AI transparency
How we use AI
We believe in full transparency about how our software works. Here is an open look at where AI assists our platform, the limits we place on it, and how your data is handled.
Last updated: October 2026
Human in the loop
All outputs are strictly drafts or suggestions. Nothing changes in your library until you approve it, and every approval is logged.
Never trained on your data
Your content is never used to train AI models. Our enterprise API terms with Google strictly forbid training on customer data.
A single, trusted provider
All AI features run exclusively on Google Gemini. We do not send your data to any other third-party models or providers.
Where AI is used, and where it is not
We apply AI purposefully, not indiscriminately. Here is the full breakdown of every AI-assisted task, including what information leaves your environment, and who holds final approval.
| Task | Uses AI | What is sent | Who decides |
|---|---|---|---|
| Ingest and convert documents | AI-assisted | PDF/document contents and metadata. AI restricted to structuring data and formatting; it does not alter underlying text. | Your institution. Migrated documents undergo internal QA sampling and require final institutional sign-off. |
| Analyze readability | No AI | Nothing leaves your environment. Processed entirely on-platform. | Deterministic. Calculated using standard readability formulas. |
| Analyze governance | AI-assisted | Document text and recorded metadata (owner, authority, dates, definitions, citations). | The policy owner. Scores and citations are purely advisory; all revisions are in human hands. |
| EXPERIMENTAL: Create plain-language draft | AI-assisted | Body text of the selected document. Triggered on demand. | The editor. Outputs are editable drafts. Human approval is required to accept changes. |
| Check for duplicate definitions | AI-assisted | The new term and definition and up to 40 potential matches from your register. | The author. Potential duplicates are flagged fro review; the author decides how to proceed. |
| Analyze definitions | AI-assisted | The selected definition and related entries. | The reviewer. Suggested refinements must be reviewed by a human and accepted. |
| Create draft definition | AI-assisted | The term, institution type, and jurisdiction. No policy text. | Assigned reviewer. Follows standard two-step institutional approval. |
Processing location
Where AI processing happens
Data residency and AI processing operate under separate, clearly defined boundaries.
Data Storage. The database that holds your library are stored in Canada.
AI Processing. When an AI action is triggered, the required text is sent encrypted to Google Cloud's secure U.S. facilities for processing, and returned immediately. Google does not store your content or use it to train models.
Questions about AI use
For further information, please contact us at privacy@academicintelligence.ca.