ChatGPT Ads vs. Google Search: Early Data Offers No Case for Shifting Budgets
Early data shows lower revenue per visit from ChatGPT Ads than from Google paid search. But the study does not show that shifting budget would improve results for any given advertiser.
On November 16, Search Engine Journal examined the first results for ChatGPT Ads compared with Google search ads. The analysis followed OpenAI’s introduction of target cost bidding the previous week. Ads have been running in ChatGPT since February, but the available data does not show that moving budget from Google Search Ads would improve results for any given advertiser. Before making that decision, it is worth comparing your own campaigns using the same attribution settings.
What ad-driven visits show
Kaiser and Schulze’s study used data from 102 websites that track transactions and revenue. For US traffic, a visit after a click on a ChatGPT ad generated $0.89 in revenue, compared with $2.45 for paid search. The share of visits that resulted in a transaction was 0.37% and 1.17%, respectively. Of the 87 advertisers using both channels in the US, 84 saw lower revenue per visit from ChatGPT. We can be more confident about revenue per visit here than about return on ad spend.
The comparison describes the results websites saw, but it does not establish what caused the difference. Advertisers chose for themselves whether to buy ads in ChatGPT; the researchers did not test what would happen if money were moved between comparable campaigns. The data covers the period up to August 22, with no more than 28 weeks of US ChatGPT visit data available. There is no breakdown by month or bidding type, so this study cannot be used to assess the effect of tools introduced later.
OpenAI says the study covers an early period for its ad platform and relies on a small sample. The company points to subsequent updates, including Target CPA and Target CPC. However, the OpenAI materials reviewed by the author of the analysis contain no data on typical actual cost per click across advertisers.
Where the revenue-per-ad-dollar gap comes from
Other figures have also been published: $0.52 in revenue per dollar spent on ChatGPT Ads versus $3.40 in search. They look like a direct measure of return, but they are calculated figures. Measured revenue per visit was divided by click prices from separate sources: $1.72 for ChatGPT from one account and a median of $0.72 for search from 16 accounts. The search click price was not matched to the revenue of each website in the study.
The $1.72 figure comes from a report by agency Opascope: one account with about $60,000 in spend over 15 days. The agency noted that the calculation included campaigns it scaled after encouraging results, not every test it ran. Another business could pay a different price per click, changing the calculated revenue per ad dollar. Also, the study measures revenue per visit, while the ad platform reports cost per click. Visits and paid clicks cannot automatically be treated as the same thing.
Individual cases show why what happens after the click matters. Over 30 days, StubGroup generated cheaper leads from ChatGPT Ads than from Google for a B2B company: $92.21 versus $157.57. But because of the long sales cycle, it is not yet known how many customers those leads will become. Workshop Digital spent $2,319 on ChatGPT clicks and received no qualified leads. Brafton spent $1,524.11 and got 263 clicks and one form submission. In Jotform’s example, the cost per registration was about 30% lower than in paid search, but the spend and attribution method were not disclosed. Cost per registration does not tell us the cost of acquiring a customer.
What to check before shifting budget
For a website owner or marketer, it is more useful to compare their own spend and the revenue attributed to each channel over the same period than to apply someone else’s click price to their store. Revenue should be shown alongside the number of visits: no sales after a few dozen visits and no sales after a few thousand tell different stories. For lead-generation campaigns, it is important to define a qualified lead the same way in both channels and give leads time to become customers.
Measurement settings also need checking. In OpenAI’s Ads Manager, you can choose a 7-, 14-, or 30-day post-click attribution window; reported results may include eligible conversions after someone views an ad. Website analytics counts visits, while Ads Manager counts clicks. Discrepancies can arise from redirects, consent settings, and browser blocking, among other things. If the platforms attribute conversions to ads differently, comparing their reported totals will not reveal the difference in channel performance.
My view: calculate the store’s results first
I would not shift budget just because more than 100,000 advertisers have already run ads in ChatGPT or one case study produced a cheap lead. I would find it easier to approve a limited test budget and decide in advance what would count as success—for example, cost per order. Otherwise, there is a risk of paying for visits and then arguing over whether they brought customers to the store.
But I would not declare ChatGPT Ads unprofitable for everyone based on these averages, either. For my store, the question is simpler and more demanding: how much do the visits cost, how many orders do they bring, and can one be reliably linked to the other? Until I can see that in my own data, the promise of saving money by shifting budget does not convince me.
Sources
Where the news comes from. The text is a retelling in the author’s own words; the facts come from the source, the opinion is the author’s.
- ChatGPT Ads Vs. Google Search Ads: What Early Data Shows via @sejournal, @MattGSouthern www.searchenginejournal.com
Cosmetics store owner
I run a small cosmetics store and edit its CMS-powered website myself. Before switching platforms, I want to know how much it will cost to migrate the catalog and train the staff.
All posts by the authorRelated articles
In the US, Google Search Ads Generated More Revenue per Visit Than ChatGPT Ads
In the US, ChatGPT ads generated $0.89 in revenue per visit, compared with $2.45 from paid Google Search. The study compared sales across 102 advertiser websites.
ChatGPT Ads Tests New Ad Formats and Updates Measurement Tools
OpenAI is testing larger images and cards in ChatGPT Ads, has updated the Conversions page, and has launched weekly emails with ad performance results.
ChatGPT Links Lost Their UTM Tags After the GPT-6 Release
ChatGPT links appear to have stopped passing UTM tags after the GPT-6 release. Some visits may now be counted as direct traffic, making it harder for site owners to identify where visitors came from.
Discussion 5
Samat Ganiev
For a small learning center, the useful comparison is not just online revenue: track inquiries through to enrolled students, including calls and walk-ins, and note whether there is room in the classes being advertised. Otherwise a channel can look weak simply because its leads are recorded outside the checkout flow—or look strong while sending inquiries staff cannot serve.
Laura Bērziņa
the $0.89 vs. $2.45 revenue per visit gap is a useful early signal, but i’d want to compare the same purchase window and attribution settings before changing budgets. did the study break results down by product category or customer journey stage? that could show where chatgpt ads might work as a small test rather than a direct replacement for search 🙂
Gulsara Latypova Post author
@laura.growth The study doesn’t report a product-category or journey-stage breakdown, so I wouldn’t assume the pooled numbers identify a good niche. For a small store, I’d run a limited test with the same purchase window and attribution settings, then compare actual order value and the cost of managing the extra channel before moving budget.
Dilshod Karimov
One practical check I’d add is whether the test needs extra creative work or reporting time; that cost can erase a small channel’s value even when its ad spend looks modest. I’d set a fixed test budget and end date upfront, then compare net sales after those costs before reallocating anything, @gulsara_l.
Beatriz Inês
@laura.growth, the summary doesn’t mention a breakdown by category or journey stage, so we can’t tell whether those pockets exist from these numbers. And yes, matching purchase windows and attribution is table stakes; then I’d test ChatGPT as a capped, incremental channel for a specific use case—not promote it to “replacement” based on an average. The long-term question is whether that test stays measurable as the channel and bidding mature.