The Quick Guide to Every ChatGPT Feature (When & How to Use Each)

Executive overview

Most users exploit only a handful of ChatGPT's features, leaving significant productivity gains unused. This walkthrough covers all 24 features — from model selection to agent mode — with clear guidance on which ones matter most.

The core insight: customised GPT projects and dictation-based input return the highest productivity gains for the lowest effort.

Model selection

• Auto mode handles 95% of tasks — let ChatGPT decide the reasoning level.
• Use Instant only when speed is critical and accuracy is secondary.
• Use Thinking for higher-stakes tasks; Standard covers most cases, Extended for when accuracy is critical.
• Legacy models exist under settings but add unnecessary complexity for most users.

Input methods

• Dictation is underused — speaking produces more context, more fluidly than typing.
• Advanced voice mode enables back-and-forth conversation; useful for interview/pitch practice or reflective thinking sessions.
• Branching conversations let you fork a chat at any point to explore two different directions from the same starting input.

Foundation settings

• Turn off "Improve the model for everyone" under Data Controls to stop your data being used in training.
• Turn memory on — it stores preferences and context across sessions without contributing to model training.
• Memory applies to both standard chats and advanced voice mode.
• Temporary chat disables all storage but also blocks deep research, canvas, advanced voice mode, and agent mode.

File inputs and canvas co-creation

• Accepts Word, PDF, Excel, CSV, and text files — up to 10 per chat.
• Complex Excel files (multiple tabs, many rows) perform better when exported as a single-tab CSV before uploading.
• Custom GPTs and projects accept up to 20 files in their knowledge base.
• Canvas functions like a shared Google Doc — AI writes alongside you and you can highlight any segment and request a targeted rewrite.
• Canvas also writes and previews code (HTML/CSS/JS), enabling non-developers to generate charts, animations, and graphics for presentations.

Image and video creation

• Images are created via the plus menu → "Create image"; style presets are available.
• A library stores all previously generated images for future reference.
• Sora (separate portal) generates AI video; output volume is capped based on subscription tier.

Web search and deep research

• Web search with Thinking (Extended) enabled handles most research — broad comparisons, industry overviews, current-events queries.
• Deep research takes 10–30 minutes and returns a 10–40 page report; reserve it for specific, complex multi-step tasks (e.g. cross-comparing clinical studies).
• Deep research can run against connected internal data sources (Google Drive, Dropbox, OneDrive) via connectors — not just the public web.

Study and learn mode

• Puts ChatGPT into Socratic mode: it asks questions rather than giving direct answers.
• Useful for students and professionals learning new software or building the ability to explain a product.

Custom GPTs and GPT projects

• Custom GPTs are shareable publicly or within an enterprise; useful for distributing a configured AI to others.
• GPT projects are private; they have access to agent mode, deep research, and project-specific memory — which custom GPTs do not.
• Project memory stores preferences scoped only to that project, keeping them separate from global memory.
• The system prompt / instructions is the most important element of any custom AI — spend time refining it.
• Upload example documents (e.g. past blog posts) as the knowledge base so the AI can match your style.

Sharing conversations

• Share a specific message within a conversation by selecting the share icon at that message.
• Share the entire conversation thread via the Share button in the top right.
• Sharing full threads is an efficient way to show others how to prompt effectively.

Scheduling

• Ask ChatGPT to repeat a task on a schedule (e.g. "send this every two days at 8am") — it creates a recurring job and notifies via email.
• Manage, pause, or delete schedules under Settings → Schedules.

Agent mode

• Creates a sandboxed virtual computer and lets the AI operate autonomously for 30–60 minutes.
• Most useful for: finding and testing active promo codes; researching rental properties against a defined list of constraints.
• Available only within GPT projects (not custom GPTs) and not in temporary chat.
• Plus users get ~25 runs per month; Pro users get more.

Connectors

• Connect ChatGPT to external data sources: Gmail, Google Drive, Dropbox, OneDrive, Box, and others.
• Practical example: connect Gmail and have ChatGPT summarise recent emails or draft replies.
• Custom connectors can be created via advanced settings — relevant only for developers.

Codex

• A coding-specific model accessible from the menu.
• Dedicated to high-quality code generation; use instead of standard Chat when the task is purely development-focused.

The 80/20 priorities

1. Pick the right model for each task (Auto for most, Thinking for high-stakes).
2. Turn on web search for anything requiring current information.
3. Build custom GPTs or projects for recurring use cases.
4. Use dictation instead of typing to get richer, higher-quality outputs.