Oct 11, 2025
ChatGPT Design Improvements
Usable Prompt Personas
Making prompt personas easier to create, switch, and reuse.
The Problem
With ChatGPT, a best practice is to define a clear “persona” and
scope so the model knows how to respond.
Today, users either:
(1) retype a persona and constraints for every
conversation
(2) search for (or build) a custom GPT.
The first path is repetitive and mentally taxing; The second adds
discovery and setup friction—users must understand what a GPT is,
where to find good ones, and how to make their own. On top of
that, the current GPT entry points aren’t integrated into the main
chat workflow, which forces context switching. As a result, this
high-value capability is underused.
Our Solution
Make prompt personas immediately accessible in the chat. Reduce friction to create, select, and switch personas so users get better results with less effort and higher satisfaction.
Implementation
Address usability at the point of action (the composer):
- Inline Persona Picker: Place a compact persona selector directly above the message field. One click to choose or switch.
- Mid-chat Switching: Change personas during a conversation (e.g., start with “UI Engineer,” then switch to “Senior Software Engineer” for scope review).
- Quick Create & Save: From the same picker, “Create new persona,” write a short description, and save. No separate page.
- Smart Scaffolding: When creating a persona, auto best-practice prompt structure on the backend(voice, tone, constraints, do/don’t) so the user only describes intent.
- Inline Edit: Edit an existing persona in a popover and resave; no full-screen detour.
- Clear Feedback: Show the active persona so users always know what’s applied.
This brings personas into the line of sight, lowers cognitive load, and encourages experimentation without context switching.
Persona demo video
Expected Impact
Easier access and creation should increase adoption and quality: faster time-to-useful responses, more consistent results across chats, and richer outcomes as users comfortably switch personas to match the desired output.
Results
These changes led me to use personas far more often than GPTs in ChatGPT. Being able to switch personas mid-thread produced better outcomes. Inline creation and edits—paired with best-practice scaffolding—removed the guesswork around “the right way” to write a persona, resulting in higher-quality setups.
There are still refinements to make. The popover UI needs polish (spacing, hierarchy), and a “response type” option I tested wasn’t useful—personas should infer that from context. Overall, the direction is promising, but it merits broader user testing (including sharing/using other people’s personas) to validate placement, defaults, and wording.
Conclusion
This design shows that small, incremental improvements to personas can meaningfully improve the ChatGPT experience—yielding better responses and giving users more control over how content is produced. That said, further research and testing on the UI/UX flow are needed to confidently integrate this directly into ChatGPT.