This past week, OpenAI rolled out cost-target bidding in its ChatGPT Ads platform. These ads first launched in February. By Aug. 31, OpenAI reported that its ads product had achieved a $1 billion annualized run rate.
The question for a paid search team is whether ChatGPT Ads now compete with Google Search Ads for advertising dollars, and whether the early results make a case for moving some of their budget to ChatGPT Ads.
ChatGPT Ads are competing for advertisers’ budgets. However, the revenue metrics, survey data, and Google’s disclosures I reviewed give no indication of how many dollars have been diverted away from Google Search Ads to ChatGPT Ads.
In terms of performance metrics, a ChatGPT ad click visit to an advertiser’s website generated about one-third the revenue of a paid search visit in U.S. traffic. The working paper in my Nov. 2 coverage provided the pooled revenue metric from 102 different websites with transaction and revenue tracking. The authors of that paper called the comparison descriptive. One of the authors’ universities also released a press statement saying that for every dollar an advertiser spends on ChatGPT Ads, they receive $0.52 back in revenue (compared to $3.40 for every dollar spent on Google Search).
Only the first of these performance metrics was actually measured. The second figure combined that measured performance metric with the average cost-per-click pricing from one ChatGPT account and the median cost-per-click pricing from 16 paid search accounts. So the second figure varies with what an advertiser pays per click on each of those channels.
What The Evidence Says About Search Budgets
An Oct. 21 engineering post from OpenAI puts the run rate above $1 billion, with more than 100,000 advertisers having run ads since launch. Alphabet reported $63.35 billion in Search & other revenue for the third quarter, up 12%, which I covered on Oct. 28.
On Alphabet’s earnings call, Chief Business Officer Philipp Schindler cited “the lower monetization we were seeing in AI Mode” as one factor in slower Search growth, without giving a figure or a baseline.
Google Search was the most-mentioned potential source of future ChatGPT ad spending in a Bank of America survey of 117 advertisers and agencies, TheStreet reported on Oct. 14. I haven’t seen the bank’s note, and the survey asked about plans. It doesn’t measure money that has moved.
Neither company’s revenue figures show spending that shifted from one platform to the other. Google can continue to grow even as some advertisers move part of their budgets elsewhere.
What The Per-Visit Numbers Show
In U.S. traffic, paid search generated $2.45 in revenue per session versus $0.89 with ChatGPT ad sessions, and transactions per session came to 1.17% against 0.37%.
These pooled figures are based on a total of 60 weeks (through Aug. 22). For U.S. traffic, there were up to 28 weeks of ChatGPT ad sessions available for analysis.
Kaiser and Schulze built a statistical model that controlled for website, device type, and week, and found that ChatGPT ad sessions generated lower rates of conversion and lower revenue per session than paid search. These results were robust even after removing individual advertisers. Out of 87 different advertisers who used both paid search and ChatGPT ads in U.S. traffic, 84 produced less revenue per session using ChatGPT ads.
About 20% of the 102 websites studied were services, software, education, or finance websites. A table footnote explains that these websites report online conversions as bookings, leads, licenses, or sign-ups. However, the appendix calls a transaction a completed order, and it does not explain how the authors defined the relationship between these two types of metrics.
Because advertisers chose to buy ChatGPT ads, the authors do not argue causally that ChatGPT is worse for advertisers than paid search. Rather, the authors view their work as “descriptive rather than causal,” a record of what these websites received from each advertising channel. They do not provide any evidence as to whether switching advertising dollars from paid search to ChatGPT under matched campaigns would be beneficial or detrimental to specific advertisers.
Data for this study were donated by Grips Intelligence, where Kaiser, a former employee, works as a scientific advisor. The authors disclose that Kaiser has received a salary and virtual company shares from Grips Intelligence. The authors add that Grips Intelligence played no role in conducting the research described in this study.
I contacted OpenAI about this study and received an email response from an OpenAI spokesperson saying the study “captures an earlier period of ChatGPT Ads’ development and extrapolates platform performance from a small sample.” The statement also says OpenAI has kept launching products since the ads launched, with Target CPA and Target CPC among the recent ones.
Why The Per-Dollar Figure Depends On Click Prices
Kaiser and Schulze express the difference in terms of value for money, revenue per session divided by click price. Treating each session as a paid click, that works out to about $0.52 per ad dollar for ChatGPT ($0.89 ÷ $1.72) and $3.40 for search ($2.45 ÷ $0.72), the two figures in the Frankfurt School release. The authors report that an ad visit from ChatGPT returns about one-third the value of a search visit at about 2.4 times the price. In their abstract, they state that ChatGPT ads are worth less than a sixth the value for money of Google Search. A table provides a reported ChatGPT price of 6.57 times what would be necessary to produce the same return on investment as Search.
The Frankfurt School release cites the difference as “6.6 times better value for money” for Google and reports this in revenue per ad dollar. Its $0.52 and $3.40 figures are not provided within the body of the paper, and the release itself gives no explanation of how they were calculated.
There is no single price given for search, but rather a median price of $0.72, which comes from 16 mostly U.S. accounts that have linked Google Ads cost records. This price was not matched to the revenue each site generated.
The ChatGPT price is $1.72, which Opascope, an agency that runs paid media for clients, published in June. It covers one account and about $60,000 in spend over 15 days, on campaigns the agency scaled after they showed promise. Opascope wrote that the numbers aren’t an average of every test it launched, and it reported a 1.49x ROAS.
Just one website in the sample did as well as Opascope’s account on both conversion rate and implied revenue per visit, and it had only nine ChatGPT ad sessions.
The authors suggest that $1.72 is a conservative price as it is toward the low end of what practitioners report and below the $3-$5 OpenAI suggests be used as a starting maximum bid. Also, bidding is based on a maximum value, which is simply the cap an advertiser sets, not necessarily what each click will cost.
Resampling advertisers, while keeping the search price constant, gives the authors a range of $0.05 to $0.57 around a ChatGPT click price of about $0.26 that would match search’s return. Neither the ChatGPT click price nor the search price was changed.
What Early Case Studies Show
Advertising agencies have provided reports about their individual ChatGPT Ads campaign data for one advertiser, and separately, Brooke Osmundson reported in September that advertisers lacked the context to assess what they were being told by the numbers in such reporting.
StubGroup provides ChatGPT ads management services. It ran Google ad campaigns and ChatGPT Ads campaigns for an advertiser who is a B2B compliance company, for 30 days. The clicks on ChatGPT ads cost $0.76 a click. The clicks on Google search campaigns cost $10.46 a click. The cost per lead with ChatGPT was $92.21; with Google it was $157.57. The conversion rates were 0.8% with ChatGPT, and 6.64% with Google.
The agency said that while the advertising campaigns run through Google have years of optimization behind them, the leads generated through those campaigns appear to be converting to customers at a greater rate so far than the leads generated through ChatGPT ads. As a result of this, the cost per customer appears to be lower for Google. How many customers will be ultimately generated from the leads created through both platforms, StubGroup does not yet know, due to the client’s long sales cycle. Additionally, the agency said that the $0.76 per click was low when compared to the CPCs of some of its other clients.
Workshop Digital used its own funds to pay for $2,319 worth of clicks on ChatGPT Ads. Workshop Digital’s CPC for ChatGPT ads was $3.14. For Google Search the CPC was $6.76. The CTR for Workshop Digital’s ChatGPT ads was 1.1%. For Google, the CTR was 5.2%.
Workshop Digital received zero qualified leads from its ChatGPT ads, and two qualified leads from Google. The agency called the two types of advertising campaigns “not a direct comparison,” although it didn’t provide the spend amount or dates for Google Search as part of the comparison.
Brafton, which sells ChatGPT Ads services, ran one click campaign with one ad from June 17 to Aug. 28. It spent $1,524.11 for 263 clicks and one form submission. Its Google and LinkedIn comparisons are outside benchmarks, and its director of paid media, Zach Shah, said ChatGPT ads are “priced like search, but seem to behave like paid social.”
All three agency reports came down to what happened after the click, the stage with the fewest results.
Additionally, OpenAI pointed me to some customer success stories via its ad page. One such example is from Jotform. According to this example, their cost per signup is approximately 30% less than with paid search. This figure is based on Jotform’s internal data collected during July; however, no costs are provided behind it, nor does it detail how the sign-ups were attributed. Sign-ups do not equal sales.
Why This Matters
A paid search team may be asked to shift some portion of search spend to ChatGPT on the strength of a per-dollar figure based upon another party’s click costs. For an account paying a number of dollars per click on search, a number greater than the median of the paper’s data, there are differing starting points for making such comparisons. The basis for comparison that makes sense begins with actual spend and attributed revenue on each channel for the same period of time and using the same attribution settings.
The size of the test will affect how much an ecommerce account testing ChatGPT Ads can learn from a weak result. In all, 18 of the paper’s 102 sites recorded transactions resulting from ChatGPT ads; a single zero does not distinguish a weak channel from a small test. So when reporting to stakeholders, the number of sessions associated with each ad campaign belongs beside the revenue figures, because what a zero means is quite different depending on whether there were a few dozen or a few thousand visits.
To compare results, lead-gen accounts also need to define a qualified lead in exactly the same terms on both channels and give those leads sufficient opportunity to convert. This is why StubGroup’s customer comparison is still open, since its lower-cost leads have not yet turned into lower-cost customers. While a cost per signup was used by Jotform as the metric in their case study, it is further removed from revenue than a cost per customer.
A team planning to present Ads Manager results either against the paper’s data or against its Google Ads performance metrics will first need to document its attribution settings. On OpenAI’s Measure Results page, Ads Manager conversion totals may include eligible view-through conversions, and advertisers select from a 7-day, 14-day, or 30-day click window. The view-through window can be set to one day or disabled, and both windows employ last-touch logic. The paper tracked visits that were attributable to the ad.
Ads Manager tracks clicks, while analytics uses sessions. There can be differences between these two numbers. On OpenAI’s basics page, the company says ad-review crawling programs may access landing pages that contain tracking parameters, and analytics tools may report these visits as paid traffic even if they do not represent billable clicks. The Measure Results page provides additional explanations for these discrepancies, including redirects, consent settings, and browser-based blocking. Only by assuming clicks and sessions align can the release’s $0.52 be calculated from reported revenue per session and cost per click.
Open Questions
OpenAI’s ads FAQ says that they do not currently provide performance benchmarks. In addition, all four OpenAI help pages I reviewed provided a suggested starting bid, but none provided any information about click prices across advertisers. Neither did OpenAI’s statement. One advertiser’s cost is documented in the research paper. Each agency CPC was unique to its specific case.
There were no monthly or bid-type breakdowns within the research paper, so it could not establish if the outcomes had been affected by OpenAI adding conversion-optimized campaigns in June. Data from the research paper ceased prior to when OpenAI implemented bid adjustments, which went into effect this month.
On Oct. 28, Google published an ads post stating that CTR for sponsored and organic product listings in AI Mode were similar. This was based on two weeks of internal U.S. data. Google gave no absolute rates. That was the single performance figure for AI Mode ads in my review of Google documentation.
Looking Ahead
Two of the bid options now offered for new ChatGPT Ads campaigns didn’t exist during the paper’s data window. Since Nov. 9, new conversion campaigns that want a cost target use Target CPA, and new click campaigns can use Target CPC, as I covered this week. The option with no cost target is Intelligent bidding, previously named Maximize Results, according to OpenAI’s bidding page. Beginning with Nov. 18, all values submitted using the legacy API bid fields are treated as the average-cost target when creating a new campaign or changing an ad group’s bid type. Max Bid and Max CPC will officially retire for existing campaigns on Jan. 1, 2027.
Performance improvement is not shown in the new options. Noting which objective, bid strategy, and attribution windows produced each fourth-quarter result will help a team decide if the metrics generated this year have enough similarities to compare next year.
More Resources:
- ChatGPT Ads Aren’t Paid Search – 5 Questions To Answer Before You Shift Budget
- ChatGPT Ads: New Acquisition Channel Or Just Another Brand Tax?
- How To Leverage AI Ad Placements For Stronger PPC Performance
Featured Image: Tada Images/Shutterstock