Google Ads vs ChatGPT Ads in Canada: What the Real Data Shows

01The rules this article follows

You asked for real numbers. So the method needs stating before the findings, because the temptation in this particular comparison is enormous — one channel has no published benchmarks, which makes inventing a tidy table almost effortless.

One consequence worth stating plainly: this article cannot tell you what ChatGPT Ads will do for your Canadian business. Nobody can, including OpenAI. What it can do is show you the observed range, the structural constraints that explain the range, and the arithmetic you would need to clear before spending. That is a narrower promise than most agency content makes, and a far more useful one.

02Why only one side has benchmarks

This is the asymmetry that governs the comparison, and it is documented rather than assumed. OpenAI's own guidance states that ChatGPT Ads "does not yet have performance benchmarks across advertisers, industries, or campaign types," and that there is "no recommended bid amount at this time" for conversion-optimised bidding. The Ads in ChatGPT help documentation likewise frames the product as early.

Google Ads, by contrast, has been benchmarked continuously for a decade. The WordStream / LocaliQ benchmark series is the standard reference, and the 2026 cross-industry medians most agencies quote descend from it.

That last row is the honest bottom line for a Canadian reader: you are comparing a channel that has existed in this country for roughly seven months against one with two decades of accumulated evidence. A fair comparison of performance is not yet possible; a fair comparison of constraint, cost and risk is.

03The two channels measure different things

Before any number, the unit. This is where most ChatGPT-versus-Google comparisons quietly mislead, including several published ones.

Impression

On Google, a served search result. On ChatGPT, an ad-bearing reply — one unit appended below a completed answer. Some panels count per reply, others per impression; the two produce different CTRs from the same campaign.

Intent moment

Google captures a declared query at the point of decision. ChatGPT captures a conversation, often earlier in the decision, sometimes after the assistant has already answered the question.

Click meaning

A Google click normally continues a search. A ChatGPT click often follows an answer the user already received — so intent can be weaker, or further along, depending on the prompt.

Attribution window

Google has two decades of modelled attribution. ChatGPT's pixel and Conversions API arrived in May 2026, and one agency reports that a majority of attributable conversions land outside the immediate click window.

The consequence is practical rather than academic. A 1% CTR on ChatGPT and a 6.6% CTR on Google are not the same measurement of the same behaviour, which is why the gap should not be read as a quality verdict. It should be read as a price input: lower CTR with a comparable CPC means you are paying more per impression and less per click than you may assume. The arithmetic in section seven handles that properly.

04Every published ChatGPT Ads test we could verify

This is the useful dataset, and it is smaller than most articles imply. Below is every real-spend ChatGPT Ads test with published figures we could trace to a named publisher in 2026. Dashes mean the metric was not disclosed — not that it was zero.

Read that table honestly and three things stand out, none of which favour either platform wholesale.

The range is not a range, it is two populations. Three of six tests with conversion data produced zero or "very few" recorded outcomes. One produced 1.49× ROAS. Averages across this set would be meaningless, and any agency presenting an average is either careless or selling.

The best result is the most expensive test. The only documented return-on-ad-spend figure came from roughly $60,000 of spend over fifteen days. That is not a coincidence: at that spend the advertiser could accumulate conversion volume fast enough for the system to learn. A CA$66.75 test cannot.

Same-advertiser comparisons are the only fair ones. Workshop Digital ran both channels and published both sides: 1.06% CTR at $3.14 CPC on ChatGPT, 5.2% CTR at $6.76 CPC on Google Search, with zero marketing-qualified leads on the former and two on the latter. That is one agency's own budget — but it is a controlled comparison, and it is more than most published claims can say.

Two figures from the same tests deserve separate mention because they cut against the "ChatGPT traffic is worthless" reading. GravityKit measured post-click behaviour and found ChatGPT Ads traffic averaged 5 seconds engaged time and 1.0 pages per session, against 65 seconds and 4.8 pages for Google Ads and 40 seconds and 3.5 pages for organic ChatGPT traffic. Meanwhile Criteo, a launch partner, reported across a sample of 500 US retailers in February 2026 that LLM-referred traffic converted at roughly 1.5× the rate of other referral channels. Both can be true: paid ChatGPT clicks arrive less engaged, while organic AI referrals arrive more qualified. That distinction matters enormously for how you plan — and it is the subject of our analysis of how ChatGPT Ads matches intent.

05The Canadian data that exists — and its limits

Canada went live on 16 April 2026, in OpenAI's first expansion outside the United States. Seven months later the public Canadian performance record consists of two tests and a set of platform-level facts. Here is all of it, without embellishment.

Test one: Choice OMG, Ontario and Alberta, 8–25 June 2026

A Canadian agency ran two CPC-bid lead-generation campaigns (Ontario and Alberta) plus one CPM-bid campaign, pointing at its own AI marketing service page, with the pixel and Conversions API wired through Google Tag Manager and its own server. Account-level results: 8,940 impressions, 58 clicks, 0.65% CTR, CA$7.16 average CPC. The CPM-bid campaign delivered 4,293 impressions at a 0.26% CTR, working out to CA$18.18 per click — roughly four times the CPC-bid price for the same destination. Lead forms submitted: none. Two CPC campaigns in June separately averaged CA$4.42 and CA$4.88 per click.

Their own diagnosis is the most valuable sentence in the entire Canadian record: "the more someone looks like our ICP, the more likely they're on a paid, ad-free plan." That is the eligibility constraint explained by someone who paid to discover it.

Test two: PINTAYA, CA$ account

2,100 impressions, 17 clicks, CA$66.75 spend, 0.81% CTR, CA$3.93 average CPC, CA$31.79 effective CPM, with five conversions recorded. Split by ad, the lead-generation creative outperformed the ROI creative on every metric (0.85% vs 0.74% CTR; CA$3.00 vs CA$6.94 CPC; CA$25.42 vs CA$51.63 CPM). At a CA$3.93 CPC and a hypothetical 5% landing-page conversion rate, cost per lead would be about CA$78.60 — a projection, not a result.

What else is documented about Canada

• Penetration: independent tracking reported for the trailing week ending 3 July 2026 put ChatGPT ad penetration at 53.6% of Free and Go traffic in Canada versus 51.0% in the United States. Canadian users were seeing sponsored units at a higher rate than users in the market OpenAI piloted first.
• Scale: Canada represented about 5.4% of ChatGPT's 900 million weekly active users per Similarweb, as reported in April 2026 — a directional figure, not an addressable-audience count.
• Platform revenue: Reuters reported roughly US$100 million booked in the pilot's first six weeks, with 600+ advertisers, and OpenAI reported a US$1 billion annualised run rate on 31 August 2026. Neither figure is Canadian.
• Pricing: CPMs fell from a US$60 launch level to as low as US$25 by the time of Digiday's April 2026 reporting, with practitioner reports clustering at US$25–45.

If you want the pre-flight version of that judgement, our ten benchmarks for Canadian businesses turns it into a scoring exercise, and the ChatGPT advertising agency Canada page covers how we scope it.

06The Google Ads Canadian benchmark

Google's numbers are better documented, but not always Canadian. The most-quoted cross-industry medians come from WordStream / LocaliQ's benchmark series, which is US-based; Canadian auctions in most local service verticals price similarly or somewhat lower, with Toronto and Vancouver metro able to reach US metro levels in competitive niches.

Two cautions. First, the 2026 medians widened in places because conversion rates rose industry-wide while cost per lead fell for the first time in five years — from $70.11 to $66.69 — so the direction of travel favours Google on efficiency even as its CPCs climb. Second, the Canadian rows come from an e-commerce dataset, where search intent is unusually transactional; do not transplant a 9.96% CTR onto a contractor's lead-generation account.

For the industry-level view most relevant to LeadGulls' clients, the practical ranges are: Home and Home Improvement around a $8.33 CPC with an 8.05% conversion rate and roughly $90.92 cost per lead; Legal at the top of the CPC table near $9.87 with a $131.63 cost per lead; and Dentists at $8.00 CPC with a 10.67% conversion rate and about $72.97 cost per lead. Those are the numbers any Canadian ChatGPT Ads test is competing against.

07The only arithmetic that matters

Cost per conversion is CPC divided by conversion rate. Everything else — CTR, CPM, impression share — is an input to that division. Run it on published figures only:

The interesting finding is in rows two and three. At its documented best, ChatGPT Ads produced a cost per conversion within about 10% of the Google cross-industry median — while spending a third as much per click. At its documented worst it produced nothing, and no amount of media buying converts "nothing" into a cost per acquisition.

So the honest summary is not "ChatGPT Ads is cheaper" or "ChatGPT Ads does not convert". It is: the channel's outcome is currently decided by account construction and offer fit, not by media price. That is a testable claim, and it is why the next two sections matter more than the metrics.

The published range for ChatGPT Ads runs from about seventy dollars per conversion to no conversion at all. A channel with that spread is not a channel you scale on a benchmark — it is one you qualify. Ahmet Dogan, CEO, LeadGulls Digital Marketing Agency

08The constraint nobody puts in a pitch

ChatGPT shows ads only to logged-in adults on the Free and Go plans. Plus, Pro, Business, Enterprise and Education are ad-free. Every performance figure above sits inside that boundary, and the boundary is not neutral — it correlates with who the person is.

Choice OMG reached the conclusion from its own Canadian data. SE Ranking reached a version of it too, reporting that its delivery was so uneven that the two largest ads received around half of all impressions. Workshop Digital, a B2B agency, spent its own money to learn that it could not reach its own buyer profile. Three independent tests, one mechanism.

The same logic applies to category eligibility. Canada sits outside the United States, where healthcare, financial services and legal services are restricted; outside the US those three are generally prohibited. If your business is in one of them, no performance benchmark is relevant to you at all — the eligibility map in our Canadian category guide settles it in a minute.

09Where each channel provably wins

Restricting ourselves to sourced evidence, the split is reasonably clear.

Note what is absent from the second column: any independently verified, multi-advertiser, Canadian conversion benchmark. That is not scepticism for its own sake. It is the accurate state of the evidence, and it is why the next section exists.

10Measurement limits in Canada

Three documented problems will distort any comparison you attempt in a Canadian account.

1. Attribution came late and is still partial. OpenAI's conversion tracking only became available in May 2026. Before that, advertisers saw impressions and clicks with limited ability to tie them to outcomes.
2. Most conversions may land outside the click window. One agency reports that roughly 60% of attributable conversions happen outside the immediate click window, which would make last-click attribution dramatically undercount the channel. That is a single-source figure — treat it as a hypothesis to test in your own data, not a planning constant.
3. Geographic targeting below country level is availability-dependent. OpenAI documents province, city and postal targeting as available only "where available", so a Toronto or Vancouver advertiser must confirm granularity inside the tool rather than assume it. This alone can invalidate a tight-radius test before it starts.

Add the Canadian legal layer and the measurement gets harder still: consent language under Canada's Anti-Spam Legislation (CASL) determines which leads you may lawfully contact, and a lead you cannot call is not a conversion even if the dashboard counted it. The full treatment is in our Canadian benchmarks article.

11How to run a fair Canadian test

Given the evidence, the useful deliverable is a test design rather than a verdict. This is the structure we use, and it deliberately borrows its discipline from the channel with the better measurement.

1. Run both channels on the same offer and the same landing page. Workshop Digital's comparison is credible because it controlled for everything except the channel. Do the same.
2. Pre-register a failure definition. Write down, before launch, what result would make you stop, at what threshold, and by when. In a channel where the spread runs from $73 to never, motivated reasoning is the default failure mode.
3. Budget for signal, not for access. The documented ROAS win came from roughly US$60,000 across fifteen days. A CA$70/day test cannot reproduce that learning curve; it can only tell you whether clicks and early-stage leads appear at all. Fund accordingly and set expectations to match.
4. Fire qualified-lead and closed-deal events server-side. The only documented measured conversion rate — 2.35% — came from an account that could actually see conversions. Our Conversions API setup guide is the implementation reference.
5. Match currency and geography in reporting. Do not compare a CA$7.16 Canadian CPC to a US$5.42 US median and call it a gap. Convert, or compare like with like.
6. Use a geo holdout for incrementality. The only method that survives a CFO's follow-up question, in either channel.

If you would rather have the test designed and instrumented for you, that is the opening phase of a ChatGPT Ads consulting engagement, and the wider comparison practice sits on the PPC agency and Google Ads and SEO management pages. For a Toronto-specific view, see Google Ads agency Toronto and PPC consultant Toronto.

12The verdict we would give a client

Given only what is published, here is what we would actually say — including to a client who came to us wanting the opposite answer.

• Google Ads remains the default in Canada, on evidence. It has measured conversion rates across every relevant vertical, keyword-level control, and a decade of Canadian benchmarks. Nothing in the ChatGPT Ads record displaces it as a primary channel.
• ChatGPT Ads is a qualified test, not a media plan. The evidence supports a funded, instrumented test in consumer-facing local services and e-commerce, where the intent moment is conversational and the audience tier matches the product.
• For Canadian B2B, be sceptical. Three independent tests found the buyer profile sitting behind the ad-free tier. Test it if the budget is genuinely discretionary, but do not build a pipeline forecast on it.
• For restricted Canadian categories — health, finance, legal — the question is moot. Paid access is generally prohibited outside the US. Spend the money on citation and organic AI visibility, where the evidence and the compounding both favour you.
• Never scale on an estimate. Every "average ChatGPT Ads conversion rate" you read is either an estimate, a single case, or a partner-reported example. The one measured rate published in full came in at 2.35% — below the estimate bands that circulate alongside it.

That is a less exciting conclusion than most agency articles on this topic, and it is the one the data supports. If you would like it checked against your own account rather than the national picture, the practical next step is the ten-benchmark assessment or a direct conversation through our contact page.

13Sources and confidence

Every figure in this article traces to one of the following. Confidence labels are ours, and they are deliberately conservative.

Where a metric was unavailable, no substitute was used. Where a figure is a single account, it is labelled as one. If you are aware of a Canadian ChatGPT Ads dataset we have missed, we would genuinely like to see it — our register is reviewed quarterly and corrections are published rather than quietly absorbed.

14Questions we get asked

15Glossary