How to Change ChatGPT’s Personality and Check the Results

Illustration: The featured illustration depicts the four personalities announced with GPT-5 in August 2025. Follow the current Settings > Personalization instructions below.

Choose a ChatGPT personality for the way you want an answer to sound. A terse answer still needs checking. A skeptical tone does not establish that an argument is right.

This guide was reviewed against OpenAI’s documentation on September 14, 2026. The instructions below are documented behavior; this revision does not claim an account-level test across every app or plan.

Where to change the setting

Open Settings, then Personalization. OpenAI’s current personalization guide lists Friendly, Pragmatic and None as default personality choices. It says available controls vary between web and desktop. Use the options your app actually shows; this guide does not establish a universal mobile layout or paid-plan requirement.

OpenAI describes personality as a change to communication style, without changing the model’s capabilities. The previous version of this article wrongly treated the presets as separate improvements in coding precision, reasoning and business judgment.

For a more specific preference, use custom instructions. OpenAI documents these as a way to carry response-style preferences across chats. A task-specific request can also state the tone and format you want.

Try a useful comparison

Use a small task whose facts you already know. Keep the task and source material the same while comparing the available styles. Judge the response against your requirements, including any facts or instructions it omitted. The following is an example prompt, not a recorded model response:

Rewrite the note below for a colleague.
Keep every date, number and commitment unchanged.
Use plain language and stay under 100 words.
Do not add a promise or an explanation that is absent from the note.

Note: The prototype review is on Thursday at 10 a.m.
Please send comments by Wednesday at 4 p.m.
We have not agreed on a release date.

Check whether Thursday, 10 a.m., Wednesday and 4 p.m. survived, and whether the draft invented a release date. Then decide which tone you prefer. This is a suggested way to compare outputs, not evidence that one preset is more accurate.

Why older guides mention Cynic, Robot, Listener and Nerd

OpenAI introduced those four names as a research preview with GPT-5 in August 2025. The announcement described optional communication presets, initially for text chat. Those launch details explain the original title and URL of this article; they should not be used as a current menu map.

Correction history

The September 14, 2026 revision replaces the unsupported sparkle-icon instructions and the claim that the feature has the same layout across platforms. It removes unsubstantiated productivity, professional-use and training-method claims. The old sample replies had no retained test evidence and have been removed rather than presented as observed output. The original publication date and URL remain.

If you are choosing a product rather than its tone, use the AI Model Picker to build a workflow checklist.