Written and reviewed by Bc. Petr Kučera, MSc., MBA
ChatGPT has shown ads in the US since 9 February 2026, and OpenAI opened self-serve buying in Europe on 31 August 2026. For this article, we collected 49 international cases published by companies and authors under their own names. They come from the US, other English-speaking markets, Europe, Brazil, Taiwan and India. Another 47 experiences come from advertiser discussions on Reddit, Hacker News and the OpenAI Community. That is how many cases we found, not the number of campaigns on the market.
For every number, we name who measured it, where, and whether they sell ChatGPT advertising. Our take is at the end.
We explain how PPC campaigns work in a separate guide, and what Czech advertisers have experienced in another article. This one looks only at results outside the Czech Republic.
OpenAI says ChatGPT Ads does not yet have performance benchmarks across advertisers, industries or campaign types, according to its help center. Every number below is one advertiser’s experience, not a norm.
The timeline
According to Similarweb, a total of 5,171 advertisers had run ads by 8 June 2026, and the number active each week grew from 2,585 to 3,770. The US accounted for 71.6% of impressions. Similarweb sells advertising data.
OpenAI recommends starting with a maximum bid of USD 3 to 5 per click (Ads in ChatGPT: the basics). Brooke Osmundson of Search Engine Journal points out that the industry has turned this recommended starting point into a “benchmark,” even though OpenAI has not published one (SEJ). In her words: “A USD 3 to USD 5 starting bid becomes a USD 3 to USD 5 ‘benchmark,’ even though OpenAI has published no such benchmark.”
StubGroup, an agency that sells ChatGPT advertising, reports a similar range. John Horn of StubGroup puts CPC at USD 2 to 5 across industries (MarTech), and Dario Roca typically sees clicks at USD 3 to 5 for clients outside his case study (StubGroup). That is one agency’s experience, not a price list.
Positive comparisons of ChatGPT ads with other channels come mostly from agencies and companies that sell ChatGPT ad management or a related tool. The four conversion rate comparisons with Google Ads all came out worse for ChatGPT (StubGroup). This reflects selection, not dishonesty: companies that sell the ads have a reason to publish results. The sample is therefore not representative, and the tests were mostly short.
Half of the 24 published international campaigns had a CTR below 1.25% (range 0.35% to 7.9%). In the 15 campaigns billed in dollars, half paid less than USD 4.41 per click; in Europe a click cost EUR 0.16 to 2.68. Conversion decides it, and in all four comparisons for the same advertiser ChatGPT converted worse than Google Ads. OpenAI publishes no benchmarks, and the positive numbers come mostly from those who sell ChatGPT advertising. Don’t treat someone else’s CTR as a plan; set your budget only once you see your own conversions.
Across 24 published international campaigns, half had a CTR below 1.25%, with a range of 0.35% to 7.9%. That is an unweighted median of published data points, not a typical market value (Synter).
Similarweb put the overall CTR in the pilot at 0.68%, with the top quartile at around 1%. We know this figure only from a slide reproduced by Search Engine Roundtable on 5 May 2026.
| Author | Market | CTR | CPC | Source |
|---|---|---|---|---|
| Andy Brice | GBP account | 7.9% (13 August) | GBP 0.10 | blog |
| Kalyvox | France | 4.83% | EUR 0.16 | JDN |
| Agência Koko | Brazil | 3.30% | BRL 0.74 | Koko |
| PromptMarketing (online store) | NL, DE, AT, FR | 2.41% | EUR 0.44 | PromptMarketing |
| Performics | US | 2.35% | USD 1.28 | JDN |
| Studio en Tête | France | 2.07% | EUR 0.55 | Studio en Tête |
| Ryan VanValin (small online store) | Not stated | 1.75% | USD 1.83 | |
| Umbau | Belgium | 1.2% | EUR 2.03 | Umbau |
| Workshop Digital | US | 1.1% | USD 3.14 | Workshop Digital |
| Common Thread Collective (online store) | Not stated | 0.94% | USD 4.41 | |
| Grow My Ads | Not stated | 0.92% | USD 13 | Grow My Ads |
| Octolens | US | about 0.8% | about EUR 3 | |
| Searchflow (three buying stages) | Netherlands | 0.73 to 0.96% | EUR 1.63 to 2.09 | Searchflow |
| Premium Online | Netherlands | about 0.75% | EUR 2.68 | Premium Online |
| Floyd Blaikie (B2B) | CAD account | 0.7% | CAD 9.29 | |
| Choice OMG | Canada | 0.65% | CAD 7.16 | Choice OMG |
| Kalyvox | US | 0.56% | EUR 3.33 | JDN |
| Synter | US, UK, CA, AU, NZ | 0.35% | USD 9.89 | Synter |
Currencies appear as each source reported them, without conversion. The median also includes data points outside the table: upGrowth in India (4.25%), growthhackers.tw in Taiwan (3.85%), two German campaigns, attractivemedia (2.94%) and ahead (2.0%), plus SE Ranking (1.30%) and Intelegencia (1.1%).
Andy Brice sells desktop software and does not sell ChatGPT advertising. From 5 to 28 August 2026, he spent GBP 289.52 and got 2,988 clicks and 14 installs, according to his 2 September summary. The 7.9% CTR comes from his 13 August article.
Performics (Publicis) ran a US pilot for a French travel advertiser: of a USD 250,000 budget, only USD 1,000 was spent. CTR was 2.35%, versus 12% on Google. According to the agency, reach was the problem.
Choice OMG (Peter Jaffray, Canada) got 58 clicks at a 0.65% CTR and a CAD 7.16 CPC, and not a single lead. According to the author, everything worked technically; the audience was the problem: “The mechanics worked. The audience did not.”
Synter (Joel Horwitz) builds technology for B2B advertising, meaning advertising aimed at other businesses. Across five markets, the US, UK, Canada, Australia and New Zealand, it logged 129,663 impressions at a 0.35% CTR. CPC was USD 9.89 (USD 10.62 in the US), and CPM came to USD 34.16. CPM was nearly the same across markets, so CTR determined CPC. Synter sells ChatGPT advertising.
Workshop Digital compared context hints, the topics you attach to an ad. Keyword hints reached a 1.2% CTR; conversational hints, 0.9%.
Four of the five highest CTRs in the set come from outside English-speaking markets: Kalyvox in France (4.83%), upGrowth in India (4.25%), growthhackers.tw in Taiwan (3.85%) and Koko in Brazil (3.30%). Without Brazil, Taiwan, India and Germany, 19 data points remain, with a median of 1.1%. The data does not tell us reliably why.
Kalyvox (Alexandre Bougra) sells B2B SaaS, software as a service for businesses, and advertised the same product in two countries. In France, it paid EUR 0.16 per click at a 4.83% CTR; in the US, EUR 3.33 at 0.56%. The bounce rate was high, meaning visitors left almost immediately, and in the author’s view a cheap click did not mean quality traffic: “Le CPC était excellent. Le trafic beaucoup moins.”
Across 15 dollar-billed campaigns, mostly in the US, CPC ranged from USD 0.76 to 14.8, and half were below USD 4.41. Again, that is an unweighted median of published data points (Opascope). Ten European campaigns paid EUR 0.16 to 2.68 per click (Premium Online). We do not give a European median because many sources report only a range.
Selected dollar-billed campaigns, sorted from lowest CPC
| Advertiser | CPC |
|---|---|
| StubGroup (case study) | USD 0.76 |
| Performics | USD 1.28 |
| Opascope | USD 1.72 |
| Ryan VanValin | USD 1.83 |
| Whole Whale | USD 3.02 |
| Workshop Digital | USD 3.14 |
| Common Thread Collective | USD 4.41 |
| InterTeam Marketing | about USD 5 |
| Q1Media | about USD 5 |
| Symphonic Digital | about USD 9 |
| Nakora (US) | USD 9.4 |
| Synter (US) | USD 10.62 |
| Grow My Ads | USD 13 |
| Nakora (enterprise case) | USD 14.8 |
European campaigns:
The data does not show that Europe is cheaper across the board. Kalyvox paid less in France than in the US for the same product, and Andy Brice’s GBP account paid GBP 0.10 a click. Neither justifies a generalization.
Symphonic Digital tested in the US with the agency’s own budget. At a USD 3 bid, the ads got no impressions at all. Total spend came to about USD 500, at a CPC of roughly USD 9.
Karel Huk, a Czech who manages a US account, ran 30 campaigns across 30 metro areas with a target CPA of USD 2.50. CPA is cost per acquisition. Together, they got a single impression. After he raised the target CPA to USD 20, delivery grew 523-fold. These are US figures.
According to OpenAI, the bid_too_low warning is guidance, not a cap on delivery. The threshold varies with the market and the competition. Andy Brice got delivery at bids of GBP 0.07 to 0.15, and Kalyvox in France at around EUR 0.10.
At Synter, a single topic accounted for 74% of US impression volume, while another ad group in the UK got one impression in 9 days. As the author puts it, no bid can create a conversation that is not happening: “No bid fixes a conversation that is not happening.” Delivery can also come in bursts. Synter saw a jump of 45,419 impressions in one day, and Enablers got thousands of impressions an hour, then nothing.
Alban Peltier, CEO of the French company Stelleo, told Journal du Net how quickly a campaign can go live. He signed up on Thursday, 3 September, and less than a week later the first small invoices arrived: the campaign was running while the company was still being verified. The first invoice was how he learned it had gone live. His domain was .fr, yet the console suggested targeting the US, and there was no way to target a city or region or to set time windows. In his words, the tool is “not in beta, more like alpha.”
The most telling comparison uses the same advertiser, period and metric in both channels. Conversion rate is the share of clicks ending in a conversion. Four such comparisons have been published.
| Author | ChatGPT | Google Ads | Sells ChatGPT ads |
|---|---|---|---|
| Andy Brice | 0.46% | 5.3% | No |
| John Horn, StubGroup | 0.21% | 3.71% | Yes |
| Falia (Canada, B2B) | 0.9% | 5.95% | Not stated |
| Dario Roca, StubGroup (case study) | 0.8% | 6.64% | Yes |
Google had the higher conversion rate in all four. Two were published by StubGroup, and the sample does not represent the market. Andy Brice’s paid ChatGPT visits lasted 7 s, versus 20 s from Google. In MarTech on 11 August 2026, Horn put engagement at 17 s on ChatGPT versus 41 s on Google.
StubGroup’s case study of one B2B compliance client over 30 days shows why one metric is not enough. CPC was USD 0.76 on ChatGPT versus USD 10.46 on Google, and CPL, the cost per lead, was USD 92.21 versus USD 157.57. According to the agency, however, Google leads turned into customers “at a significantly higher rate,” so Google’s cost per customer was lower.
Other comparisons look at cost per lead, MQLs (marketing-qualified leads) or CPC. The results are mixed, and different metrics cannot be pooled.
The positive comparisons come mostly from InterTeam, Nakora, Intelegencia and StubGroup, which all sell ChatGPT advertising. At StubGroup, cheaper leads did not mean cheaper customers; for Porrini, they did not bring in revenue.
Obility, a US B2B agency, shared its numbers in a Reddit post, without a direct comparison to Google. Over four months: CTR of 0.5 to 1.0%, CPC of USD 3 to 8, CPM of USD 25 to 45, conversion rates of 3 to 6% and a cost per MQL of USD 120 to 250. The agency sees the channel as a way to build awareness.
Andy Brice: paid ad clicks converted at 0.46%, while unpaid visits from ChatGPT recommendations converted at 4.2%. Paid visits lasted 7 s, unpaid ones 23 s.
Octolens (Charlotte Schmitt) makes software that tracks mentions on social media. Over two weeks, in the US only, on a euro account with a daily budget of EUR 50, it got about 25,000 impressions and just under 200 clicks at a CPC of roughly EUR 3. It spent a little over EUR 600 and got no sign-ups from those clicks, verified using the pixel, server-side events and its own analytics. Over the same period, unpaid ChatGPT recommendations brought 10 sign-ups. Octolens sells brand monitoring, including in AI, so the conclusion favors its own offer.
Whole Whale saw little difference. Paid visits lasted 1 min 22 s with 36% engagement; unpaid visits, 1 min 35 s with 40%. Platform clicks roughly matched sessions in GA4 (Google Analytics 4).
Q1Media, a US agency, reports across several accounts a CTR of 0.5 to 2.5%, a CPC of about USD 5 and a conversion rate of about 2%, roughly half that of unpaid ChatGPT traffic. Q1Media sells ChatGPT advertising.
Does paying for ads get you mentioned in ChatGPT’s answers? The data on this is thin.
We cover how to show up in ChatGPT’s answers in a separate article.
ROAS, return on ad spend, is revenue divided by ad spend. The online stores that published results report a positive ROAS, but it rests on platform or tool attribution, not verified revenue. PromptMarketing says so explicitly.
Agência Koko in Brazil advertised an info product from 7 to 14 August 2026. It spent BRL 3,690.85 for 151,334 impressions and 4,996 clicks: a 3.30% CTR and a BRL 0.74 CPC. Contracts came to BRL 4,936.08, a return of 1.34, while cash collected in the period was BRL 1,875.55, about BRL 0.51 collected for every BRL 1 spent. The dashboard showed zero conversions, yet there were three sales: the oppref parameter did not carry over to the third-party Hotmart checkout. Koko sells ChatGPT advertising. The dashboard suggested a loss; contract value beat ad spend, while cash fell short of it. Profit after other costs cannot be determined.
The gaps run in both directions. In OMR, Carlo Siebert described a US client, a B2B company, that spent about USD 10,000 and, in his assessment, found that it did not pay off. OpenAI’s reporting showed 36 conversion events, yet the company had only 2 inquiries, and those demonstrably came from outside ChatGPT. At Koko it was the reverse: zero in the dashboard, three actual sales. OpenAI notes that its reporting may include modeled conversions.
Clicks and visits may not match, either
A gap between clicks and visits is not, on its own, proof of fraud. For Brice and Whole Whale the numbers matched, which shows how much the conversion tracking setup matters. At PromptMarketing, 96% of clicks were on mobile; at Kalyvox, about 95%. We cover server-side conversion tracking in a separate article.
Most published tests lasted days or weeks. The longest published test with a cost curve, by Intelegencia, lasted six months. Costs improved after a change in approach, not simply over time.
Whether context hints work the same way everywhere is still open. In the MarTech test and at Workshop Digital, keywords got more delivery than full questions. The available data does not show whether this holds generally.
The numbers below are unverified claims from anonymous Reddit accounts, not benchmarks.
Ten forum accounts reported CTRs from 0.57% to 3.5%, with half at 1% or below. Thirteen reported CPC, most often USD 3 to 5, with a median of USD 4.2. These are anonymous claims on Reddit.
ChatGPT ads are currently a test for specific offers, not a replacement for search.
Two facts help explain this. OpenAI’s Ads in ChatGPT page says ads are shown to Free and Go users and to logged-out users, while Plus, Pro, Business, Enterprise and Edu stay ad-free. And advertisers do not see the specific queries their ads appeared for, only aggregate metrics.
In our reading of the data, positive results are reported mainly by online stores (based on platform or tool attribution) and by companies whose offer solves a specific problem; R3 saw results improve only after rebuilding its ads around specific services. Testing makes sense for products or services that people actually research and compare in ChatGPT, and for companies with budget to spare beyond the channels that already work. Workshop Digital acknowledges that ChatGPT may play a larger role in longer buying decisions than a short test can show.
Andy Brice, who compared Google, ChatGPT ads and ChatGPT recommendations, summed up his experience on 13 August 2026: “It feels like shovelling money into the void.” Our position: ChatGPT does not replace Google Ads; it complements it for some offers. We recommend funding the first test from a test budget. We cover how search campaigns fit into the mix in our article on PPC advertising in 2025.
For the launch itself, follow our step-by-step guide.
See what we do in PPC on our PPC services page.
There is no official average, and OpenAI has no benchmarks. In 24 published international campaigns, CTR ran 0.35% to 7.9%; half were below 1.25% (unweighted median of published data points).
In 15 published dollar-billed campaigns, mostly in the US, CPC ranged from USD 0.76 to 14.8, with half below USD 4.41 (unweighted median). Ten European campaigns paid EUR 0.16 to 2.68 per click. OpenAI recommends a starting maximum bid of USD 3 to 5, but that is not a price list.
In four conversion rate comparisons where the same advertiser measured both channels over the same period, Google came out ahead every time. Andy Brice saw 0.46% on ChatGPT versus 5.3% on Google. Comparisons on other metrics are mixed, and the positive ones come mostly from companies that sell ChatGPT advertising.
Online stores that published results report ROAS starting at 1.49× (Opascope); Common Thread Collective reports 3.3× to 6.8×, depending on the attribution model. That is platform or tool attribution, not verified revenue, and most of these results come from agencies that sell ChatGPT advertising.
The gap runs both ways. Koko’s dashboard showed zero conversions against three actual sales; Siebert’s reporting showed 36 conversions against 2 real inquiries, demonstrably from outside ChatGPT. OpenAI says its reporting may include modeled conversions.
Andy Brice saw a 4.2% conversion rate from recommendations versus 0.46% from ads. In a two-week, US-only test, Octolens got 10 sign-ups from recommendations over the same period and none from ads; note that Octolens sells social media mention tracking. Whole Whale saw little difference.
Not sure whether a ChatGPT ads test makes sense for your business? Contact us and we’ll review it together before you launch.
Contact us
About the author
Bc. Petr Kučera, MSc., MBA
CEO & Performance Strategist, PPC ONE
Petr discovered the power of performance campaigns back in 2011, and they have been an inseparable part of his professional life ever since. His edge comes from working with more than a hundred companies.
Data verified as of 29/09/2026.
80 client reviews