Two years ago, 71.8% of Canadian businesses said they had no plans to use artificial intelligence over the coming year. That share is now 52.7%. But the businesses that left the “no” column did not all arrive in the “yes” column, because the question has a third box, and Raw POV’s reading of the underlying tables puts nearly a quarter of the retreat there.
I started this one expecting a simple story about adoption, and found a counting problem instead. The headline number from Statistics Canada’s latest business conditions survey is that a quarter of businesses, 25.2%, now plan to use AI in the next 12 months. The companion number, 52.7% with no plans, was reported alongside it. Add them together and you get 77.9%, which leaves 22.1% of Canadian businesses unaccounted for in every summary I read. That missing fifth is not a rounding error and it is not a gap in the data. It is a published answer category, and it is the fastest growing one of the three.
The survey question has three answers, not two
The release published on 31 August 2026 covers collection between 2 July and 6 August 2026, and it describes the results as applicable to employer businesses in Canada. It quotes two of the three figures in prose. The full table behind it, catalogue 33-10-1207-01, carries all three: 25.2% yes, 52.7% no, and 22.1% recorded as unknown. The three sum to 100.
The same structure holds in the two earlier rounds, which matters, because it means the series can be compared without guessing. In the third quarter of 2024 the split was 10.6% yes, 71.8% no, 17.6% unknown. In the third quarter of 2025 it was 14.5%, 66.7% and 18.9%. Three years, three answers, one consistent question.
What “unknown” means in this survey, and how we checked
A category labelled unknown could mean two very different things. It could be a business that answered the question by saying it did not know. It could also be an administrative residue, a business that skipped the item entirely, which would say nothing at all about how Canadian firms are thinking. Treating the second as the first would be the easiest mistake to make here, so we tested it rather than assuming.
The test used the agency’s own prose against its own tables, inside the same release. On tariffs, the release states that 19.8% of businesses were unsure about the impact of United States tariffs on goods they sell. The table behind that sentence, 33-10-1188-01, labels that exact cell “Level of impact, unknown”, value 19.8. On passing costs to customers, the release says 18.7% were unsure. Table 33-10-1193-01 labels that cell “unknown”, value 18.7. In both cases the agency writes “unsure” in the text for the cell it labels “unknown” in the table.
That does not make the AI figure a statement about individual firms, and we are not treating it as one. It does establish, from primary documents rather than inference, that this category in this survey is a declared answer and not a filing artefact. Worth noting for anyone repeating the exercise: the label alone would not have settled it.
Where the retreat actually went
Between the third quarter of 2024 and the third quarter of 2026, the share of businesses with no plans for AI fell 19.1 percentage points. Over the same two years, the share with plans rose 14.6 points. The undecided share rose 4.5 points, from 17.6% to 22.1%.
Those three movements are the whole story, because the three shares always sum to 100. Of the 19.1 points that left the “no” column, 14.6 points went to a plan and 4.5 points did not. That is 23.6% of the retreat landing in the undecided middle, and it is the figure Statistics Canada does not publish anywhere in this release or its tables. In relative terms the undecided group grew by 25.6% over the two years, which is faster growth than the raw point change suggests, because it started from a small base.
One more comparison is worth making. In 2024, for every business that planned to use AI there were 1.66 that could not say. By 2025 that ratio was 1.30. In 2026 it is 0.88. The undecided group has not shrunk. It has simply been passed, for the first time in the series, by the group with a plan.
The reasons look different once they sit on a common base
The release also lists why businesses without plans say they are staying out, and this is where a denominator quietly changes underfoot. Those percentages are calculated on the businesses that report no plans, not on all businesses. In 2026 that base is 52.7% of the economy. In 2024 it was 71.8%. A reason can therefore climb as a rate while covering fewer businesses in absolute terms, which is exactly what happened.
Raw POV converted each reason to the common base by multiplying the reason’s rate by the share of businesses reporting no plans in that year. Statistics Canada publishes both components separately and does not publish the product.
| Reason given for no AI plans | 2024, share of the no group | 2024, share of all businesses | 2026, share of the no group | 2026, share of all businesses | Change on the common base |
|---|---|---|---|---|---|
| AI is not relevant to what we make or deliver | 74.2% | 53.3% | 79.1% | 41.7% | 11.6 points lower |
| Concerns about privacy or security | 6.8% | 4.9% | 10.8% | 5.7% | 0.8 points higher |
| Lack of knowledge on AI capabilities | 9.3% | 6.7% | 9.9% | 5.2% | 1.5 points lower |
| The technology is not mature enough | 8.8% | 6.3% | 9.2% | 4.9% | 1.5 points lower |
| AI is too expensive | 6.1% | 4.4% | 4.4% | 2.3% | 2.1 points lower |
Read down the last column and one row goes the other way. Every reason for staying out covers fewer Canadian businesses than it did two years ago, except privacy and security, which covers slightly more. Cost is the reason that collapsed hardest, falling by roughly half on the common base, which sits oddly beside the assumption that price is the barrier. Our earlier reading of the adoption data found much the same thing from a different direction, that the divide was not really about budget.
The relevance answer deserves its own sentence. Within the group that says no, it became more dominant, rising from 74.2% to 79.1%. Across the whole economy it fell from 53.3% to 41.7%. Both are true, and only one of them is usually quoted.
What the data cannot show
This survey is a series of separate cross sections, not a panel. It does not follow the same businesses from one quarter to the next. When we say 23.6% of the retreat became indecision, we are describing how a distribution moved, not tracking firms that changed their minds. The businesses answering in 2026 are not the businesses that answered in 2024.
The survey also does not ask what the undecided are waiting for. It records that they could not say, and stops. Anyone who tells you which way that group will break is guessing, and the tables do not support the guess in either direction. Nor does the data speak to whether the reported plans turn into anything: a separate reading of the same survey programme found that among firms already using AI, the training gap by company size is far wider than the adoption gap, which is a reminder that intention and practice are measured separately for a reason.
Finally, the reasons table applies only to businesses with no plans. It says nothing about the undecided group, whose reasons this survey does not collect at all. The largest single unknown in the AI question is, fittingly, unexplained.
Frequently asked questions
What share of Canadian businesses plan to use AI in the next 12 months?
25.2%, according to the Canadian Survey on Business Conditions for the third quarter of 2026, collected between 2 July and 6 August 2026. That is up from 14.5% in the third quarter of 2025 and 10.6% in the third quarter of 2024.
How many Canadian businesses are undecided about AI?
22.1% of businesses fall into the answer category Statistics Canada records as unknown, which is the same category the agency describes as “unsure” when writing about other questions in the same release. It was 17.6% in 2024 and 18.9% in 2025.
Why do Canadian businesses say they will not use AI?
Among businesses reporting no plans, 79.1% say AI is not relevant to the goods they produce or the services they deliver, 10.8% cite privacy or security concerns and 9.9% report a lack of knowledge about what AI can do. Those percentages describe the no group, which is 52.7% of all businesses, not the whole economy.
Is cost the main barrier to AI adoption in Canada?
Not according to this survey. Cost is cited by 4.4% of businesses without plans, down from 6.1% two years earlier, and on a whole economy basis it fell from 4.4% to 2.3% of all businesses. Relevance, not price, is the answer given by the large majority.
Sources and method
From the source. Statistics Canada, The Daily, “Canadian Survey on Business Conditions, third quarter 2026”, published 31 August 2026, collection 2 July to 6 August 2026 (release). The three answer shares and the reasons come from the primary tables, not from the prose summary: table 33-10-1207-01 and table 33-10-1208-01 for 2026, tables 33-10-1045-01 and 33-10-1046-01 for 2025, and tables 33-10-0878-01 and 33-10-0879-01 for 2024. All figures are for Canada, all industries. Statistics Canada rates every estimate used here as quality A, its highest band.
Raw POV’s own analysis. Three things in this piece are ours and are not published by Statistics Canada. First, the split of the two year retreat: no plans fell 19.1 points, plans rose 14.6, undecided rose 4.5, so 23.6% of the movement out of “no” did not become a plan. Second, the conversion of every reason to a common base, by multiplying each reason’s rate by the share of businesses reporting no plans in the same year, which is what makes the privacy row visible as the only one rising. Third, the ratio of undecided businesses to planning businesses across the three rounds, 1.66 then 1.30 then 0.88.
A correction to our own working note. We had expected the 22.1% undecided share to be a figure we would have to derive by subtraction. It is not. Statistics Canada publishes it directly in table 33-10-1207-01, and publishes 18.9% for 2025 and 17.6% for 2024. What is absent is the comparison across years, not the numbers themselves.
On the word “unknown”. We did not assume that the table’s unknown category means a business that could not decide. We checked it against two other questions in the same release where the agency’s prose says “unsure” for the cell its table labels “unknown”, on tariff impact (19.8%) and on passing costs to customers (18.7%). Both matched exactly. This establishes the category is a declared answer within this survey. It does not establish what any individual business was thinking.
Related reading from Raw POV. On what Canadian law actually obliges companies to do with the technology already in customers’ hands, see our reporting on the gap between the repair law and any duty to sell a part. On another Statistics Canada release that separates a group’s characteristics from where its credentials came from, see our reading of the overqualification data.
