ChatGPT is OpenAI’s AI assistant: you type or speak a request and it answers, writes, codes, and reasons through problems in a chat window. It has grown well past plain text chat. The newest features let you talk to it by voice, generate images inside the conversation, upload documents and photos for analysis, get answers based on current web pages, build custom assistants for tasks you repeat often, and revise writing in a dedicated editing view.
Voice mode: talk instead of type
Voice mode replaces typing with a spoken conversation. You tap the voice button, start talking, and ChatGPT replies out loud. It handles interruptions, keeps the thread of the conversation, and offers a choice of voices, so a long brainstorming session feels closer to thinking out loud than dictating prompts.
The change that matters most: voice now works with your files and projects. You can pull up a document, start a voice conversation about it, and ask questions, request a summary, or talk through an outline while your hands are busy. The same applies to saved projects, which carry their recent chats, sources, and instructions into the call.
Voice fits situations where typing gets in the way: talking through ideas on a commute, rehearsing a presentation, practicing a language, or explaining a problem you cannot quite phrase in writing. It is weaker where precision matters, like code or exact figures, where reading the answer on screen is still better.
Image generation: pictures inside the chat
ChatGPT can generate images from a text description without leaving the conversation. Describe what you want, get an image, then keep refining it with follow-up requests. The editing got sharper recently: generated images open in an expanded viewer where you can switch between a focused view and a canvas view, add comments on specific areas of the image, keep the versions you like, and ask for targeted changes instead of starting over each time.
This is practical for quick visual work: a blog header, a mockup for a design review, a diagram to explain an idea, a birthday card concept. The images work well for drafts and placeholders. They are not a replacement for a designer when the result has to be exact, and text inside generated images still comes out garbled often enough that you should check it before you publish anything.
File analysis: upload documents and ask questions
You can upload PDFs, spreadsheets, screenshots, and photos, and ChatGPT will read them and answer questions about what it finds. A long report becomes a summary. A contract becomes a list of the clauses that matter. A spreadsheet becomes charts and trends with an explanation of what changed.
The habit that makes this feature useful is pairing each upload with a specific question. Uploading a file and asking what it is gives you a generic overview. Uploading the same file and asking for the three risks in a proposal, explained in plain language, gives you something you can act on. It reads the content well, but it does not verify facts outside the file, so treat its answers as a reading aid rather than a source of truth.
Web browsing: answers based on current information
ChatGPT no longer answers only from its training data. It can browse the web and base answers on current pages, which means it handles questions about recent events, prices, and anything else that changed since its training cutoff. When it browses, it cites the pages it used, and you can open those pages yourself to check the claim.
Use this as a starting point rather than a final answer. Ask for current information, then verify anything that matters against the cited source. Browsing makes ChatGPT far more useful for research, travel planning, and shopping questions, but the citations exist for a reason: the summary can miss nuance that the original page has.
Custom assistants: your own reusable helpers
Custom assistants, called Custom GPTs, let you set up a version of ChatGPT with fixed instructions, tone, and reference material, then reuse it whenever you need the same kind of work. A writing assistant that follows your style guide, or a study coach that quizzes you chapter by chapter. Some people build a coding reviewer that checks pull requests the same way every time.
Building one is mostly a matter of writing clear instructions and uploading anything it should know: documents, examples, rules. You do not need to repeat your preferences in every chat because they live inside the assistant. OpenAI runs a gallery where people share assistants they made, and you can keep yours private if the material is sensitive.
An editing view for writing and code
Longer drafts get their own editing view. When you ask ChatGPT to help with a document or a piece of code, it can open the text in a separate panel beside the chat. You select a paragraph, ask for a specific change, and only that part gets rewritten. There are shortcuts for common tweaks like shortening a passage or adjusting the tone, and the chat stays open for discussion about the draft.
This beats copying text back and forth. You keep one conversation, the draft stays in one place, and every change happens where you can see it. It is the least flashy of the features here and, for regular writing work, the one people reach for most.
What you get on each plan
The feature set depends on your plan. Free accounts get everyday text chat with limited access to images, voice, and advanced tools. Paid plans raise those limits and include the full versions of voice, image generation, deep research, and custom assistants. New features usually arrive on paid tiers first and spread to free accounts later, so a missing feature may simply be rolling out.
A sensible order to try them: voice first, because it changes how you use the tool day to day; file uploads second, because they turn ChatGPT into a reader for work you already have. Image generation can wait until you have a visual job that used to need a separate tool. Browsing needs no setup, and custom assistants earn their place once you catch yourself repeating the same instructions.
OpenAI documents every release on the ChatGPT What’s New page, which is the reliable place to check what changed recently. Setup questions are answered in the OpenAI Help Center, and OpenAI’s official site covers plans and announcements. Independent reviews of major releases appear in the tech press, including The Verge.
