How do we structure knowledge so Harriet answers with the right version first?

Split versions by audience and geography, write clear titles, and avoid one giant document that mixes jurisdictions.

When several documents could match a question, quality depends as much on how you publish as on the AI.

Split by audience and jurisdiction

Prefer separate files for materially different rules (countries, entities, union vs non-union).
Give each file a specific title and summary in the console so the right document is identifiable at a glance.

One topic, one primary doc per context

For a single topic (for example parental leave), have one canonical document per jurisdiction or employee population.
Link related docs in the body (“If you are in India, see …”) instead of pasting all variants into one file.

Tags and groups

Use groups to enforce who may see sensitive variants.
Use tags consistently for maintenance and filtering.
Align country / region metadata with the actual rules in the file.

Example

Instead of one 80-page “Global handbook,” publish UK handbook, US handbook, and India handbook with matching metadata. A UK employee’s question hits the UK file first; HR in a cross-border group can still open the others if their groups allow it.

Guardrails

Duplicating the same policy in three places without a clear owner creates drift—assign an owner per document.
If two documents contradict each other, Harriet may still surface both; resolve conflicts in the source content.