reports. If you do buy ads, keep paid and earned ChatGPT visits apart in your reporting.
What the study can’t tell you
- It’s a working paper, version 1 from October 2026, still under review.
- Advertisers chose to advertise, so the paid results describe what happened rather than prove what caused it. They also cover only the first 28 weeks of advertising, while ad formats and prices are still changing. Self-serve buying opened in May 2026 and conversion-optimized campaigns in June.
- The data counts the click that brought a visitor, not what the person saw before. Purchases that happened later through another channel aren’t credited to the ad, though the authors found no detectable lift in the channels the campaigns didn’t buy.
- The data comes from Grips Intelligence, which provided it free of charge. Kaiser serves as its scientific advisor, and the authors state the provider had no role in the questions, analysis or writing.
The full paper, slides and a fact sheet are on the authors’ site.
