How to Ask Copilot Best About Learning with Knowledge Objects

A copy/paste playbook for storing, using, updating, and forgetting structured knowledge.

Focus: structured “knowledge objects” Outcome: better answers, fewer clarifications Tip: say whether to remember or not

1) What “learning” means (2 modes)

A) In-session learning (always)

I can use what you tell me in this chat to answer better right now.

B) Cross-session memory (when you ask)

If you explicitly ask me to remember something (and it’s stable/useful), I can store it for future chats.

Tip: If you want it to persist, say: “Remember this for next time…”

2) What is a “knowledge object”?

A small, structured bundle of facts with clear metadata, so it can be reused consistently:

  • Type (preference / profile / project / decision / definition / dataset)
  • Content (the facts)
  • Scope (this chat only vs. remember)
  • Source (you, email, link, meeting notes)
  • Validity (time-bound or until changed)
  • Confidence (high/medium/low if uncertain)

3) The best way to ask: use a template

These two templates are the fastest route to accurate, consistent answers.

Template 1 — Store as memory (persists)

Please remember this as a knowledge object: Type: Preference | Profile | Project | Decision | Other Content: … Scope: Remember for future chats Source: Me (the user) Valid until: (optional date) or “until I change it” Notes/Constraints: (optional)

Template 2 — Use now only (do not store)

Use this as a knowledge object for this conversation only: Type: … Content: … Scope: This chat only (do not remember) Source: … Valid until: …

4) Control prompts (remember / forget / update)

What do you remember?

What do you remember about me from previous chats? List it as bullet points.

Use only what’s provided

Answer using only what I tell you and what’s in the knowledge objects. If missing, ask.

Forget something

Please forget this memory: [describe the item]. Confirm it’s removed.

Update a memory

Update the stored knowledge object: Old: … New: … Effective date: …

5) Make knowledge objects “machine-friendly”

  • IDs & versions for projects/decisions (helps track changes).
  • Clear definitions to prevent ambiguity.
  • Time + timezone for market-sensitive items.

Example: IDs & versions

Knowledge Object ID: PROJECT-VAR-DIV-001 Version: 1.2

Example: time context

Timezone: Europe/Oslo As-of date: 2026-04-17

6) Copy/paste examples

Example 1 — Preference memory

Please remember this as a knowledge object: Type: Preference Content: When I ask about shipping markets, reply with a short summary first, then the data table. Scope: Remember for future chats Source: User Valid until: I change it

Example 2 — Project context (durable)

Please remember this as a knowledge object: Type: Project Content: I’m tracking VLCC (TD3C/TD15/TD22) and Norwegian E&P dividends (VAR, EQNR, AKRBP). Scope: Remember for future chats Source: User Valid until: end of 2026

Example 3 — Sensitive info (do not store)

Use this as a knowledge object for this conversation only (do not remember): Type: Dataset Content: My portfolio holdings are: … Scope: This chat only Source: User

7) Tell me what to do with the object

End your request with one of these lines for best results:

  • “Store it.” (remember it)
  • “Use it only now.” (don’t remember)
  • “Ask if uncertain.”
  • “Prefer accuracy over speed.” (I’ll verify/clarify)
  • “Prefer speed over accuracy.” (I’ll estimate and label it)