The assistant answers a question about one client using an exception agreed for another. Both documents are relevant to onboarding, but only one governs the current case.
Choose AI knowledge-collection boundaries according to which information should be considered together. Keep shared approved guidance distinct from project-specific decisions when combining them would create ambiguity. The right arrangement depends on the task and the actual platform's capabilities, not a universal rule to upload everything into one space.
Photo by CHUTTERSNAP on Unsplash.
Start with the question boundary
List the questions the collection should answer and the context needed to select the correct source. A general staff question about the standard service differs from a delivery question about an individual engagement.
Do not rely on a model to infer every relationship from a filename. Record the service, project and effective period where those distinctions matter.
Collection design does not replace access controls. Keep permission decisions with the appropriate owner and verify the actual system rather than assuming a separate folder guarantees separation.
Separate three kinds of knowledge
Use this working distinction:
Shared rules: approved procedures and standard service definitions.
Local decisions: authorised project or client-specific variations.
Working material: drafts, ideas and unresolved questions.
Decide which kinds belong in each task context. A general answer should not quietly inherit a local exception. Equally, a project-specific answer should not ignore an approved variation merely because the standard procedure is easier to retrieve.
An illustrative support arrangement
Suppose the normal service operates on weekdays, while one engagement includes separately agreed weekend cover. A shared public FAQ should describe the standard offer. A private engagement task may need both the standard rule and the specific agreement.
The source relationship matters more than the number of collections. The answer should identify which rule applies and why. If the tool cannot preserve that context reliably, narrow the task or keep the decision with a person.
Compare maintenance effort
Separate collections can reduce irrelevant material, but they can also create duplicate copies of shared guidance. Ask who updates each copy when a rule changes.
A shared authoritative source with clearly selected local context may be easier to maintain than several independently edited versions. Check what the platform supports before choosing that arrangement.
Document the intended relationships so the person maintaining the resource can explain which source governs a conflict.
Try cross-boundary questions
Use an ordinary shared question, a project-specific question and one where a local exception conflicts with the standard rule. Inspect the sources and resulting answer.
Check what happens when the user omits the project name. A request for clarification may be preferable to a confident answer from the wrong context.
Select the smallest useful arrangement and review it when the service or collection expands. The aim is relevant, attributable answers, not a more elaborate folder structure for its own sake.
For related practical guidance, see defining minimum information access and keeping shared knowledge current.
Hili Consulting can help define useful knowledge-collection boundaries for shared and project-specific work. Arrange a free 30-minute consultation and bring one example of the work you want to improve.
