ChatGPT ads vs Google Ads ROAS: what four months of data actually show

Here is the short version, four months into ChatGPT advertising. On a modeled basis, ChatGPT ads and Google paid search land within about 3% of each other on cost-per-conversion. At a $4 blended cost-per-click and a 7.1% conversion rate, a ChatGPT click costs roughly $56 to convert. At Google's $4.51 average search CPC and a 7.8% paid-search conversion rate, the number is roughly $58. Two channels, nearly identical unit economics.

But the averages hide the story. In high-consideration retail categories - consumer electronics, home and garden, lifestyle and wellbeing - Criteo reports AI-referred traffic converting at close to twice the rate of traditional search. In pure transactional ecommerce, a 973-site study found ChatGPT referrals converting worse than both paid and organic Google search. The ROAS question does not have one answer. It has a distribution, and where you sit on it depends almost entirely on your category and your buyer's intent.

Why this matters

Every marketing team we talk to is getting the same question from a CFO right now: should we be buying ChatGPT ads, and what return should we expect? Until recently it was unanswerable. ChatGPT ads only went live in the US on February 9, 2026, on the Free and Go tiers, and OpenAI has spent the months since changing the pricing model out from under advertisers - from a $60 CPM at launch to a $3-5 cost-per-click by April.

That instability is exactly why a modeled answer matters more than a measured one. There is not yet enough clean, attributed account-level data to publish a true ROAS benchmark for ChatGPT ads the way WordStream or Triple Whale publish one for Google. What there is: reliable inputs - cost-per-click, conversion rate, intent signals - from OpenAI, Criteo, Similarweb, and independent academic work. Assemble those into a unit-economics model and you get a defensible answer, as long as you stay honest that it is modeled, not measured. That is what this report does.

We build agents that run these channels for marketing teams - Oz is Claude Code for marketers - so our practical question is not whether ChatGPT is a real ad channel (it is) but what an agent needs to know to allocate budget to it correctly. An honest ROAS read is the first input.

Method

This is a synthesis-and-model study, not a measured account study. We did not run a controlled spend test across the two platforms. We (1) collected published 2025-2026 benchmarks for cost-per-click, conversion rate, and ROAS on both channels from primary and reputable secondary sources, and (2) combined them into a cost-per-conversion and modeled-ROAS comparison, then ran that comparison across category and intent scenarios.

Inputs. ChatGPT pricing: OpenAI's launch was CPM-based at roughly $60, switching to a $3-5 recommended CPC by April 2026 (with software and finance verticals running $8-18), per Digiday and Search Engine Land reporting. Conversion: Similarweb clickstream data across a tracked panel (April-May 2026) puts ChatGPT referral conversion at 7.1%, against 7.8% for paid search. Category lift: Criteo's May 5, 2026 update reports AI-referred conversion approaching 2x traditional search in consumer electronics, home and garden, and lifestyle and wellbeing, across 1,000+ live brands. The counterweight: a working paper covering 973 ecommerce sites and $20B in revenue (August 2024 to July 2025) found ChatGPT referrals converting below organic search, which ran 13% higher, and affiliate, 86% higher, with only paid social converting worse.

Google inputs. Average Google search CPC of $4.51 across 10,000-plus commercial keywords (WebFX); 2026 median Google Ads ROAS around 3.5:1, with search campaigns at 5.17:1 and Performance Max at 2.57:1, and a 2025 year-over-year ROAS decline of roughly 10%, per Triple Whale and WordStream benchmark aggregations.

What we modeled. Cost-per-conversion equals CPC divided by conversion rate. Modeled ROAS equals conversion rate times average order value, divided by CPC - run at a representative $80 order value and stress-tested up and down. We treat the conversion-rate figures as channel-level clickstream events, not verified purchases, which is why we lead with the relative comparison between channels rather than any single absolute ROAS number. Every external figure is linked at the point of use.

Findings

1. Modeled cost-per-conversion is nearly identical: about $56 vs $58

On the blended averages, the two channels are a statistical tie. At a $4 ChatGPT CPC and 7.1% conversion, modeled cost-per-conversion is $56.34. At Google's $4.51 search CPC and 7.8% conversion, it is $57.82 - a 2.6% gap. Translate to ROAS at an $80 order value and both land near 1.4x on a last-click basis. The takeaway is not the absolute number, which moves with your AOV and margin; it is that ChatGPT ads have already closed to within a rounding error of Google paid search on raw efficiency, four months after launch. That is faster convergence than display, social, or retail media managed in their first year.

2. ChatGPT's CPC advantage is real but concentrated in B2B

Where ChatGPT undercuts Google, it undercuts hard. For general ecommerce, ChatGPT's $3-5 CPC sits right on top of Google's $4.51 average - no meaningful arbitrage. The gap opens in high-value verticals: ChatGPT software and finance clicks run $8-18, while equivalent Google search clicks in legal and insurance top $50-67. A SaaS or professional-services advertiser paying $25 a click on Google can find materially cheaper inventory inside ChatGPT today, because the auction is young and thin. That arbitrage compresses as advertisers pile in - over 1,000 brands are already live through Criteo's integration - so the cheap-CPC window is a 2026 phenomenon, not a structural moat.

3. The intent premium is category-specific, not channel-wide

ChatGPT does not convert better. High-consideration categories convert better inside ChatGPT. This is the most misread stat in the space. Criteo's roughly 2x lift shows up specifically in consumer electronics, home and garden, and lifestyle and wellbeing - categories where buyers research at length and reach the ad already deep in consideration. Similarweb's panel shows ChatGPT referrals at 7.1%, essentially matching paid search at 7.8% on average. The lift is not the channel; it is the buyer. A shopper who asks ChatGPT for the best standing desk under $400 is a different person from one who types standing desk into Google, and the conversion math reflects it. We see the same pattern in our own AEO work - AI-sourced visitors convert several times better than generic organic, but only when the underlying query carried real purchase intent (more in our citation engineering playbook, and on why getting cited at all is its own discipline in our query fan-out analysis).

4. In pure transactional ecommerce, ChatGPT currently loses

The counter-evidence is strong and should not be waved away. The most rigorous dataset in the space - 973 ecommerce sites, $20B in revenue, an SSRN working paper - found ChatGPT referrals converting below organic search and well below affiliate, beating only paid social, and trailing both paid and organic search on revenue per session. If your category is fast, transactional, and low-consideration - commodity consumables, impulse retail - the current ChatGPT ROAS picture is unfavorable, full stop. The discovery-driven categories where Criteo sees a 2x lift are precisely the ones a 973-site average dilutes. Both findings are true; they describe different populations.

5. The volume ceiling caps total return, not per-click return

ChatGPT can be efficient and still small. AI referral traffic is roughly 0.2% of ecommerce sessions today - about 200 times smaller than Google organic - and ChatGPT alone is more than 87% of it. The pilot's early click-through rate was reported around 0.91%, a fraction of search-ad CTRs. So even at parity ROAS, the absolute revenue you can route through ChatGPT ads in 2026 is a sliver of what Google moves. The offsetting signal: OpenAI reportedly booked $100M in ad revenue in the first six weeks and is expanding the pilot to the UK, Mexico, Brazil, Japan, and South Korea. The ceiling is rising fast. Today it is a high-efficiency, low-volume channel - a tactical add, not a Google replacement.

6. Attribution understates ChatGPT ROAS

Whatever number you measure is probably low. Conversions that begin in a ChatGPT conversation frequently complete as direct or branded-organic visits, so last-click reporting systematically under-credits the channel. OpenAI returns only aggregate performance data - views and clicks - to advertisers, with no access to chat content, which limits granular optimization. Criteo's role as the ad-serving layer adds another attribution hop. The practical implication: do not judge ChatGPT ads on last-click ROAS alone in 2026. Use incrementality tests or holdouts - the same discipline that rescued paid-social measurement after iOS 14.

What this means for marketers

Five takeaways for an in-house team or agency deciding how to treat ChatGPT ads this year.

Budget it as a high-intent, low-volume tactical line, not a Google substitute. Expect Google-comparable efficiency at a fraction of the volume. Size the test accordingly: a small, measurable allocation, not a channel migration.

Let category decide your conviction. If you sell considered purchases - electronics, home, wellness, B2B software - the intent premium is on your side. If you sell fast transactional goods, wait for the auction and the data to mature.

Chase the B2B CPC arbitrage while it lasts. High-value verticals paying $25-plus per Google click should test ChatGPT now, before the 1,000-plus advertisers already on the platform bid the cheap inventory away.

Measure with incrementality, not last click. Last-click will under-report ChatGPT. Build a holdout from day one or you will kill a profitable channel on bad data.

Win the organic citation first. ChatGPT shows organic answers above the ad. If you are not already cited in the response, your ad is fighting a recommendation that ignored you - earned AEO visibility and paid ChatGPT placement compound.

Limitations

This is a modeled comparison built on third-party benchmarks, not a controlled spend test we ran ourselves. Treat every number as directional.

The conversion figures are not the same metric. Similarweb's 7.1% is a clickstream conversion event across a tracked panel; the 973-site study measures ecommerce purchase conversion; Criteo's 2x is a category-level retail multiple. They are not strictly comparable, which is why we lead with relative gaps inside each source rather than splicing them into one absolute ROAS. The ChatGPT data window is four months and the pricing model changed twice inside it, so any benchmark has a short shelf life. Vendor studies from Criteo and Similarweb carry an interest in a flattering AI-commerce narrative; the independent 973-site paper is the useful counterweight, and we weighted it accordingly. Average order value, margin, and your own conversion rate move the absolute ROAS far more than the channel choice does, which is why the model uses an illustrative $80 order value. And OpenAI's aggregate-only reporting means no one outside OpenAI can yet audit true account-level ROAS. When that data arrives, some of these estimates will move.

ChatGPT ads have not beaten Google on ROAS. They have matched Google's unit economics in four months while moving a two-hundredth of the volume - which is either trivial or the most important thing happening in paid media, depending on how fast the volume curve bends.

The honest four-month verdict: ChatGPT ads are a real, efficient, category-dependent channel with a small but fast-rising ceiling. The teams that win 2026 will not be the ones who switched everything to ChatGPT ads, or the ones who ignored them - they will be the ones who tested early, measured with incrementality, and let an agent reallocate budget across both channels as the data came in. That cross-channel reallocation, run continuously instead of in a quarterly spreadsheet review, is exactly the kind of work we built Oz to do.

If you want an agent watching ChatGPT and Google ROAS side by side and moving budget to whichever is winning your category this week, join the Oz waitlist.