ChatGPT Plus gets custom instructions, allowing it to remember how you want it to behave

OpenAI today unveiled a potentially impactful feature for users for its ChatGPT Plus subscription service ($20/month): Custom instructions, a new setting that users can toggle on when logged into their ChatGPT Plus account. The setting allows the AI chatbot to store information about how the user wants it to respond and behave, retaining this perspective even when the user closes one chat and begins another.

The feature, currently available in beta release outside of the UK and EU, could save enormous time for regular users of the service, as it prevents them from having to begin with the stock ChatGPT interface and then “priming” it with the perspective the user wants every time they open a new chat window. In other words, you can type up your overarching prompt one time, and ChatGPT Plus will save it for as long as you wish, even as you prompt it with new requests and questions going forward and close and begin new chat conversations.

Potential use cases

OpenAI cited the hypothetical example of “a teacher crafting a lesson plan.” With the custom instructions feature enabled, the teacher “no longer has to repeat that they’re teaching 3rd grade science,” every time they begin a new chat with the service. Instead, ChatGPT Plus will retain this perspective and answer with it mind going forward.

Or, if you are a developer who likes to code in Python, you can store that information in the new custom instructions setting, and ChatGPT will return results in Python every time you ask for coding help instead of you having to keep reminding it to do so.

How to use the Custom Instructions on ChatGPT Plus now

Users can try the setting now on the web or on the ChatGPT iOS app. On the web, it’s accessible by clicking your account username in the lower left corner of the ChatGPT interface, then clicking Settings, Beta Features, and toggling on “Custom Instructions.”

Then, the user has to close the menu and click on their name again. A new menu option should appear in the pop-up, labeled “Custom Instructions.” Clicking on it will give you a new screen with two questions that OpenAI asks you to answer in 1,500 characters or fewer.

“What would you like ChatGPT to know about you to provide better responses?” and “How would you like ChatPT to respond?”

OpenAI provides “thought starters” for each question to guide your answers, including “where are you based?”, “what do you do for work?”, “what are your hobbies and interests?” for the former, and “How formal or casual should ChatGPT be?”, “how long or short should responses generally be?”, “How do you want to be addressed?” and “Should ChatGPT have opinions on topics or remain neutral?”

Based on these frameworks, it seems as though OpenAI is trying to help guide users into priming ChatGPT to respond in a custom way for each of them, and retain that customization for as long as they have the setting toggled on.

While OpenAI presently only allows ChatGPT Plus users to feed in one set of custom instructions at a time, the instructions are entirely open ended provided they fit in the 1,500-character long text boxes. This means that you can actually have ChatGPT respond from multiple perspectives, as well, if you enter those into the text box.

Initial experiments show potential

VentureBeat experimented with this by typing “I’m a novelist writing a new work of science fiction. Please keep in mind each character’s motivations, personalities, and relationships in mind as you build the story,” in the first text box, then providing character descriptions in the second.

The raw results were technically and grammatically sound, although clearly short of the unique writing voice and rigor we’d expect from a published novel…for now. And they could presumably be edited by a person into something appealing to some readers.

OpenAI’s beta release of the new custom instructions feature comes just a few weeks after it released another big new feature, Code Interpreter, allowing users to upload documents, create visualizations based on data they provide and have ChatGPT write and run code in Python.

The new features come at a time of growing opposition to OpenAI and ChatGPT, including lawsuits from authors who allege OpenAI scraped their books in violation of copyright and a similar lawsuit by a famous comedian. There have also been complaints over alleged degradation in ChatGPT response quality from when the service first became available in November 2022 and when the model underlying it was updated to GPT 4 in March 2023. The U.S. Federal Trade Commission is also reported to be investigating the company over a data breach.

Still, OpenAI is clearly moving ahead with what it views as improvements and useful new capabilities for its signature service and forging new alliances with established names in media, including The Associated Press and the American Journalism Project.