OpenAI introduces custom instructions for ChatGPT
The beta feature reached Plus subscribers first, with 1,500 characters each for user context and preferred response style, and was withheld from the EU and UK at launch.
- Models & capabilities
- Minor
OpenAI introduced custom instructions for ChatGPT, a feature letting users set standing preferences that applied to every new conversation rather than being restated each time. Users filled in two fields, up to 1,500 characters each — one describing themselves and their context, the other specifying how they wanted ChatGPT to format or style its responses — and the settings took effect from the next chat onward.
The feature launched in beta to ChatGPT Plus subscribers, with a wider rollout to free users promised in the following weeks, and was not initially available to users in the European Union or United Kingdom, reflecting the more cautious pace at which OpenAI extended new features into jurisdictions with stricter data-protection regimes. OpenAI said the change followed feedback, gathered through conversations with users across 22 countries, about the friction of having to re-explain context — a user’s job, preferred tone, or formatting requirements — at the start of every session. Submitted instructions passed through OpenAI’s moderation system, which could refuse to save or apply content that violated its usage policies, and the company noted the data could be used to improve its models unless a user opted out via data controls.
Custom instructions were a modest but widely used addition to a product whose interface had otherwise changed little since launch, and previewed a direction — persistent, user-controlled context — that OpenAI and competitors expanded substantially over the following years through features such as memory across conversations.