ChatGPT Ads should earn a bounded test, not an automatic transfer of Google Search budget. An October 2026 working paper reports substantially lower tracked revenue per paid ChatGPT visit in its early-rollout US data. That is a reason to demand credible unit economics and an incrementality plan. It is not proof that Google causes better outcomes, that conversational advertising cannot work, or that your account will reproduce the study average.
What the November 2 reporting actually establishes
On November 2, 2026, Search Engine Land reported that ChatGPT ads generated about 64% less revenue per visit than paid Google Search in a new study. The underlying evidence is Maximilian Kaiser and Christian Schulze's Can Advertising Fund Free AI? Evidence from ChatGPT's International Advertising Rollout, version 1, dated October 2026 and under review. Kaiser is affiliated with WU Vienna; Schulze with Frankfurt School of Finance & Management.
This is new study reporting, not a new ChatGPT advertising feature. Its observation window ends August 22, 2026. It does not evaluate every subsequent change in formats, bidding, targeting, or measurement. Our earlier analysis of ChatGPT visual formats and reasoned-intent measurement addresses platform capabilities; this paper asks what observed visits returned.
The headline sample needs particular care. Nearly 10 billion sessions means all-source sessions in the study's 1,898-site organic-referral panel over 60 weeks, beginning June 29, 2025. It does not mean 10 billion paid ChatGPT visits, or 1,898 advertisers buying ChatGPT ads.
The separate paid-referral panel contains 387 websites with at least one paid ChatGPT session. Of those, 102 have the financial tracking required for revenue and conversion analysis. The US website-by-website comparison includes 87 advertisers buying both paid channels. These denominators answer different questions and cannot be substituted for one another.
Lower revenue per visit is a material warning, not a causal verdict
The paper's US pooled channel totals within the 102-site financial panel show the following results. These are observed, unadjusted levels, not randomized treatment effects or forecasts for a new account. The full paper's Appendix J and Table J.1 provide the definitions.
| Observed US metric | Paid ChatGPT | Paid Google Search |
|---|---|---|
| Website revenue per session | $0.89 | $2.45 |
| Completed-order conversion rate | 0.37% | 1.17% |
The rounded revenue difference is approximately 64%. Separately, 84 of the 87 US dual-channel advertisers earned less revenue per paid ChatGPT visit: the rounded 97% finding. That is a share of advertisers in this comparison, not 97% of all ChatGPT advertisers, clicks, or purchases.
The authors also estimate adjusted channel differences accounting for website, device, and week. Those models strengthen the evidence that the observed gap is not simply a comparison of unrelated stores. They do not remove advertiser self-selection or establish that switching the same prospective customer from one channel to another would cause the reported difference.
Revenue is also not profit. The paper uses website-reported order value, including taxes and excluding shipping. Product costs, returns, servicing costs, and customer acquisition overhead still matter before a finance team can make a profitability judgment.
The $0.26 estimate is ROAS parity, not profit break-even
The researchers estimate a ChatGPT cost per click of about $0.26 at which tracked revenue per advertising dollar would match their paid-search benchmark. Calling that simply a “break-even CPC” loses the crucial qualifier: it is a relative return-on-ad-spend parity price, not the point at which an advertiser necessarily becomes profitable.
The calculation combines the revenue-per-session ratio with a $0.72 median Google Search CPC. That cost benchmark comes from only 16 linked accounts whose paid-search traffic was at least 80% US-based, not all 102 financially tracked sites. The paper's Appendices A, J, and L make this smaller denominator explicit.
The estimate assumes comparable margins and counts the tracked visit. Resampling advertisers produces a wide $0.05–$0.57 range while holding the search click-price benchmark fixed. One advertiser supplies 76% of US paid ChatGPT revenue, so concentration materially affects this calculation. The authors report that removing the largest advertiser widens the observed performance shortfall; concentration does not make the result disappear.
The frequently quoted $1.72 actual CPC comes from Opascope's single-account disclosure, last updated June 15. It covers 15 complete days and selected campaigns that were scaled after showing signal. It is not a market average, a current quote, or evidence that every advertiser pays above the study's parity price.
What the study cannot settle about your next customer
These are early-rollout website results, not a complete record of customer value. The researchers observe the last click bringing a visitor to a website. They do not observe the preceding prompt, ad impressions, user subscription tier, or the full sequence of later purchases through other routes.
That leaves plausible explanations to investigate, not conclusions to announce. Different purchase readiness, campaign choices, landing pages, and platform maturity could contribute to the gap. Later purchases could add value. None of those possibilities establishes how much value is missing or proves that conversational context will eventually outperform search.
The paper's matched cross-channel analysis found no detectable traffic lift, but could rule out only large increases. It does not establish zero indirect benefit. Nor should retail order metrics be treated as validated benchmarks for B2B qualified pipeline or long sales cycles. Search Engine Journal's examination usefully separates those limitations from the headline comparison.
The data supplier also deserves disclosure. Grips Intelligence provided the data free. Kaiser previously worked there and discloses a scientific-advisor role, salary, and virtual company shares. The paper says Grips had no role in the research and no right to approve or delay publication. That relationship belongs beside the evidence, without treating it as proof either for or against the findings.
The paper separately examines unpaid ChatGPT referrals after ads arrived. Its baseline volume estimate is not statistically significant; stronger declines depend on regional-growth assumptions. Those results should not become an unqualified claim that ads reduced trust or cannibalized organic traffic.
Modi's POV: contextual relevance has to earn its price
My view is that conversational relevance is an input to a media hypothesis, not a commercial outcome. An ad can fit a detailed discussion without reaching someone ready to buy now. Equally, a weak immediate conversion rate does not settle whether a campaign creates valuable demand later.
The useful response is neither “replace search” nor “abandon ChatGPT.” Ask whether this inventory can add profitable customers at a price your business can support. OpenAI's current advertising FAQ says it does not yet have performance benchmarks across advertisers, industries, or campaign types. Neither a favorable case study nor this unfavorable panel should become a universal budget rule.
A test decision that follows from the evidence
First, define the commercial comparison
Choose one offer, market, and customer outcome. Keep product availability, pricing, landing-page quality, and observation windows comparable where practical. Separate branded search from nonbrand acquisition rather than letting an established brand-capture campaign serve as an unexplained comparator. Record differences in audience and delivery that cannot be controlled.
Do not describe two simultaneous channel campaigns as a randomized experiment. They can establish account-level descriptive economics, but the visitors still selected into different environments. Set the acceptable acquisition cost using contribution margin and a finance-approved payback horizon, not the paper's $0.26 estimate.
Second, reconcile sessions, charges, and outcomes
Keep paid ChatGPT and organic ChatGPT referrals separate. Preserve campaign identifiers and reconcile billed clicks, website sessions, orders, and deduplicated CRM records. OpenAI's ads basics documentation warns that review crawlers can visit tracked landing URLs and appear as paid traffic in third-party analytics without being billable clicks. Check for that discrepancy; do not assume it explains the paper's results.
Report immediate orders separately from later cohort revenue. For B2B, retain lead qualification, sales acceptance, closed revenue, and elapsed time rather than relabeling every form fill a customer. Our analysis of why GA4, Search Console, and a CRM are not full AI attribution explains why joining systems still requires accountable interpretation.
Third, test incremental value where feasible
Before launch, agree a spend ceiling, minimum detectable commercial effect, evidence requirement, conversion-maturation window, and stopping rules with an analytics owner. There is no universal spend or order count that makes every test decisive. If expected volume cannot answer the question within the affordable loss limit, narrow the question or defer the test.
Where feasible, use an appropriately designed randomized holdout or a reviewed geographic experiment to examine incremental outcomes. Check contamination, overlapping campaigns, seasonality, and statistical power. If only an observational comparison is possible, label it accordingly. A wider attribution window can reveal additional credited conversions; it cannot by itself prove that the ads caused them.
Finally, make the next decision explicit
Continue a bounded test when tracking is reconciled and uncertainty is still worth resolving. Pause when billing or outcome definitions cannot be explained. Consider expansion only when sufficiently mature account economics meet the agreed margin and payback requirements, with incremental evidence appropriate to the investment at risk.
Integrated.Social's PPC and Performance Max service is the relevant path for reviewing the comparator, conversion definitions, and budget decision. The study gives that conversation a sharper starting point: make conversational advertising earn its place with measured commercial evidence, rather than treating either channel's reputation as a result.
