Do ChatGPT Ads Actually Work? The Three Public Numbers, and What Each One Measured
By The InPromptAds Team 8 min read
Last verified 15 August 2026. Version 1.0. Next scheduled review 15 September 2026.
The short answer. Nobody can tell you whether ChatGPT ads work as a category, because OpenAI states there are no performance benchmarks across advertisers, industries or campaign types, and the three figures being used to fill that vacuum each measure something narrower than the claim they are attached to.
That is not a reason to avoid the channel. It is a reason to distrust any page that answers this question with a number.
The three numbers
1.5x — Criteo, on referred traffic
What gets claimed: ChatGPT ads convert 1.5 times better than other channels.
What was actually measured: Criteo's aggregated data reports that traffic referred from LLM platforms including ChatGPT converts at roughly 1.5 times the rate of other referral sources.
Referred traffic. Not ad traffic. That figure describes what happens after someone arrives at a site from an LLM, and the overwhelming majority of LLM referral traffic in the period it covers came from organic citations rather than from ads, because the ad product barely existed. Quoting it as an ad performance figure attributes an organic result to a paid channel.
It is still a useful number. It says people arriving from an AI assistant are further along than people arriving from a typical referral, which is a reasonable prior for the ad surface too. It is not evidence about ads.
1.49x — Opascope, fifteen days, one account
What gets claimed: ChatGPT ads return about 1.5x.
What was actually measured: roughly 60,000 USD of spend returning about 89,000 USD over fifteen days, at a 1.72 USD CPC and a 2.35% conversion rate. One advertiser. One category. Fifteen days.
This is the most substantive public disclosure in the channel and Opascope deserves credit for publishing it. They did not present it as a benchmark. Everyone quoting it does.
Fifteen days is short enough that a single strong week moves the blended figure materially, and a single advertiser tells you about that advertiser's category, creative and landing page as much as about the platform.
0.91% — one client trial, against a 6.4% comparison
What gets claimed: ChatGPT ads underperform Google search badly.
What was actually measured: a reported client trial showing 0.91% CTR against a 6.4% Google search benchmark in the same sector.
Two problems, and they point opposite ways. It is one trial, so it carries the same single-account limit as Opascope's number. And the comparison is structurally awkward: ChatGPT Ads supports both CPM and CPC buying, through Reach and Clicks objectives, so a CTR figure means different things depending on which objective the campaign ran, and the reported trial does not say.
A low CTR in a conversational surface where the user is mid-reasoning rather than mid-scan may be a bad result or may be a category norm. There is no benchmark to say which.
The pattern
All three numbers get quoted as ChatGPT ad performance. None of them is.
| The number | What it actually measures | What it does not tell you |
|---|---|---|
| 1.5x conversion, Criteo | Quality of LLM-referred traffic, mostly organic | Anything about ad performance |
| 1.49x ROAS, Opascope | One advertiser's fifteen days | Whether any other advertiser sees this |
| 0.91% CTR, one trial | One advertiser's clickthrough, in a view-priced channel | Whether that CTR is good or bad here |
There is a fourth pattern worth naming, because someone else caught it first. Digital Applied has publicly flagged that a widely circulated set of vertical conversion benchmarks in this category are modelled estimates extrapolated from organic conversion rates rather than figures measured in an ad account. That correction is right, it is not ours, and the fact that it needed making tells you how the category's numbers are being produced.
What would count as an answer
Not rhetorical. The test is specifiable and cheap enough that someone with an account should run it.
A holdout, not a before-and-after. Same brand, same creative, same period, ad exposure varied across matched geographies. Before-and-after tests in a channel this new confound the ad effect with everything else changing.
Enough conversion events to distinguish signal from noise. At the 25 USD daily campaign minimum, a two-week test yields single-digit conversions for most businesses, which cannot separate a working channel from a broken one.
Organic measured throughout, not just paid. The most commercially relevant open question in this channel is whether paid placement lifts or cannibalises organic citation. The only public data point pointing either way is Seer Interactive's report that organic clickthrough rose 156% for a client brand while paid ran alongside and fell back when the paid placement was removed. One account, and it points at interaction rather than the independence OpenAI's Answer Independence principle asserts.
Published with the method, including the failures. A result without a stated method is a claim.
Nobody has published that test. Until someone does, "do ChatGPT ads work" has no category-level answer, and we are not going to manufacture one to have something to say.
What the honest answer looks like today
For a consumer brand in an approved category, in one of the seven serving markets, with a measured organic baseline and a budget large enough to produce interpretable conversion volume: the channel is worth a test, the auction is not yet dense, and you will be generating your own evidence rather than validating someone else's.
For everyone else, the reasons not to are in who should not buy ChatGPT ads yet, and they are structural rather than performance-related.
The people currently spending most confidently on this channel are the ones who have said out loud they are buying learning rather than returns. eMarketer's February 2026 reporting has Omnicom Media, Dentsu and WPP Media in the pilot at roughly $60 CPMs, with participation skewed toward brands buying influence and insight rather than performance. That is the honest posture available in this channel right now, and it is available to a smaller advertiser too, as long as it is budgeted and reported as what it is.
What we cannot tell you
Whether ChatGPT ads work. This is the page and that is the answer. No published evidence supports a category-level claim in either direction.
Whether they work for your category. No vertical benchmarks exist that were measured rather than modelled.
Whether paid lifts or cannibalises organic. One data point, one account, pointing at interaction. Not enough.
Our own results. We do not run campaigns and hold no advertiser account. Everything here is documentation and published reporting.
Quick answers
Do ChatGPT ads work? There is no category-level answer. OpenAI states no performance benchmarks exist across advertisers, industries or campaign types, and the three figures circulating as evidence each measure something narrower than they are quoted as measuring.
What is the ROAS on ChatGPT ads? The only published figure is Opascope's fifteen-day single-account result: roughly 60,000 USD spend returning about 89,000 USD, a 1.49x blended ROAS. One advertiser, one category, fifteen days, and not presented by Opascope as a benchmark.
Do ChatGPT ads convert better than other channels? Criteo's aggregated data reports LLM-referred traffic converting at roughly 1.5 times other referral sources. That measures referred traffic, most of it organic, rather than ad traffic, so it is not evidence about ad performance.
What is a good CTR for ChatGPT ads? Unknown. OpenAI publishes no benchmarks. One reported trial showed 0.91% against a 6.4% Google search benchmark in the same sector, but ChatGPT Ads supports both Reach and Clicks objectives, so a CTR figure is not comparable without knowing which objective produced it.
Are the ChatGPT ads vertical benchmarks reliable? A widely circulated set of vertical conversion benchmarks has been publicly flagged by Digital Applied as modelled estimates extrapolated from organic conversion rates rather than measured in an ad account.
Sources
| Claim | Source | Tier |
|---|---|---|
| No performance benchmarks across advertisers, industries or campaign types | OpenAI advertiser FAQ | Confirmed, primary |
| CPMs priced on views rather than clickthrough | eMarketer, 10 Feb 2026 | Reported |
| Omnicom Media, Dentsu, WPP Media in pilot at ~$60 CPMs; learning over ROI | eMarketer, 10 Feb 2026 | Reported |
| LLM-referred traffic converts ~1.5x other referral sources | Criteo aggregated data, via eMarketer | Reported, vendor |
| 15-day live account: ~60,000 USD spend, ~89,000 USD return, 1.49x ROAS, 1.72 USD CPC, 2.35% conversion | opascope.com | Reported, single account |
| 0.91% CTR against a 6.4% Google search benchmark in the same sector | Reported client trial | Reported, single account |
| Circulating vertical benchmarks are modelled from organic conversion rates | digitalapplied.com | Reported |
| Organic CTR rose 156% alongside paid, fell back on removal | Seer Interactive | Reported, single account |
| Minimum daily spend 25 USD per campaign | OpenAI advertiser documentation | Confirmed, primary |
Related reading
- Who is actually running ChatGPT ads? — named advertisers, agencies and partners
- How to measure ChatGPT ads — what the platform reports and what it does not
- Who should not buy ChatGPT ads yet
Changelog
15 August 2026. v1.0 published.
Field kit
Tools
Site
OpenAI (primary)
The InPromptAds Team
Writes about AI-native advertising and the shift from search to conversation.
