ChatGPT Guides

See also: Guides, ChatGPT and Prompt engineering

ChatGPT Guides is the AI Wiki gateway to how-to material for ChatGPT, the conversational assistant built by OpenAI. The article links every walkthrough, tutorial, and reference page that explains how to sign up, write prompts, build Custom GPTs, use voice and vision, work with files, run code, browse the web, run agents, and connect ChatGPT to other tools. It also collects context on pricing tiers, privacy controls, and common pitfalls. ChatGPT was first released on 30 November 2022 ^[1] and has since grown into a multimodal product that OpenAI said reached more than 800 million weekly active users by October 2025 ^[18], running on the web, on iOS and Android phones, on macOS and Windows desktop apps, inside the ChatGPT Atlas browser, and through the OpenAI API. As of mid-2026 the consumer product runs on OpenAI's GPT-5.5 model family ^[22]. The guides on this hub cover the consumer product (ChatGPT) rather than direct API integration, which has its own gateway at OpenAI API.

what does this hub cover?

The pages indexed here treat ChatGPT as a working tool. Guidance focuses on tasks an end user can actually perform: opening an account, choosing a plan, writing a prompt that returns useful output, attaching files, calling a Custom GPT, turning on memory, switching to voice, scheduling a task, and running an agent. Theory pages such as Large Language Model, Transformers, and Reinforcement Learning from Human Feedback (RLHF) are linked from this hub but live in their own articles.

how do you get started with ChatGPT?

how do you sign up?

A new user can create a free ChatGPT account at chatgpt.com using an email address, a Google account, an Apple ID, or a Microsoft account. OpenAI requires a phone number to verify the account on first sign-in, which is used to deter abuse. After verification the user lands on the chat interface and can start a conversation immediately ^[25]. The account also unlocks access to the iOS app, the Android app, and the desktop apps for macOS and Windows, all of which sync conversation history across devices when signed in.

what does ChatGPT cost in 2026?

OpenAI offers a tiered set of consumer and business plans. Pricing and quotas have changed several times since 2023, so the table below lists each plan and the model access OpenAI has publicly described ^[24].

Scroll sideways for more →

Message caps and model availability shift over time as OpenAI rolls out new models. In January 2026 OpenAI announced it would begin testing advertising in the US on the Free and Go tiers, with ads shown at the bottom of an answer when a relevant sponsored product fits the conversation and Pro, Business, and Enterprise plans excluded; the company said users must be able to trust that ChatGPT's answers are "driven by what's objectively useful, never by advertising" ^[21]. The official source of truth for prices and quotas is the OpenAI pricing page and the in-product model picker ^[24].

which devices and apps run ChatGPT?

ChatGPT runs in any modern browser at chatgpt.com. OpenAI also publishes:

• iPhone and iPad app on the App Store, with widgets, Siri shortcuts, and integration with Apple Intelligence on supported devices.
• Android app on Google Play.
• Native macOS app with a system-wide keyboard shortcut to open a focused chat window over any other application.
• Native Windows app with a similar shortcut.
• ChatGPT Atlas, a ChatGPT-native web browser that OpenAI launched on macOS on 21 October 2025 for Free, Go, Plus, and Pro users. Atlas adds browser memories that recall context from sites the user visits, and an agent mode that can research, automate tasks, and book appointments in the browsing window ^[19].

All clients share the same account, the same conversation history, and the same memory store. Users can start a conversation on the phone and finish it on the desktop.

how do you write better ChatGPT prompts?

The quality of any ChatGPT answer depends on the prompt. The hub links to long-form material on writing better prompts:

• Prompt Engineering is the full reference article on the discipline.
• Prompt engineering for text generation walks through writing assistants, summarization, and structured output.
• Prompt engineering for image generation covers prompts for image models inside and outside ChatGPT.
• 26 Principles of Good Prompts is a checklist drawn from the academic paper of the same name ^[32].
• Chain-of-Thought Prompting explains the technique of asking the model to reason step by step ^[33].
• ReAct (prompting) covers the reasoning-and-acting pattern used in agent frameworks.
• Meta Prompting explains how to use a model to write its own prompts.
• Tree of Thoughts covers a search-style prompt strategy.
• System prompt explains the role-priming text that sits above the conversation.

A few practical habits cover most everyday prompts. State the role you want the model to take. State the task in a single clear sentence. Provide context, examples, and constraints. Ask for the format of the answer (a bulleted list, a JSON object, a table). For complex tasks, ask the model to think step by step or to draft an outline before writing the final answer. When an answer is wrong, push back with a follow-up message instead of starting a new chat: ChatGPT can revise inside the same context.

what is a custom GPT?

A Custom GPT is a packaged version of ChatGPT with a name, an icon, a system prompt, optional knowledge files, optional tools (web browsing, code interpreter, image generation), and optional API actions. OpenAI launched the feature in November 2023 at DevDay ^[3] and opened a public catalog called the GPT Store in January 2024 ^[4].

how do you build a custom GPT?

A Plus, Pro, Business, or Enterprise subscriber can build a Custom GPT in a guided UI ^[30]:

1. Open the GPT builder from the sidebar.
2. Describe the GPT in natural language. The builder writes a draft system prompt, name, description, and conversation starters.
3. Refine the system prompt with explicit rules, tone instructions, and formatting requirements.
4. Upload reference files. ChatGPT will retrieve from these documents at runtime through retrieval-augmented generation, an approach covered in detail at Retrieval-augmented generation.
5. Toggle which built-in tools the GPT can use.
6. Add custom actions. These are HTTP endpoints described with an OpenAPI schema, which lets the GPT call any web API on the user's behalf.
7. Choose the visibility: private, link-only, or published to the GPT Store.

OpenAI has said that more than 3 million Custom GPTs have been created since the builder launched, spanning categories such as writing, productivity, research, programming, lifestyle, and education ^[4]. Notable examples include GPT-4 (Custom GPT), ChatGPT Classic, and many third-party tools for specific workflows.

what are apps in ChatGPT?

At DevDay on 6 October 2025 OpenAI introduced apps in ChatGPT and a companion Apps SDK, which let developers build interactive apps that run inside a conversation ^[17]. An app can surface automatically when it is relevant or be called by name: starting a message with "Spotify, make a playlist for my party this Friday" pulls the app into the chat and passes it the relevant context. The Apps SDK builds on the Model Context Protocol (MCP), the open standard ChatGPT uses to connect to external tools and data, and extends it so a developer can define both the logic and the interface of an app. Launch partners included Spotify, Canva, Zillow, Figma, Coursera, Expedia, and Booking.com ^[17]. Apps in ChatGPT are the successor to both the legacy plugin system and, for consumer discovery, the GPT Store.

what happened to ChatGPT plugins?

Before Custom GPTs, OpenAI shipped a plugin system that let third parties expose tools to ChatGPT through manifests and OpenAPI specs. The flagship example was the ChatGPT Retrieval Plugin, which let a user point ChatGPT at a private vector store of documents.

OpenAI deprecated the plugin store on 9 April 2024 and removed plugin features from ChatGPT during 2024. Existing functionality moved into Custom GPTs (for tool-style integrations), Connectors (for managed enterprise data sources), and later the Apps SDK. The plugin pages on this wiki are kept for historical reference.

how does ChatGPT handle voice, vision, and images?

ChatGPT is multimodal: it accepts and produces text, images, audio, and (in some clients) live video.

• Voice mode lets the user speak to ChatGPT and hear a spoken reply. The original voice mode launched in September 2023 with text-to-speech voices designed by professional voice actors ^[2]. Advanced Voice Mode, released in 2024, runs on a speech-native model and supports interruptions, emotion, and lower latency ^[29]. The technology is closely related to the OpenAI Realtime API, which exposes the same speech-to-speech capability to developers.
• Vision lets the user attach a photo, screenshot, or PDF and ask questions about it. ChatGPT can read text in images, describe scenes, identify objects, and analyze charts.
• Image generation is built into ChatGPT. On 25 March 2025 OpenAI shipped native 4o image generation, a model that generates images inside the conversation and replaced DALL-E 3 as the default image tool; the same GPT Image line is exposed to developers as gpt-image-1 ^[13]. Because the image model is natively multimodal, it renders legible text and follows detailed instructions better than the older hand-off pipeline. Its launch drove a viral Studio Ghibli-style photo trend that OpenAI said added one million new ChatGPT users in a single hour ^[13]. A user can ask the chat to generate, edit, and iterate on images directly in the conversation.
• Video is supported through file uploads of short clips and, on some plans, through a live video session in the mobile app. Text-to-video generation lives in the separate Sora product.

Vision and voice work together with the rest of the system. A user can speak a question, attach a photo, and receive a written or spoken answer that references both inputs.

does ChatGPT remember you?

ChatGPT has a memory feature that stores small pieces of information across conversations. When the user mentions a fact about themselves (their name, their job, an ongoing project, a writing style), ChatGPT can save it and use it later without being told again. Memory can be inspected, edited, and cleared from Settings under Personalization ^[26]. A temporary chat starts with no memory access and is not saved.

Custom Instructions are a related feature. The user fills in two text boxes that describe themselves and how they want ChatGPT to respond. Those instructions are added to every new chat as a hidden system message. Users often combine Custom Instructions with memory to lock in tone, formatting, and recurring constraints.

can ChatGPT run code and analyze data?

ChatGPT can execute Python in a sandboxed virtual machine. The feature was announced in March 2023 as Code Interpreter, renamed Advanced Data Analysis later in 2023, and rolled into the default toolset of capable models in 2024. Inside the sandbox the model can:

• read CSV, Excel, JSON, PDF, image, and audio files that the user uploads.
• run pandas, NumPy, scikit-learn, matplotlib, and other common libraries that ship with the environment.
• draw charts and return them as images.
• compress, convert, edit, and merge files.
• run small simulations and statistical tests.

For straightforward data work, this is often faster than writing a notebook from scratch. The OpenAI Codex line and the Codex (OpenAI) coding agent share lineage with this feature; the related Programming with ChatGPT guide is linked from this hub.

how do projects, canvas, and scheduled tasks work?

Alongside plain chat, ChatGPT gained several tools for organizing and automating work:

• Projects launched on 13 December 2024 ^[9]. A project is a folder that groups related conversations, uploaded files, and per-project custom instructions in one workspace, so a body of work (a research topic, a codebase, a course) keeps its own context. Projects rolled out first to Plus, Pro, and Team, then to Enterprise and Edu, and later to Free users.
• Canvas launched on 3 October 2024 as a side-by-side editing surface for long-form writing and code, with inline shortcuts such as "adjust the length," "change reading level," "fix bugs," and "port to a language" ^[6]. In the GPT-5.5 era OpenAI folded Canvas into in-chat writing blocks and code blocks: Canvas is no longer offered on GPT-5.5 Instant or GPT-5.5 Thinking, and editing now happens directly in the response ^[22].
• Scheduled Tasks launched in January 2025 ^[11]. A user can ask ChatGPT to run a prompt in the future or on a repeating schedule, for reminders, recurring reports, or monitoring a topic and notifying the user when something meaningful changes. Tasks appear on a dedicated Scheduled page. Active-task limits run from 3 on Go and 5 on Plus up to 10 on Business and Edu and 15 on Pro and Enterprise, and a task cannot run more than once per hour ^[11].

how does ChatGPT search the web?

ChatGPT Search lets the assistant fetch live information from the web. The feature was launched as SearchGPT in July 2024 ^[7], opened to ChatGPT Plus and Team in October 2024 ^[8], and rolled out to free users during 2025. Search returns linked citations alongside the model's answer, so a user can click through to the source. Typical use cases include checking news, researching products, finding recent papers, and pulling specific facts that postdate the model's training cutoff. ChatGPT chooses when to search automatically, but the user can also force a search with a button or a slash command. The ChatGPT Atlas browser extends the same live-web behavior across every page the user visits ^[19].

what is ChatGPT agent?

ChatGPT agent is OpenAI's unified agentic system, launched on 17 July 2025, that lets ChatGPT act on the web rather than just answer questions ^[14]. It merged three earlier efforts: OpenAI Operator, a research preview that controlled a remote browser to fill forms, click buttons, and complete checkouts (launched 23 January 2025 ^[10]); the deep research tool described below; and ChatGPT's own conversational reasoning. OpenAI described the result as bringing together "Operator's ability to interact with websites, deep research's skill in synthesizing information, and ChatGPT's intelligence and conversational fluency" ^[14]. With agent mode a user can ask ChatGPT to, for example, analyze competitors and build a slide deck; the agent navigates sites, filters results, runs code, performs analysis, and returns editable slideshows and spreadsheets. Agent mode rolled out to Pro, Plus, and Team, the standalone Operator preview site was retired, and its work shares infrastructure with the OpenAI Agents SDK for developers who want to build similar agents on the OpenAI API.

what is deep research?

Deep Research is an agentic capability, launched on 2 February 2025, that runs a multi-step investigation and returns a cited report ^[12]. Given a prompt, it autonomously browses the web for roughly 5 to 30 minutes and synthesizes many sources into a structured answer; OpenAI describes the agent as one that will "find, analyze, and synthesize hundreds of online sources to create a comprehensive report at the level of a research analyst" ^[12]. It first shipped to Pro on a specialized version of o3 and later moved to a GPT-5-generation model. Monthly query limits have run around 250 on Pro and 25 on Plus, Team, and Enterprise, with a small allowance for free users ^[12]. Deep research is now selectable inside ChatGPT agent as one of its modes.

what is study mode?

Study mode is a learning experience OpenAI launched on 29 July 2025 that guides a user through a problem step by step instead of handing over the final answer ^[15]. When a learner turns it on, ChatGPT asks calibrating questions, breaks a topic into stages, and checks understanding as it goes. OpenAI said the mode is "powered by custom system instructions" written with teachers, scientists, and pedagogy experts to encourage active participation, manage cognitive load, and build metacognition ^[15]. Study mode is available to logged-in users on the Free, Plus, Pro, and Business plans, with ChatGPT Edu following.

what is ChatGPT Pulse?

ChatGPT Pulse is a proactive briefing feature OpenAI launched on 25 September 2025 that delivers a set of personalized cards each morning without being prompted ^[16]. Pulse synthesizes the user's memory, chat history, feedback, and opt-in connectors such as Google Calendar and Gmail to surface what matters, then stops after a few reports rather than running an endless feed. It began as a Pro-tier preview on iOS and Android and reflects OpenAI's broader push toward assistants that work for the user asynchronously.

is my ChatGPT data used for training?

OpenAI documents data handling in its Privacy Policy and in the in-product Data Controls page ^[27]. Key points for individual users:

• By default, conversations on the consumer Free, Go, Plus, and Pro plans may be used to improve OpenAI models. The user can disable this in Settings under Data Controls. When training is off, conversations are still saved to history but are not used for model training ^[28].
• Temporary Chat is a chat mode that does not save to history, does not appear in memory, and is not used for training. Conversations may still be retained for a short period to monitor for abuse.
• Business and Enterprise data is not used for training by default ^[24].
• The user can export the full account data archive (conversations, memory, settings) from Settings.
• The user can delete individual chats, the entire conversation history, or the whole account.

For research on adversarial attacks against this kind of system, see Prompt injection and How to Steal ChatGPT-4, GPT-4 and other Proprietary LLMs.

which models power ChatGPT?

The model picker shows different model names depending on the plan. The list below sketches the main families that have appeared inside the chat product. Detailed history lives in the linked articles.

Scroll sideways for more →

Since GPT-5, OpenAI has moved away from a long list of separate model names toward a single family with selectable effort. GPT-5.5 Instant is the fast default, higher-effort "Thinking" modes spend longer reasoning, and "Pro" modes serve paid power users. OpenAI said GPT-5.5 Instant "reduces hallucination in sensitive areas such as law, medicine, and finance, while maintaining the low latency of its predecessor," and reported an 81.2 score on the AIME 2025 math benchmark ^[22]. Older versions are retired on a rolling basis, with existing chats migrated automatically: GPT-5.1 left the ChatGPT picker on 11 March 2026 and GPT-5.2 on 12 June 2026 ^[23]. Reasoning-heavy modes tend to take longer and are better at math, code, and multi-step problems; the Instant default tends to be faster and more conversational. Most users mix the two depending on the task.

what does ChatGPT connect to?

ChatGPT connects to other software in several ways:

• File uploads accept Office documents, PDFs, code, images, audio, and short video clips.
• Connectors on Business and Enterprise plans (and, for ChatGPT Pulse, on Pro) link ChatGPT to Google Drive, OneDrive, SharePoint, GitHub, Gmail, Google Calendar, and other sources, so the assistant can search across data with the user's own permissions.
• Apps in ChatGPT run third-party apps such as Spotify, Canva, and Booking.com directly in a conversation through the Apps SDK ^[17].
• Apple Intelligence integration on iOS and macOS routes Siri queries and writing tools to ChatGPT when the user opts in.
• Custom Actions in Custom GPTs let a single GPT call any HTTP API.
• Direct integrations with note apps, IDEs, and browsers exist through third-party extensions, and the ChatGPT Atlas browser brings ChatGPT to every web page.

For server-to-server integration without the chat UI, developers usually go straight to the OpenAI API, OpenAI Responses API, or Azure OpenAI Service.

what are common ChatGPT pitfalls?

A short list of issues that come up in the day-to-day use of ChatGPT:

• Hallucination. ChatGPT can write a confident answer that is wrong. The fix is to ask for citations, to enable web search for time-sensitive questions, and to verify any fact that matters before acting on it. Newer models such as GPT-5.5 Instant are tuned to hallucinate less in law, medicine, and finance, but the risk never reaches zero ^[22].
• Stale knowledge. Each model has a training cutoff. Information about events after that date is missing or guessed unless the model uses search.
• Context limits. Long conversations or large file uploads eventually fall out of the context window. The model loses track of details from earlier in the chat. Starting a fresh conversation, or summarizing the prior context into a single prompt, often fixes the problem.
• Tool confusion. When a chat has many tools enabled, the model sometimes picks the wrong one. Telling the model which tool to use, or building a Custom GPT with a narrower toolset, helps.
• Over-trust on math and code. Even with Code Interpreter, the model can write a plausible script with a subtle bug. Run the code, read the output, and treat the assistant as a fast first draft.
• Privacy slips. Pasting confidential data into a personal account is risky. Business and Enterprise plans, or a Temporary Chat with training disabled, are safer for sensitive material.
• Prompt injection. A document or web page can carry hidden instructions that try to hijack the assistant. The risk grows with agent mode and browser tools, which read untrusted pages. See Prompt injection for the threat model.

related pages on aiwiki

• ChatGPT
• ChatGPT Uses
• Programming with ChatGPT
• Fine-tune ChatGPT with Perplexity, Burstiness, Professionalism, Randomness and Sentimentality Guide
• GPT Store
• Custom GPTs
• OpenAI Apps SDK
• GPT-4 (Custom GPT)
• ChatGPT Classic
• ChatGPT Agent
• Deep Research
• ChatGPT Atlas
• OpenAI Pulse
• OpenAI Operator
• OpenAI Codex
• OpenAI o-series
• GPT-5.5
• Sora
• DALL-E
• Sam Altman
• Guides

references

1. OpenAI. "Introducing ChatGPT." OpenAI Blog, 30 November 2022. ↩
2. OpenAI. "ChatGPT can now see, hear, and speak." OpenAI Blog, 25 September 2023. ↩
3. OpenAI. "Introducing GPTs." OpenAI Blog, 6 November 2023. ↩
4. OpenAI. "Introducing the GPT Store." OpenAI Blog, 10 January 2024. ↩
5. OpenAI. "Hello GPT-4o." OpenAI Blog, 13 May 2024. ↩
6. OpenAI. "Introducing canvas." OpenAI Blog, 3 October 2024. ↩
7. OpenAI. "Introducing SearchGPT." OpenAI Blog, 25 July 2024. ↩
8. OpenAI. "ChatGPT search is now available." OpenAI Blog, 31 October 2024. ↩
9. OpenAI. "Projects in ChatGPT." OpenAI Help Center / announcement, 13 December 2024. ↩
10. OpenAI. "Introducing Operator." OpenAI Blog, 23 January 2025. ↩
11. OpenAI Help Center. "Scheduled Tasks in ChatGPT." Accessed 8 July 2026. ↩
12. OpenAI. "Introducing deep research." OpenAI Blog, 2 February 2025. ↩
13. OpenAI. "Introducing 4o Image Generation." OpenAI Blog, 25 March 2025. ↩
14. OpenAI. "Introducing ChatGPT agent: bridging research and action." OpenAI Blog, 17 July 2025. ↩
15. OpenAI. "Introducing study mode." OpenAI Blog, 29 July 2025. ↩
16. OpenAI. "Introducing ChatGPT Pulse." OpenAI Blog, 25 September 2025. ↩
17. OpenAI. "Introducing apps in ChatGPT and the new Apps SDK." OpenAI Blog, 6 October 2025. ↩
18. Wiggers, Kyle. "Sam Altman says ChatGPT has hit 800M weekly active users." TechCrunch, 6 October 2025. ↩
19. OpenAI. "Introducing ChatGPT Atlas." OpenAI Blog, 21 October 2025. ↩
20. OpenAI. "Introducing ChatGPT Go, now available worldwide." OpenAI Blog, 2025. ↩
21. OpenAI. "Our approach to advertising and expanding access to ChatGPT." OpenAI Blog, 16 January 2026. ↩
22. Zeff, Maxwell. "OpenAI releases GPT-5.5 Instant, a new default model for ChatGPT." TechCrunch, 5 May 2026. ↩
23. OpenAI Help Center. "GPT-5.5 in ChatGPT" and "Model Release Notes." Accessed 8 July 2026. ↩
24. OpenAI. "ChatGPT Pricing." pricing page, accessed 8 July 2026. ↩
25. OpenAI Help Center. "What is ChatGPT?" ↩
26. OpenAI Help Center. "Memory FAQ." ↩
27. OpenAI Help Center. "Data Controls FAQ." ↩
28. OpenAI Help Center. "How your data is used to improve model performance." ↩
29. OpenAI Help Center. "Voice mode FAQ." ↩
30. OpenAI Help Center. "Custom GPTs FAQ." ↩
31. Wikipedia contributors. "ChatGPT." Wikipedia, accessed 8 July 2026.
32. Bsharat, Sondos M., Aidar Myrzakhan, and Zhiqiang Shen. "Principled Instructions Are All You Need for Questioning LLaMA-1/2, GPT-3.5/4." arXiv:2312.16171, 2023. (Source for 26 Principles of Good Prompts.) ↩
33. Wei, Jason, et al. "Chain-of-Thought Prompting Elicits Reasoning in Large Language Models." arXiv:2201.11903, 2022. (Background for Chain-of-Thought Prompting.) ↩