Shortcuts and key features (Memory, Projects, Custom GPTs), prompt-structure patterns, and habits for getting better answers faster.
| Category | Action | macOS | Windows |
|---|---|---|---|
| Chat | New chat | ⇧⌘O | Ctrl+Shift+O |
| Chat | Focus chat input | ⇧Esc | Shift+Esc |
| Navigate | Search chat history | ⌘K | Ctrl+K |
| Input | Toggle voice input | ⇧⌘U | Ctrl+Shift+U |
Key Features
Custom Instructions
Persistent context about you and how you want responses shaped — applied automatically to every new conversation without re-explaining each time.
Memory
Lets the model retain specific facts across separate chats when enabled — distinct from Custom Instructions, and viewable/editable/deletable in settings.
Projects
Groups related chats and files together with shared instructions — useful for keeping a long-running task's context in one place.
Code Interpreter / Data Analysis
Runs actual Python in a sandbox — for real calculations, file parsing, and chart generation rather than the model just describing what it would do.
Custom GPTs
A saved configuration of instructions, knowledge files, and enabled tools — packaged as a reusable, shareable assistant for a specific task.
Prompt Structure Patterns
Role
Task
Context
Format
Iteration Techniques
| Technique | When to Use |
|---|---|
| Few-shot examples | Show 2-3 examples of the exact input/output pattern wanted before the real request. |
| Chain-of-thought ("think step by step") | For reasoning/math-heavy tasks — asking for stepwise reasoning improves accuracy on multi-step problems. |
| Ask it to critique its own answer | A follow-up "what's wrong with this?" often surfaces errors the first pass missed. |
| Constrain the output format explicitly | "Return only valid JSON, no markdown" — prevents extra commentary from breaking a programmatic parse. |
Front-load context, don't drip-feed it
Paste the relevant code/document/background in the first message rather than adding it piece by piece across several replies.
State the output format you want
"As a table", "as a bulleted list", "under 100 words" — removes an entire category of unhelpful-formatting follow-ups.
Split a big ask into stages
"First outline, then I'll ask you to write" beats one giant prompt for anything with real structure — each stage is checkable before moving on.
Set Custom Instructions once
Your role, preferred tone, and common constraints saved once apply to every future chat — far less repetitive than restating them each session.