ChatGPT ads in Canada: results from our B2B test
Partner · Strategy, advertising and measurement
We tested ChatGPT ads for B2B as soon as they reached Canada, on two campaigns and for the same client we run on Google Ads. The result: a 0.9% conversion rate in ChatGPT versus 5.95% on Google Ads, a gap of more than six times. The two other B2B tests published since point the same way. The format is not the issue: the ads only reach Free and Go accounts, where your decision-makers are rare.
- Our B2B test converted at 0.9% in ChatGPT versus 5.95% on Google Ads: same client, same period, same conversion tracking.
- Two US agencies published their own B2B tests in late summer 2026. Neither got a single qualified lead, from close to 1,000 clicks combined.
- Ads only show on Free and Go accounts. An executive working in their company's Business or Enterprise account never sees your ad.
- Clicks cost less than on Google ($3 to $6 US in the published B2B tests), but they produce far fewer leads.
- Measurement has caught up in part: a pixel and a Conversions API since May, plus attribution partners and early incrementality work announced on October 5, 2026.
- Treat this channel as a small-scale learning budget, not a substitute for the campaigns already driving your sales.
On this page
Our B2B test in Canada: ChatGPT versus Google Ads
What follows is for executives and marketing managers weighing a budget for ads in ChatGPT, especially in B2B. It is not aimed at low-ticket consumer campaigns, where the math is different. It does not replace a test run on your own data.
We launched two B2B campaigns in OpenAI's self-serve ads manager in spring 2026, shortly after Canada opened. To compare fairly, we measured the conversion rate with the same tracking as our Google Ads campaigns, over the same period and for the same client. The exact sectors and budgets remain confidential.
| Our B2B test, spring 2026 | Google Ads | ChatGPT ads |
|---|---|---|
| Client | The same B2B client | The same B2B client |
| Period | The same period | The same period |
| Conversion tracking | The same tracking | The same tracking |
| Conversion rate | 5.95% | 0.9% |
The verdict is clear: a gap of more than six times. We knew the ads only reached free users, and we wanted to test anyway. The numbers confirmed the theory. What follows explains why this outcome was predictable, and what you can still draw from the channel without risking the budget that already drives your sales.
Many companies will open a budget on this channel, compare it to their Google search campaigns, and conclude too fast that it does not work. The real risk is not losing the small stake of the test, it is judging a young visibility channel by the yardstick of a mature conversion channel, then cutting before you have understood who it speaks to. Bad measurement costs more than the spend itself.
What other published B2B tests show
Our test is no longer alone. Two US agencies published the results of their own B2B campaigns in late summer 2026, paid for with their own money. Neither got a qualified lead.
| Test | Spend and clicks | Cost per click | Result |
|---|---|---|---|
| Falia, Canada, spring 2026: two B2B campaigns, same client as on Google Ads | Confidential | Confidential | 0.9% conversion versus 5.95% on Google Ads |
| Workshop Digital, United States, published August 26, 2026: marketing services, manufacturing, financial services | $2,319 US, 739 clicks | $3.14 US | No qualified lead, versus two from Google Search |
| Windmill Strategy, United States, published September 2, 2026: industrial manufacturing | $1,470 US in one month, 259 clicks | $5.68 US | No conversion |
Clicks cost less than on Google: at Workshop Digital, $3.14 US versus $6.76 US on Google Search for the same advertiser. Yet Google Search produced the only two qualified leads of their test, with a 5.2% click-through rate versus 1.1% in ChatGPT.
The click-through rate of these B2B tests (1.1% at Workshop Digital, 0.66% at Windmill Strategy) is in line with the average Similarweb measured across all advertisers in the spring:
ChatGPT ads aren't a science experiment.
Overall CTR: 0.68%.
Top quartile: 1%.
Best brands: 1.57%.
Peak: 5.4%.
For comparison: display sits around 0.35%, podcast 0.5–1%.
A click-through rate twice that of a display banner says nothing about the sale. In B2B, all three published tests show it: the click comes, the demand does not follow.
What is an ad-exposure tier
An ad-exposure tier is the subscription level that determines whether or not a user sees ads in a service. In ChatGPT, only the Free and Go tiers display ads; the higher tiers show none. It sets, before any targeting, the population a campaign can actually reach. On a platform where decision-makers pay not to see ads, this tier decides the reach of a budget far more than the targeting itself.
How ads in ChatGPT work in October 2026
Ads appear below the end of an answer, labeled as sponsored and kept separate from ChatGPT's text. At launch, a single ad showed per answer. That is no longer the case: one or more units can appear, each featuring one or more items from one or more advertisers. The product-feed carousel arrived in August. OpenAI will also test a visual format shown during image generation in the United States later in October.
Targeting relies on context hints: plain-language descriptions of the conversations where your offer is useful, with no exact match and no keywords in the Google Ads sense. Country and geographic area, platform and custom audiences have since been added. Eligible advertisers can also exclude phrases. No job-title or company targeting appears among these options. During Windmill Strategy's test, a custom audience required at least 25,000 matched users: few B2B account lists reach that threshold.
According to OpenAI, the system picks the ad based on expected relevance and outcomes. It reads the conversation context, the landing page, the title, the copy and your hints. If the user has left personalization on, it also reads signals from their other chats and memory. The bid matters when several ads are eligible, but relevance weighs just as much. Workshop Digital got a better click-through rate with hints written as specific keywords (1.2%) than with conversational phrases (0.9%).
Buying is by the thousand impressions (CPM) or by the click (CPC). According to OpenAI, CPC and outcome-optimized bidding now make up the majority of campaigns. The self-serve Ads Manager is open to businesses in Canada, France and Switzerland: the legal entity that advertises and gets billed must be based in a listed country.
Why the audience decides everything in B2B
The ad-exposure tier is the key to reading the channel. Ads only show on Free and Go accounts, the latter at $8 US a month in the United States. Plus, Pro, Business, Enterprise and Edu accounts see none. Sam Altman announced it this way in January:
We are starting to test ads in ChatGPT free and Go (new $8/month option) tiers.
Here are our principles. Most importantly, we will not accept money to influence the answer ChatGPT gives you, and we keep your conversations private from advertisers.
It is clear to us that a lot of people want to use a lot of AI and don't want to pay, so we are are hopeful a business model like this can work.
(An example of ads I like are on Instagram, where I've found stuff I like that I otherwise never would have. We will try to make ads ever more useful to users.)
OpenAI's CEO says it himself: the ad model targets people who want a lot of AI without paying. In B2B, this is a deep problem. An executive working in their company's Business or Enterprise account never sees your ad. You mostly speak to people looking for information, rarely to the people who make the buying decision.
The pool shrinks further. A free user can remove ads by accepting fewer messages per day. ChatGPT shows none in temporary chats or to people under 18. Yet OpenAI lists "choosing software for their business" among the uses an ad can support. The need is real: it is the person who carries it who is missing when they work in a paid account.
The imbalance goes beyond B2B. An audience defined by an unwillingness to pay for a subscription leans toward price-sensitive profiles. Someone who will not spend a few dollars a month on a tool they use every week is not the typical prospect for a high-purchasing-power offer. For a management team, the consequence is direct: on this channel, the audience constrains returns before targeting even comes into play, and no budget buys back an audience that is not there.
Where the channel shows signs of life is in online retail, on products one person buys alone. The agency Common Thread Collective published this result in July:
34 days on ChatGPT Ads. One account:
$7/day → $1,005/day. 142x spend growth.
~5x ROAS. $4.41 CPC. Both attribution models positive.
High-AOV ($1,000+ AOV) product. The audience on ChatGPT isn't scrolling. They're asking.
A product over $1,000 US that a single person decides to buy: the opposite of a B2B purchase where several managers decide over weeks. The results OpenAI highlighted on October 5, 2026 (WeightWatchers, Dose, Portland Leather) are all consumer brands. Our overall position on this channel, and what we recommend instead, is in why we advise against ads in ChatGPT.
What a ChatGPT ad costs in B2B
OpenAI does not publish a price list: the cost comes out of the auction. The published tests still give benchmarks. In B2B, a click cost $3.14 US at Workshop Digital and $5.68 US at Windmill Strategy, versus $6.76 US on Google Search for Workshop Digital. In online retail, Common Thread Collective reports $4.41 US.
OpenAI states there is no minimum spend. Yet Windmill Strategy could not set a campaign below $25 US a day: a thirty-day test therefore takes at least $750 US per campaign. Our own budgets remain confidential.
The math that matters is elsewhere. A $3 click that produces no lead costs more per sale than a $6 click that does. Always compare the cost per qualified lead, never the cost per click alone.
What you can measure, and what is missing
At launch, advertisers only received impressions and clicks. In March, SEO consultant Glenn Gabe summed up what the first advertisers were reporting:
Low tech, no performance data, and limited targeting. Wow, sounds amazing :) -> Advertisers that bought ChatGPT's first ad campaigns say the process was low tech and that they haven't received much data showing if their ads worked
"Two executives at agencies working with early ChatGPT advertisers said they haven’t yet been able to prove the ads have driven any measurable business outcomes for their clients."
"OpenAI hasn’t yet offered marketers any automated way to buy ad space. Buyers have had to rely on making phone calls and sending spreadsheets and emails to OpenAI representatives, one ad executive said. More importantly, advertisers found it hard to tell whether the ads were paying off." https://www.theinformation.com/articles/openais-first-advertisers-prove-chatgpt-ads-work?rc=mzkwis
That is no longer true. Since May 2026, a pixel and a Conversions API track conversions on the browser side and the server side. That is what let us compare conversion rates between the two platforms rather than stop at the click. On October 5, 2026, OpenAI added attribution partners (including AppsFlyer, Triple Whale and Northbeam) and integrations that send conversions from your existing systems (Hightouch, Tealium, LiveRamp).
Three things are still missing. Incrementality measurement (the sales advertising actually causes) is in its early days: OpenAI is exploring geographic experiments with Haus, Measured and WorkMagic. Brand safety is only verified by DoubleVerify and IAS in a pilot. And the results OpenAI publishes are all consumer: no B2B case among them. The difference between an attributed sale and a sale caused by advertising remains, and we cover it in our article on incrementality tests.
One nuance deserves naming, because it qualifies our own result. Visitors referred by AI platforms convert, according to Criteo, about one and a half times better than other referral channels: the conversation does a qualification job that a query does not. The traffic is rare, but when it clicks, it is often more engaged. Our B2B results do not reflect this effect, precisely because the audience we were after was not there.
Advertising in ChatGPT in Québec
OpenAI's documentation does not address Québec specifically. Three points matter for an advertiser here.
- Language. According to OpenAI, an ad's language may differ from the conversation's: the system considers location, settings and conversation language. Plan a French version of every ad and its landing page, as for any advertising in Québec.
- Territory. During Windmill Strategy's test, country was the only level of geographic targeting. OpenAI now announces geographic-area targeting: check in the ads manager how granular it gets before building a campaign aimed at Québec only.
- Consent. OpenAI's pixel tracks your visitors on your site. Like the Google or Meta pixel, it must go through your consent banner to comply with Québec's Law 25.
What to do right now
Our recommendation comes down to five points, all aimed at protecting your main budget.
Choose the offer that speaks to a free audience. Put forward accessible offers, consumer products, useful content. Avoid pushing a high-purchasing-power sale there. In B2B, accept that you are mostly speaking to people looking for information, not to decision-makers.
Test small, with decision thresholds set in advance. Treat this channel as a learning budget, not a substitute for the campaigns already delivering. Give the test at least thirty days and write your criteria before launch, for example: scale up if the cost per acquisition stays under one and a half times the Google Ads benchmark, cut if it goes past double.
Set up measurement before the first dollar. Conversion pixel, Conversions API and a dedicated segment in your analytics tool. The channel now has the tools to be judged on sales; without them, you will judge on the click, and the click of this format almost always disappoints next to search.
Say in the ad who it is for. With no job-title or company targeting, the copy does the filtering. "Digital marketing for B2B manufacturers" screens out bad clicks better than "Improve your marketing." Write your context hints as specific keywords and put the essentials in the first words, which the display sometimes cuts off.
Judge the channel on what it can deliver today. Visibility during the research phase and a little qualified traffic, not a direct conversion engine. Compare its returns to your visibility campaigns, not to your brand campaigns on Google.
What to settle before opening a budget on this channel
The test spend is small, but the real cost is management time and the risk of drawing the wrong conclusion. These five questions get settled before you commit the first dollar.
- Is your typical client on a free tier, or do they pay to stop seeing ads?
- What learning budget are you willing to lose without it touching your profitable campaigns?
- Is conversion measurement in place before launch, or will you judge on the click?
- Against which benchmark do you compare returns: your brand campaigns, or your visibility campaigns?
- If the cost per acquisition goes past double your benchmark after thirty days, what do you stop, and who decides?
A good answer names a client, a tier, a cost threshold and a decision. A vague answer talks about being present on AI. A vendor who pitches you this channel without asking where your buyer is is selling a presence, not acquisition.
Deciding whether this channel has a place in your acquisition, and at what budget, is exactly the kind of question we settle in a paid audit.
The companies that lose money on this channel do not lose it in the bid, they lose it in the comparison. They pit a young visibility channel against their best search campaigns, see the gap, and cut without having learned anything. The ones that get something out of it do the opposite: they set a small budget, a visibility benchmark and a cost threshold in advance, they put measurement in place, and they read the result for what it is. Our own test confirmed the theory rather than disproving it, and it cost us a learning stake, not a share of the budget that keeps our clients alive.
The overall view of paid channels is in how to choose an advertising platform.
Steering an acquisition budget on what it produces, channel by channel, is at the heart of the Optimize the profitability of your digital campaigns goal.
Already running your campaigns and want backup to frame this channel? See our work in paid advertising.
Frequently asked questions about ChatGPT ads for B2B
Are ads in ChatGPT available in Canada?
Yes. Ads have shown to Canadian users since spring 2026, and OpenAI's self-serve ads manager accepts businesses based in Canada, as well as in France and Switzerland. That is what let us run our own B2B tests there.
Do ChatGPT ads work for a B2B company?
Poorly, for now. Our B2B test converted at 0.9% in ChatGPT versus 5.95% on Google Ads, for the same client. The two other B2B tests published in 2026 got no qualified leads. The channel works better for consumer purchases that one person decides alone.
ChatGPT ads or Google Ads: which brings more B2B leads?
Google Ads, by a wide margin. In our test, Google Search converted six times better for the same client. At Workshop Digital, it produced the only two qualified leads of the test, with a more expensive click ($6.76 US versus $3.14 US in ChatGPT).
How much does a ChatGPT ad cost?
OpenAI publishes no price list: the cost comes out of the auction, per click or per thousand impressions. In the published B2B tests, a click cost $3.14 to $5.68 US. Windmill Strategy could not set a campaign below $25 US a day, so at least $750 US for a thirty-day test.
Do executives see ChatGPT ads?
Rarely. Only Free and Go accounts show ads. Plus, Pro, Business, Enterprise and Edu accounts show none, so an executive working in their company's paid account never sees your ad.
Can you target an industry or a job title in ChatGPT?
No. Targeting relies on context hints (plain-language descriptions of the conversations you want), country and geographic area, platform and custom audiences. There is no job-title or company targeting. The ad copy has to say who it is for.
Can you advertise in ChatGPT in Québec only?
Possibly. OpenAI announces geographic-area targeting, whereas country was the only level available during Windmill Strategy's test in late summer 2026. Check in the ads manager how granular it gets before building a Québec campaign. Plan a French version of the ad too.
Can you measure conversions from ChatGPT ads?
Yes. A pixel and a Conversions API have existed since May 2026, and OpenAI added attribution partners on October 5, 2026. Incrementality measurement is in its early days. You can judge the channel on sales, with a dashboard that is still incomplete.
- OpenAI, Testing ads in ChatGPT, official announcement, February 2026, accessed July 2026. Source of the February 9, 2026 launch date.
- OpenAI, Ads in ChatGPT, official documentation accessed October 2026. Source of which accounts see ads, the free ad-free option, the number of units per answer, how ads are selected and ad language.
- OpenAI, Ads Manager Availability, official documentation accessed October 2026. Source of self-serve access in Canada, France and Switzerland.
- OpenAI, New ways to buy ChatGPT ads, official announcement, May 2026, accessed July 2026. Source of the self-serve manager, cost-per-click bidding and measurement through the pixel and Conversions API.
- OpenAI, A milestone in expanding access to AI, official announcement, August 31, 2026, accessed October 2026. Source of bidding types, geographic-area targeting, custom audiences and the business software example.
- OpenAI, Building advertising for the way people use AI, official announcement, October 5, 2026, accessed October 2026. Source of the 1.2 billion weekly users, the visual format, measurement partners, incrementality work, phrase exclusions and the advertiser examples.
- OpenAI, Introducing ChatGPT Go, official page accessed July 2026. Source of the Go tier price of $8 US a month.
- Workshop Digital, ChatGPT Ads: What We Learned From Our First B2B Test, August 26, 2026, accessed October 2026. Source of the spend, clicks, click-through rates, costs per click and qualified leads compared with Google Search.
- Windmill Strategy, Our First 30 Days with ChatGPT Ads, September 2, 2026, accessed October 2026. Source of the industrial manufacturing test, the observed daily minimum and the custom-audience threshold.
- Criteo, update on the integration with OpenAI, official press release, 2026, accessed July 2026. Source of the higher conversion rate of visitors referred by AI platforms.
Gabriel almost always takes your first call and carries out your audit. He builds the strategy starting from your growth goal: where to put your budget, which market to test and how to connect each lead to a real sale in your CRM. He mainly leads engagements for three goals: Optimize the profitability of your digital campaigns, Develop a new market, and Generate demand and growth. With Geneviève, he also works on organic search (SEO), AI visibility (GEO) and conversion rate optimization (CRO). The sales a Google Ads or Meta Ads campaign brings in depend on the page that receives the click. He writes mainly about marketing strategy, paid advertising and measurement.
About Falia →
SEO
Paid advertising
Conversion
AI visibility
Web design