AI tutoring for schools: a buyer's guide to the questions worth asking before a pilot

This is the question that separates a tutoring product from an answer product, and it is the one to ask first. The useful form is: with no settings changed, what happens when a student types just give me the answer. A vendor should be able to show you, in the product, rather than describe a philosophy.

CANtutor AI's answer is mechanical. Every account starts in Learn mode, which teaches the idea with small steps and questions rather than handing over the solution. Practice mode waits for the student's attempt, names the one place it went wrong, and gives the smallest hint that gets them moving. There is an Answer mode; it is labelled, and every surface that shows it carries the same warning that turning in AI-written work as your own may breach the school's academic-integrity policy. The AI Grader ignores the mode, so a student cannot soften a mark from the sidebar. Ask any vendor for the equivalent.

Ask for the published policy and read it against a short list: who the processing providers are, whether student content is used to train models, where it is stored and processed, how long it is kept, and how deletion works. A vendor that answers by category only, or that cannot name providers, is asking the school to trust an architecture it will not describe.

CANtutor AI's policy names its providers by company, states that its agreements do not permit them to use content to train general-purpose models, and says plainly that some processing happens outside Canada, including in the United States. Where a school enables the service, the school determines what information is provided and the service handles it on the school's instructions. The full checklist, with each answer tied to a section of the policy, is in the companion article on privacy.

Most schools run a Google or Microsoft environment and an LMS, and the practical question is whether the tutor can see the coursework students are actually assigned without a second copy being maintained by hand. Ask precisely what an integration reads, whether it writes anything back, and what happens when the district's policy blocks it.

Google Classroom connects for personal Google accounts and for school accounts where the district permits third-party apps. Some districts block it; if yours does, everything else still works and you upload your material instead. When it is connected, the sync is read-only and one-way: courses, assignments, due dates, attachments and rubrics come in, and nothing is submitted or written back to the teacher. Those assignments and rubrics are then what the tutor, the explainer and the grader work from. A page written to be forwarded to a Workspace administrator is linked below.

Not grades, at least not in a term. No vendor has outcome data that transfers to your school, and one that promises a grade improvement is promising something it cannot know. Measure use and behaviour instead, because those are observable in weeks.

CANtutor AI is metered by credits, which students can see broken down by feature, and the plans are on /plans. For a school or district deployment, the /for-schools page has the contact route; Ontario boards have their own page covering the privacy questions boards ask, with each answer linked to the policy section that supports it.