Knowledge card L2 · Context engineering informational

Tell ChatGPT What to Remember On Purpose

What it is

An active habit: explicitly tell ChatGPT to remember durable facts ('remember I use British spelling', 'remember my team is five people') rather than leaving it to infer what's worth keeping.

curate the inputsNotice a durable factSay 'remember that…'Reused across chats
Deliberately saved facts beat whatever the model happens to absorb.

Why it works

Left alone, Memory captures whatever the model decides is notable, which is a mix of the useful and the random. Directing it — stating the specific, stable facts you want carried forward — fills Memory with high-signal entries you'll actually benefit from, instead of a drift of half-relevant details.

When to use it

Whenever a genuinely reusable fact comes up mid-conversation — a preference, a recurring constraint, a stable detail about your work or setup.

When not to use it

Transient or task-specific details ('this draft is due Friday') that will be stale next week — those clutter Memory and skew later answers.

Prompt

Remember this for future chats: <specific, durable fact — a preference, constraint, or stable detail>. Keep it concise so it stays useful across unrelated conversations.

Example

'Remember I write in Australian English and prefer metric units' means every future draft arrives correct without you re-specifying it each time.

Common mistakes

  • Letting Memory fill with whatever the model happens to notice.
  • Saving time-bound details that are wrong within a week.
  • Storing so much that useful entries drown in trivia.

Related