A headline says ChatGPT Ads return $0.52 per ad dollar and Google Search returns $3.40. Your boss asks whether to move budget. Before you answer, know that only one of the numbers behind that comparison was measured. The other came from someone else's click prices.
This post shows how to run the comparison with your own numbers, calculate the click price ChatGPT would need to break even, and report a result without overselling it. The figures come from Search Engine Journal's analysis of a university working paper, which pooled data from 102 websites.
What was measured and what was borrowed
The paper measured revenue per session. In U.S. traffic, paid search produced $2.45 per session and ChatGPT ad sessions produced $0.89. Transactions per session were 1.17% for search and 0.37% for ChatGPT.
The per-dollar figures add a click price. Dividing $0.89 by a $1.72 ChatGPT CPC gives about $0.52. Dividing $2.45 by a $0.72 search CPC gives about $3.40.
Search Engine Journal reports that the $1.72 comes from one agency account: about $60,000 over 15 days, on campaigns the agency scaled after they showed promise. The $0.72 is a median from 16 mostly U.S. Google Ads accounts, and it was not matched to the revenue each site earned.
Change either price and the per-dollar result moves, even though session revenue stays put. That is the whole mechanism, and it is why you should run the math with your own CPCs.
The break-even CPC formula
Start with the return you need. For most teams that is the current Google Search return per ad dollar, or a lower threshold finance will accept for a new channel.
Break-even CPC = ChatGPT revenue per session ÷ target return per ad dollar
With the paper's figures: $0.89 ÷ $3.40 = about $0.26. That matches the roughly $0.26 click price the article says would equal search's return. The paper reports the ChatGPT price at 6.57 times the level needed.
You can also skip the return step. The revenue-per-session ratio is $0.89 ÷ $2.45, about 0.36. So ChatGPT needs a CPC of roughly 36% of your Google CPC to match your return.
How the result flips with click price
The table below is illustrative. It pairs the paper's pooled revenue per session with Google click prices from agency reports the article mentions, which is the same kind of mixing the headline does. It shows the mechanism. It is not a forecast for anyone. This pairs well with how to evaluate aeo research before changing your seo explained.
| Google Search CPC | Search return per $1 | ChatGPT break-even CPC | |---|---|---| | $0.72 (paper's median) | $3.40 | $0.26 | | $6.76 (one agency's Google CPC) | $0.36 | $2.46 | | $10.46 (another agency's Google CPC) | $0.23 | $3.80 |
If your search clicks cost $0.72, a $1.72 ChatGPT click sits far above break-even. If they cost $6.76, a $1.72 click clears the bar on session revenue alone. A $3.14 click (the ChatGPT CPC one agency reported) still misses it at $2.46. Same ChatGPT session value, opposite verdicts, and the only thing that changed was the Google price.
The formula also needs your real ChatGPT session revenue. Until you have it, any placeholder, $0.89 included, is borrowed. Swap it out as soon as your own test produces enough sessions.
Match the settings before you compare the totals
Platform reports count conversions differently. Per the article, OpenAI's Measure Results page says Ads Manager totals may include eligible view-through conversions. Advertisers pick a 7-day, 14-day or 30-day click window, the view-through window can be one day or off, and both windows use last-touch logic. Check your Google Ads conversion settings against that list.
Run through this checklist before anyone sees a chart:
- Test period: Use the same dates for both channels. Don't compare a 15-day ChatGPT burst with a quarter of Google data.
- Click window: Set the same 7, 14 or 30 days in both platforms.
- View-through: Turn it off in ChatGPT Ads, or report both versions separately. Confirm what Google is counting.
- Conversion definition: Use the same action on both sides. A completed order is not a sign-up, and the paper does not explain how it reconciled the two for the roughly 20% of sites in services, software, education or finance.
- Qualified leads: For lead gen, write one definition (for example, passed sales review) and apply it to both channels.
- Spend and sessions: Record spend, clicks and sessions per channel, not only revenue.
- Conversion lag: Wait long enough for leads to close. The B2B compliance test in the article ran 30 days, and the agency still couldn't compare cost per customer because of the client's long sales cycle.
Keep the metric close to money, too. Jotform's reported cost per sign-up sits further from revenue than cost per customer, and sign-ups are not sales. This pairs well with ai performance review questions for seo teams that work in depth.
How many sessions before a zero means anything
Only 18 of the paper's 102 sites recorded transactions from ChatGPT ads. A zero on a small test can't tell you whether the channel is weak or the test was just small.
What follows is arithmetic, not a finding from the study. Assume your ChatGPT conversion rate matched the pooled 0.37%. With 100 sessions you'd expect 0.37 orders, and a zero would turn up roughly 69% of the time. With 1,000 sessions you'd expect about 3.7 orders, and a zero would turn up around 2.5% of the time.
A rough rule: divide 3 by your expected conversion rate to get the sessions needed before a zero starts to mean something. At 0.37% that is about 810 sessions. If you sell high-ticket items and your rate is far lower, you need far more sessions. Or measure an earlier step, such as qualified leads, and say so.
The agency examples in the article show the problem. One agency reported zero qualified leads from ChatGPT ads against two from Google, and it gave no spend or dates for the Google side. Another logged one form submission from 263 clicks. Neither supports a firm verdict.
Reporting template and verification
Show sessions beside revenue, every time. A small table does the job:
Also read: related topic: how to audit brand presence on reddit: a worked example
| Metric | ChatGPT Ads | Google Search | |---|---|---| | Dates | same | same | | Click window / view-through | 7/14/30 day, on or off | 7/14/30 day, on or off | | Spend | | | | Clicks and sessions | | | | CPC | | | | Conversions (defined as: ___) | | | | Revenue or qualified-lead value | | | | Revenue per session | | | | Return per ad dollar | | | | Break-even CPC | | |
Add a limitations note under it. For example: "Small test. Results reflect these dates, this budget and these attribution settings. Lead outcomes still maturing. Not evidence of what a budget shift would do."
To verify, tag ChatGPT ad URLs with a distinct source value. Then compare three things for the same dates: platform-reported conversions, analytics sessions and conversions for that source, and orders or CRM records that tie back to them. When the three disagree, find out why before the number goes into a deck. Likely suspects are the click window, view-through, deduplication and tracking gaps.
What the study can and cannot tell you
The paper is descriptive, and the authors say so. Advertisers chose to buy ChatGPT ads, so the study doesn't show that switching dollars would help or hurt a specific advertiser.
The pooled figures cover up to 28 weeks of ChatGPT sessions in U.S. traffic, through Aug. 22, which makes it an earlier period. The ChatGPT click price comes from one account. The search price is a median from 16 accounts. An OpenAI spokesperson told Search Engine Journal the study captures an earlier stage of ChatGPT Ads and extrapolates from a small sample, and pointed to the newer Target CPA and Target CPC options. Grips Intelligence donated the data, and one author is a scientific advisor there. The authors say the company played no role in the research.
In the paper, 84 of 87 advertisers using both channels in U.S. traffic earned less revenue per session from ChatGPT. That describes the past. It doesn't predict your account.
If your Google CPCs are low and your revenue per session is solid, ChatGPT would need very cheap clicks to match you, so treat a test as a small learning budget. If your Google CPCs are high, the break-even CPC may be within reach, and a matched test is worth running. Either way, hold the test to equal dates, equal settings, equal definitions and enough sessions. Promise stakeholders a clean read, not a result.



