OpenAI · ChatGPT · 15 July 2026
More room for instructions does not mean better instructions.
Eligible paid and organisational plans can now store up to 5,000 characters of custom instructions instead of 1,500.
Tidy up; do not fill the new space. Better placement and fewer durable rules beat a 5,000-character autobiography.
Why this may matter
Stable context can reduce repetition, but a long hidden rulebook can make answers strange and hard to debug. Project-specific rules still belong with the project, not in a global profile.
Notice when an instruction changes a high-stakes answer or conflicts with the current request. The current task should remain visible and reviewable.
Inspect the change, example and test detail
What changed
The limit increased for Plus, Pro, Enterprise, Business and Education users, creating more room for persistent response preferences and working rules.
Real example
Keep spelling, accessibility needs and a rule to label uncertainty globally. Keep a client’s terminology, audience and deliverable rules inside that client project.
A bounded test
Ask of every current instruction: should this apply to every conversation? Delete stale rules, move project rules and retain only persistent preferences.
Sources and review
Sources retain their own provenance. A provider announcement establishes what was announced; it does not become independent testing.
Canonical claim: The limit increased for Plus, Pro, Enterprise, Business and Education users, creating more room for persistent response preferences and working rules.
Readiness: Use the dated guidance with the stated human check; this is not a completed Human Bit test.
- Released
- 15 July 2026
- Reviewed
- 25 July 2026
- Review by
- 8 August 2026
- Review owner
- The Human Bit editorial
- Revision
- custom-instructions-grow@v1
- Volatility
- volatile