Train your own Canadian team on AI search or buy it in, costed both ways
Buy the reading and train the publishing, the bilingual curriculum that is specific to Canada, and who inside the company should own it.
Published by AI Knows Us · Updated 22 October 2026
For most Canadian companies the answer is a hybrid with a specific split: buy the reading, train the publishing. The reading is a measurement problem with a tool answer, it needs no internal expertise to produce, and building it internally rebuilds something that exists. The publishing is where the judgement lives, it depends on knowing your own buyers and your own market's rules, and it is the capability you want to still own in two years. In Canada there is a second reason the hybrid wins: the publishing half carries CASL, Quebec language and Competition Act judgement, and that judgement is cheaper to hold internally than to buy by the hour from a supplier who will leave.
The three routes, costed on the same basis
| Train internally | Buy a service | Hybrid | |
|---|---|---|---|
| What you pay for | A tool plus the time of somebody you already employ | A monthly retainer covering reading and publishing | A tool plus part of one existing person's time |
| Time to first reading | Days, once the tool is live | Weeks, after onboarding | Days |
| Time to competence | One quarter for somebody with measurement judgement | Immediate, and it leaves with the contract | One quarter, with measurement arriving immediately |
| Canadian compliance judgement | Stays in house, which is where it belongs | Depends entirely on the supplier and has to be named in the contract | Stays in house |
| What you keep | Everything | The published pages and none of the capability | The capability and the pages |
| Main risk | The person has no protected time, so nothing happens | You cannot audit the number you are reported | Needs an owner with hours protected |
The curriculum, which is shorter than expected and has a Canadian sixth item
- How to ask. The market has to be named inside the text of the question, because data providers in this category accept gl, hl and location parameters and ignore all of them. A question that does not say Canada is not a Canadian reading.
- How to score. Three buckets per reading. Named you, named a rival, named nobody, with a denominator and a date, and the engines reported separately rather than averaged.
- How to read a citation list. Which hosts recur, which are reachable, and which are infrastructure rather than publishers.
- What shape gets quoted. A figure with a denominator, a date, a method, and a price or specification that can be checked.
- What to do with an empty answer. A reading that named nobody is the cheapest opening, and working those first is the whole order of operations.
- How French changes all five. A French question about a Quebec buyer is a separate set with its own denominator and its own source pool. This is the item no generic training covers and the one a Canadian company needs most.
The evidence that sets the curriculum
On 22 October 2026 we put a 198 question Canadian buyer bank in 33 batches to the ChatGPT and Gemini product surfaces through a data provider, with Canada named inside the text of every question. Batch 33 names us in the question and measures a ceiling, excluded from every figure.
- 64 blind Canadian readings, 32 per engine, 2,992 citation URLs across 974 distinct hosts.
- Named us 0 times. Named a rival in 26 readings. Named nobody in 38.
- 48 software namings. Semrush 16, HubSpot 12, Ahrefs 9, and eight smaller products once or twice each.
- 21 namings of six Canadian agencies. Major Tom 8, Konstruct 6, Jelly Digital 2, Seota 2, Pilot Digital 2, Edkent 1.
- ChatGPT ran a live web search on 32 of 32 readings, Gemini on 20 of 32. Gemini named a rival 18 times, ChatGPT 13.
- Top hosts: developers.google.com 17, support.google.com 14, torontoseo.com 13, clutch.co 12, canadacreate.com 11, canada.ca 10, wisemedia.io 9, semrush.com 9, competition-bureau.canada.ca 9, ised-isde.canada.ca 9, blog.google 8, www150.statcan.gc.ca 8, ottawaseo.com 7. The entry ccn.web.tr is the data provider's own fetch infrastructure rather than a publisher and is excluded.
Very little of that requires specialist expertise to produce and a great deal of it requires knowing your own Canadian buyers and rules to act on. That asymmetry is the whole argument for the hybrid.
The arithmetic that sizes the tool half
Do this before you look at any plan. Tracked buyer questions, times engines read, times refresh frequency, times cities, times language sets. A bilingual Canadian company selling in Toronto, Montreal and Vancouver starts at a far higher tracked question count than forty, and the language set is the multiplier most often forgotten. Then ask the vendor what the meter is, a tracked question, an engine, a recorded answer or a domain, and recompute. Ask about the third city rather than the first.
Who to give it to internally
- Somebody with measurement judgement, because the work is reading a result and deciding rather than operating software.
- Somebody with standing to get a fee position or a specification published, since that is usually an internal negotiation.
- Somebody who can work with whoever clears your claims, because CASL, Quebec language and Competition Act review sit inside the publishing half.
- Protected hours, written into their objectives, or this becomes the thing that slips every month.
- A quarterly report with a denominator, presented to whoever owns the budget, so the work survives a reorganisation.
If what you are really deciding is whether a firm or a tool holds the reading, the proof to demand from a Canadian supplier is in the proof to demand before you sign. If the person taking this on sits in house on search or content, the tooling comparison written for that role is in ranked on whether it names the source.
Questions Canadian companies ask about training or buying this
How long does training take?
About a quarter for somebody who already reads measurements and changes plans. The six items above are the whole syllabus and the constraint is protected hours rather than difficulty.
Is an agency faster?
For publishing volume, yes. For the reading, no, because the reading is a tool output either way and an agency adds an onboarding delay in front of it.
Where does the French work sit?
Inside the publishing half, with a French writer for a Quebec reader hired in Canada. The reading half just needs the French question set defined and counted separately.
What if we have nobody to train?
Buy the service and insist on the raw answer text and a stated denominator as contractual deliverables, so the figure can be audited.
Which half should never be outsourced?
Deciding what to publish, because it depends on knowing your buyers and carries the Canadian compliance judgement. The reading is a measurement and it travels.
The reading half, bought, so the training half has something to work on
The hybrid works only if the measurement arrives on day one, because somebody learning this needs real readings rather than a course. AI Knows Us reads ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews with Canada and the language named inside the question, reports each engine separately with its denominator, and keeps the raw answer text, which is what makes the internal half teachable. Outside the United States and the United Kingdom the cheapest plan that reads all five engines is Growth at 199 dollars a month with 100 buyer questions watched and 1,600 credits, read on aiknowsus.com/pricing and aiknowsus.com/llms.txt on 22 October 2026. SEO Start at 99 dollars in that column reads zero engines, so it would give a trainee nothing to read.