Almost every AI statistic that circulates publicly is missing something — the country it was measured in, the sample size, the definition of “use,” or the difference between a fact and a vendor’s estimate. Statistics Canada’s own releases are a useful case study, because reading them properly means noticing what an incomplete version of the same numbers would hide.
Key takeaways
A statistic about AI travels well precisely because it is easy to repeat and hard to check. The fix is not scepticism about every number, but a short, repeatable set of questions applied to whichever number is in front of you — illustrated here with Statistics Canada’s own AI-use releases, because they are unusually well documented and still get misread.
Statistics Canada’s Q2 2026 release states that 19.2% of businesses reported using AI to produce goods or deliver services over the preceding 12 months. Read alone, that number can be made to sound like either a lot or a little depending on the writer’s intent. The same release also states that 40.0% of businesses indicated the use of AI is not relevant to the business — and a separate StatCan release for the third quarter of 2025 found that two-thirds (66.7%) of businesses reported no plans to adopt AI over the next 12 months, with 18.9% uncertain. None of these numbers contradicts the others; together they describe a Canadian business population where a minority actively use AI, a large share see no relevance to it, and most have no near-term plan to change that. A single headline figure, quoted alone, erases that texture.
StatCan’s Q2 2026 figures come from the Canadian Survey on Business Conditions, collected 1 April to 6 May 2026, with a total sample of 21,105 businesses and results based on 9,251 responding businesses. That is a dated, defined, methodologically disclosed Canadian survey — a business self-reporting whether it used AI “to produce goods or deliver services,” not a third-party audit of AI usage or performance. A great many circulating AI statistics disclose none of this: no date, no country, no sample, no definition of what counting as “use” required. If a statistic can’t answer “measured how, of whom, and when,” that is itself the finding.
A related StatCan study modelling AI adoption reports that firms using data analytics are 15.0 percentage points more likely to adopt AI than firms that do not, based on a probit regression across the pooled 2019 and 2021 waves of the Survey of Digital Technology and Internet Use. That is a real, disclosed, Canadian statistical result — and it is still a correlation from an observational survey, not a controlled experiment proving that adopting analytics causes AI adoption. The study itself frames it as firms with certain capabilities being “significantly more likely” to adopt, which is the honest and more limited claim. A version of this statistic stripped of its methodology (“analytics leads to 15% more AI success”) would be a different, unsupported claim wearing the same number.
The Pan-Canadian Artificial Intelligence Strategy’s Phase I is described as $125 million through Budget 2017, funding the Canada CIFAR AI Chairs programme and the Amii, Mila and Vector institutes. That is real, sourced Canadian government spending — on research and training infrastructure. It is not a market-size estimate, not a return-on-investment figure, and not evidence about what any individual business will gain from adopting AI. Dollar figures attached to AI in Canada are almost always one of these two very different kinds of number: programme input, or claimed output. Confusing the two is one of the most common ways an AI statistic misleads without technically being false.
The NIST AI Risk Management Framework is widely cited in AI governance writing, including by Canadian sources — but it is a United States government framework, “intended for voluntary use,” released January 26, 2023, and the same page states plainly that “the AI RMF 1.0 is being revised as part of the White House AI Action Plan.” Nothing about NIST is Canadian, and nothing about it is settled. A Canadian source can legitimately reference it — Canada’s own privacy commissioners do exactly that, discussed further below — but a statistic or framework that quietly drops the “US” and “under revision” qualifiers on the way into a Canadian article has been rewritten into something more confident than the source actually supports.
A five-question checklist
Before repeating an AI statistic: (1) is this the whole finding, or a fragment that omits a counterweight in the same release; (2) does the source disclose what was measured, of whom, and when; (3) is a correlation being stated as a cause; (4) is a dollar figure describing spending or a return; (5) is a non-Canadian or vendor source’s qualifier — “voluntary,” “under revision,” “United States” — still attached, or has it quietly disappeared.
Take a bare, unsourced version of a claim along the lines of AI use having tripled in Canada. StatCan’s own release supports something close to that, precisely stated: “this proportion has tripled since the second quarter of 2024 (6.1%)”, referring specifically to the 19.2% Q2 2026 adoption figure. Stated that precisely, the claim survives the checklist — it names the metric, the dates, and the source. Stated as a bare tripling with no metric, no dates and no country attached, the same underlying fact has been stripped of everything that let it be checked, even though the number itself did not change.
Related: how Canadian businesses actually use AI, and AI in Canadian small business.
They don’t contradict — they describe different slices of the same population. 19.2% use AI, 40.0% say it’s not relevant, and two-thirds report no adoption plans (Q2 2026), (Q3 2025); all three can be true of the same business population at once.
Not on its own. The Pan-Canadian AI Strategy’s $125 million (Budget 2017) is programme spending on research infrastructure, not a measured return — the two are frequently conflated in public discussion of AI dollar figures.
Only with its qualifiers intact: it is a United States framework, voluntary, and under revision. Canada’s own privacy regulators cite it for a narrow, specific purpose — it should not appear as if it were Canadian policy.
A short call is enough to check what it actually measured before you act on it.
