Complete practical guide

Guide for ChatGPT: Learn it, use it, verify it.

A clear, practical guide to conversations, prompts, files, images, voice, web search, research, coding, memory, projects and safe AI use.

English defaultMobile friendly30 chaptersUpdated October 2026

ChatGPTOfficial OpenAI service reference

01–10

Foundations & tools

GPT, ChatGPT, prompts, files, images, coding, search, research and voice.

11–20

Mastery & workflows

Safety, prompting, study, business, websites and evidence-based work.

21–28

Reference library

Research, files, visual prompting, debugging, templates, privacy and quick reference.

Guide contents

What is GPT?

GPT refers to a family of generative AI models designed to understand instructions and produce useful outputs such as text, code and other supported content. ChatGPT is the conversational product built around these models and its available tools.

Think of GPT as a language-and-reasoning engine rather than a conventional database. It can generate a convincing answer without guaranteeing that every statement is correct.

What is ChatGPT and how does a chat work?

ChatGPT is designed for conversational interaction. You can ask a question, refine it with follow-ups, provide context, attach supported files or images, and use available tools depending on your account and settings.

State the task and desired outcome.

Add the context ChatGPT needs.

Review the response instead of assuming it is automatically correct.

Follow up with corrections, examples or constraints.

What can ChatGPT do?

Available capabilities vary by plan, region, model, device and settings. OpenAI currently documents question answering, writing, reasoning, translation, web search, deep research, image input/generation, file work, data analysis, voice, Canvas, memory, Projects and custom GPTs among supported capabilities.

Write & rewrite

Draft, summarize, translate, polish, brainstorm and restructure.

Learn

Ask for explanations, examples, quizzes, study plans and step-by-step help.

Create

Work with text, code, images and other supported formats.

Research

Use search or deep research where available for current, source-backed work.

Analyze

Use uploaded files and data-analysis tools when available.

Talk

Voice can support natural spoken conversations where available.

Prompting: the skill that changes results

A good prompt makes the task, context, constraints and output format clear. You do not need magic keywords; you need useful information.

Weak

Make a website.

Stronger

Create a responsive landing page for a food guide. Audience: beginners. Tone: professional. Include hero, search, category cards, FAQ and mobile navigation. Return a single self-contained HTML file.

Prompt formula

Goal: What should be done?

Context: What does the model need to know?

Constraints: What must or must not happen?

Format: What should the final answer look like?

Quality check: What should be verified before finishing?

Working with files

ChatGPT can work with supported documents, spreadsheets, presentations, PDFs and other files. Depending on the available capability, you can ask it to summarize, extract information, compare documents or analyze data.

For important documents, ask for citations, page references, extracted quotations or a structured checklist so you can verify the result against the original file.

Images: understand and create visual content

Supported ChatGPT experiences can analyze uploaded images and can also generate or edit images from natural-language instructions. Availability varies by account and feature access.

Create a clean product hero image for a beginner AI guide. Use a dark modern background, one glowing AI orb, strong contrast, and plenty of empty space for a headline.

Coding with ChatGPT

ChatGPT can help explain code, find bugs, refactor, generate tests, design APIs and reason about architecture. Treat generated code as a draft: run tests, inspect dependencies and review security-sensitive changes.

Good workflow

Share the relevant code and environment.

Describe the bug or desired behavior.

Ask for a minimal change first.

Run tests and report the exact error.

Useful prompt

Review this function. Identify the bug, explain why it happens, propose the smallest safe fix, then provide tests covering the original failure and two edge cases.

Web search and current information

When search is available, ChatGPT can look up current information and provide citations. Search is especially useful for changing facts such as product details, current events, schedules, prices and recent documentation.

For high-stakes or time-sensitive claims, verify the cited source and check the date. A fluent answer is not proof of accuracy.

Deep research and structured research

Deep research is intended for multi-step research tasks: it can gather and synthesize information from multiple online sources into a cited, structured result when the feature is available.

Define the research question and audience.

Set the scope: date range, geography, sources and exclusions.

Ask for evidence, citations and uncertainty.

Review the primary sources behind important claims.

Voice, memory and context

Voice enables spoken interaction in supported experiences. Memory can allow ChatGPT to remember useful information you have shared when the feature is enabled. Conversation context also matters: a follow-up can build on earlier turns in the same chat.

Voice

Speak naturally, listen to responses and continue a conversation without starting over.

Memory

Useful remembered facts can personalize future conversations; controls let you manage memory.

Context

Keep related work together and restate critical constraints when accuracy matters.

How to use ChatGPT — step by step

This section explains the everyday workflow from opening ChatGPT to getting a useful, verifiable result. The exact buttons and available tools can vary by plan, platform, model and rollout.

Start a new chat: Open ChatGPT and begin a new conversation when the task is unrelated to previous work.

Write the goal: Say what you want done, not just the topic. Example: “Explain photosynthesis to a Class 8 student in simple English.”

Add context: Include your audience, source material, current situation, deadline, platform or other facts that affect the answer.

Choose the right tool: Use normal chat for general work, Search for current information, Deep Research for multi-source research, file upload for documents, image tools for visual work, and voice when spoken interaction is useful.

Specify the output: Ask for a table, checklist, steps, code, email, summary, lesson or another exact format.

Review the result: Check important facts, dates, calculations, citations and assumptions.

Follow up: Tell ChatGPT what to change. You can say “make it simpler,” “use examples,” “keep the same structure,” or provide the exact error.

Save the useful result: Keep important work organized in the appropriate chat, project or file workflow available to you.

Simple example

Explain the water cycle for a 12-year-old. Use simple English, 5 steps, one everyday analogy, and a 5-question quiz at the end.

Better professional example

Review this project brief. First extract requirements. Then identify missing information and risks. Do not invent facts. Finally create an implementation checklist with acceptance criteria.

Best habit: treat the first answer as a working draft. Give feedback, verify important claims, and iterate until the result meets your actual requirement.

ChatGPT settings — what each area is for

Settings control how ChatGPT behaves, how conversations and personalization work, and which product features are available. Names and locations of options can change as the interface evolves, so use this section as a practical map rather than a promise that every account will show every item.

Account & profile

Manage account-related information and access the settings area. Some workspace or subscription options may also appear here.

Appearance

Control visual preferences such as theme or other interface appearance options when offered by your app or web interface.

Language

Choose supported interface language options where available. The language of the interface and the language of your conversation are not necessarily the same.

Personalization

Set preferences that help ChatGPT adapt its responses, such as communication style or other personalization controls available to your account.

Custom instructions

Provide standing guidance about what ChatGPT should know about you and how you want responses written. Keep instructions clear and avoid putting secrets in them.

Memory

Manage whether ChatGPT can use remembered information to personalize conversations. Review, delete or manage saved memories using the controls your account provides.

Data controls

Review available controls for conversation history, data use and related privacy choices. Exact options depend on the current product and account.

Security

Review account security options such as password, sign-in methods or multi-factor authentication when available.

Notifications

Manage product notifications where the interface provides these controls.

Voice

Where available, choose voice-related preferences and permissions. Microphone permission is controlled by your device as well.

Connected apps / integrations

Some accounts can connect supported services or apps. Only connect services you trust and understand what access they provide.

Subscription & billing

Where applicable, manage plan-related information, billing and subscription settings. Availability depends on how the account is subscribed.

Custom instructions: a useful setup

About me: I prefer clear, practical explanations. My experience level is [beginner/intermediate/advanced]. My usual goals are [study/work/coding/etc.].

How to respond: Use headings and short steps. Explain unfamiliar terms. State assumptions. For important current claims, provide sources when available. Do not invent missing information.

Settings do not replace good prompts. A standing preference can guide style, while each individual prompt should still contain the task-specific context and requirements.

Limits, accuracy and safety

ChatGPT can make mistakes. It can produce incorrect facts, misunderstand an instruction, miss context or present uncertain information too confidently. Important information should be checked against reliable sources.

Verify when

Medical, legal, financial, safety-critical, identity, security, current-event or high-impact decisions are involved.

Improve reliability

Give sources, dates, constraints and examples; ask it to distinguish facts from assumptions; independently verify critical claims.

Advanced prompting patterns

Once basic prompting is comfortable, use reusable structures that make the desired process explicit.

Role + task

“Act as a technical editor. Rewrite this documentation for beginners.”

Critique + revise

Ask for weaknesses first, then request a revised version that addresses each one.

Constraints

Set length, audience, tone, required sections, forbidden elements and output format.

Examples

Provide one or two examples of the style or structure you want.

Checklists

Ask for a final checklist that confirms each requirement was met.

Iteration

Build complex work in stages rather than requesting everything in one vague prompt.

Real-world prompt examples

Study

Teach me photosynthesis as if I am 14. Start with a simple explanation, then give a labeled process, three common misconceptions, five quiz questions, and an answer key.

Business

Turn these rough notes into a one-page project plan. Include objective, deliverables, milestones, risks, owners and acceptance criteria. Flag missing information instead of inventing it.

Website

Audit this landing page for mobile usability, accessibility, content clarity and broken interactions. Return findings by severity and give exact fixes.

Research

Research this topic using recent reliable sources. Separate verified facts, source interpretations and unresolved uncertainty. Cite every important current claim.

Common mistakes

Too vague: “Make it better.” Say what better means.

No audience: A beginner guide and expert report need different explanations.

No output format: Specify table, bullets, JSON, code, checklist or prose.

Trusting first output: Ask for a review and verify important claims.

Changing too much at once: Make controlled edits and test after each meaningful change.

Ignoring dates: For current topics, state the relevant time period and verify sources.

Final ChatGPT checklist

Before asking

Define your goal, audience, context and constraints.

While working

Give feedback, examples and corrections. Break large tasks into stages.

Before trusting

Check important facts, dates, calculations, citations and assumptions.

For current info

Use search when available and inspect the sources.

For files

Ask for page/section references and compare critical results with the original.

For code

Run tests, inspect dependencies and review security-sensitive changes.

How to get consistently better answers

The biggest improvement usually comes from giving ChatGPT a clear job, enough context, a definition of success, and a way to handle uncertainty. Treat the conversation like a working brief rather than a single question.

1. Define the outcome

Say exactly what finished means. Instead of “help with my website,” say “audit the homepage, list broken interactions, propose fixes, then provide the corrected HTML.”

2. Give relevant context

Include audience, platform, existing work, constraints, deadlines, source material and what has already been tried.

3. Set boundaries

Tell it what not to change. This is especially important for code, documents and production websites.

4. Define the output

Ask for the exact structure: headings, tables, checklist, code file, JSON, short answer or detailed report.

5. Add verification

Ask it to check every requirement before finishing and to clearly identify anything it could not verify.

6. Iterate

Use follow-up instructions to correct direction. Complex work is usually easier to manage in small, testable stages.

Template: Goal: [outcome]. Context: [relevant facts]. Audience: [who]. Constraints: [must/must not]. Output: [format]. Verification: [what to check]. If information is missing, ask or flag it instead of inventing it.

Prompt engineering patterns

Prompt engineering is not about secret words. It is about communicating a task precisely and making the desired reasoning process, evidence standard or output structure explicit.

Zero-shot: Give the task directly when the task is simple.

Few-shot: Provide examples when you need a specific style, classification or format.

Decomposition: Break a complex job into research, draft, review and finalization.

Self-check: Ask for a requirements checklist and a list of uncertain claims.

Contrast: Ask for two approaches with trade-offs instead of one unexplained answer.

Transformation: Give source material and specify exactly how it should be converted.

Example

Analyze these product requirements. First extract the requirements into a checklist. Then identify contradictions and missing information. Next propose an implementation plan. Finally produce a concise acceptance-test checklist. Do not invent requirements.

Using ChatGPT for study and teaching

ChatGPT can act as a tutor, practice partner, explainer and feedback assistant. The strongest learning workflow makes you participate instead of only reading generated answers.

Learn

Ask for a concept at your level, then request an analogy, example and short quiz.

Practice

Ask for one question at a time and require yourself to answer before seeing the solution.

Feedback

Paste your answer and ask for errors, missing concepts and a corrected example.

Revision

Turn notes into flashcards, a revision schedule, practice questions and a final self-test.

Teach me [topic] for a [level] learner. Start with the simplest mental model. Ask me three diagnostic questions first. Then teach one concept at a time, give an example, ask me to explain it back, and correct my misunderstandings.

Using ChatGPT for work and business

For professional work, give ChatGPT the same information you would give a human colleague. Clearly separate facts, decisions, assumptions and items that still need confirmation.

Planning

Convert goals into milestones, owners, dependencies, risks and acceptance criteria.

Meetings

Turn notes into decisions, action items, owners, deadlines and unresolved questions.

Writing

Draft emails, reports, briefs and documentation while preserving the intended audience and tone.

Analysis

Ask for patterns, anomalies and explanations, but validate important numbers against the source data.

Do not paste confidential or regulated information unless your organization's rules and the product's available privacy controls permit that use.

Building websites and digital products

ChatGPT is particularly useful when a website task is treated as an engineering workflow: inspect first, make a bounded change, test it, then audit the result.

Audit: Understand the current files, routes, components, assets and deployment setup.

Plan: Define exactly which user problem the change solves.

Implement: Make the smallest coherent change.

Verify: Test mobile, desktop, navigation, forms, images, console errors and edge cases.

Regression check: Confirm existing working sections still work.

Deploy: Only after the local or staging checks pass.

Audit this website before editing it. Do not change existing working features. List routes, major components, data sources, image mappings and known errors. Then propose a bounded fix plan. After changes, verify navigation, responsive layout, broken links, duplicate images and console/runtime errors.

Research, sources and evidence

Research quality depends on the quality and freshness of the evidence. For changing topics, ask for dates and sources. For important claims, prefer primary documentation, official records, original papers or direct statements where appropriate.

Strong research request

Specify the question, date range, geography, population, exclusions, preferred source types and citation requirements.

Evidence review

Separate directly supported facts from interpretation, inference and unresolved uncertainty.

Research [topic] for [audience]. Focus on information current as of [date]. Prefer primary and authoritative sources. For each major claim, provide the source, publication date and a short explanation of what the source actually supports. Clearly label uncertainty and conflicting evidence.

Files, PDFs and structured documents

When working from a document, ask ChatGPT to stay anchored to the supplied material. This reduces the chance that outside knowledge gets mixed into a document-specific answer.

Extraction

Extract names, dates, totals, headings or requirements into a structured table.

Comparison

Compare two versions and identify additions, removals and changed wording.

Audit

Check a document against a supplied policy, rubric or checklist.

Summary

Request an executive summary plus detailed section-by-section notes.

Use only the supplied document for this answer. Extract the key requirements into a table with section/page references. Mark any item that is ambiguous or missing. Do not fill gaps from assumptions.

Image and visual prompting

For image generation or editing, describe the subject, composition, environment, lighting, mood, camera or illustration style, important objects, text requirements and things that must not appear. Iterative edits are often easier than one giant prompt.

Subject: What is the main thing?

Composition: Close-up, portrait, wide shot, centered, rule of thirds, empty space.

Environment: Location, background, time and atmosphere.

Visual direction: Photorealistic, editorial, cinematic, diagrammatic or another clear style.

Constraints: Objects to include, exclude, or keep consistent.

Iteration: Change one important variable at a time when refining.

Debugging and problem solving

When something fails, provide the exact error, environment, expected behavior, actual behavior and the smallest reproducible example. Guessing from “it does not work” wastes time.

Here is the exact error: [paste]. Environment: [OS/framework/version]. Expected behavior: [x]. Actual behavior: [y]. What I already tried: [steps]. First explain the likely cause. Then give the smallest safe fix. Do not rewrite unrelated code. Finally give a verification checklist.

For production systems, remove secrets, API keys, passwords and private tokens before sharing logs or code.

Prompt library: copy-ready templates

Summarize

Summarize this for [audience]. Start with 5 key points, then explain the important details. Preserve numbers, dates and qualifications.

Rewrite

Rewrite this for [audience] in a [tone] tone. Keep the meaning and facts unchanged. Improve clarity and structure.

Compare

Compare A and B by features, cost, limitations, risks and best-fit use cases. Do not declare a winner; explain the differences.

Audit

Audit this against the following checklist. For every failed item, give the exact evidence and a specific fix.

Plan

Turn this goal into phases, tasks, dependencies, risks, verification steps and a definition of done.

Teach

Teach me this step by step. Check my understanding after each major concept and adapt the explanation to my mistakes.

Research

Research this question using reliable current sources. Cite important claims and separate facts from interpretation.

Code review

Review this code for correctness, maintainability, performance and security. Prioritize concrete bugs before style suggestions.

Privacy, security and responsible use

Use good data hygiene. Before sharing content, consider whether it contains passwords, API keys, payment information, private identifiers, confidential business material or information about other people.

Remove secrets

Redact API keys, access tokens, passwords, private certificates and session cookies.

Minimize data

Provide only the information necessary for the task.

Verify outputs

Do not use generated content as the sole authority for high-impact decisions.

Respect permissions

Only analyze or transform files and information you are authorized to use.

Master workflow: from idea to verified result

For large projects, use a repeatable loop rather than trying to solve everything in one prompt.

Discover: Explain the goal and ask what information is missing.

Plan: Produce a task breakdown and definition of done.

Build: Complete one bounded stage at a time.

Review: Ask for a critical audit against the requirements.

Test: Verify the result in the real environment or against the source material.

Refine: Fix only confirmed issues, then run the final regression check.

You are helping me complete a multi-step project. Maintain a requirements checklist throughout. Before each major change, state what will change and what will remain untouched. After each stage, verify the checklist and report failures. Never claim a test passed unless it was actually performed.

Quick reference: choose the right approach

Simple question

Ask directly and specify the desired depth.

Current fact

Use current search or authoritative sources and check dates.

Large document

Upload the file and request structured extraction with references.

Complex project

Use staged planning, implementation, testing and audit.

Code problem

Provide the exact error and a minimal reproducible example.

Creative work

Specify audience, style, constraints and examples, then iterate.

High-impact decision

Use ChatGPT for organization and analysis, but independently verify critical facts and obtain appropriate professional advice.

Learning

Ask for explanations, questions, practice and feedback rather than only final answers.

COMMANDS & SHORTCUTS

ChatGPT commands & image editor actions

Practical reference for documented shortcuts and editor controls. Availability can vary by platform, plan and rollout.

/Search

Documented slash shortcut for selecting Search in supported ChatGPT experiences.

/Deepresearch

Documented shortcut for starting Deep Research where supported.

@Deepresearch

Current interfaces may also support invoking Deep Research with an @ mention.

Study Mode

Study Mode has its own documented entry point and can vary by interface.

Select

Image editor action for selecting a specific region.

Add / Remove

Describe what should be added to or removed from an image region.

Edit / Resize

Describe changes to existing content or adjust size where supported.

Undo / Redo / Save

Standard image-editor controls for revising and saving an edit.

Editor actions are interface controls, not slash commands. Do not count capabilities as commands.

FAQ