AI Usage at a Glance
Apr 30, 2020
Creative GenPractice documented: Jukebox is an OpenAI research project that generates music — including singing — in a wide range of genres and styles, entirely from scratch using AI. It was published in 2020 as an open-source demonstration but was never turned into a product. It is likely no longer actively maintained.
Practice DocumentedView practice →Jul 18, 2022
ModerationPractice documented: When users ask OpenAI's image generation tools to create a picture, the system runs multiple safety checks — on the request and on the resulting image — before anything is shown. Requests for violent, explicit, or hateful images are declined automatically. The filters also block attempts to generate realistic likenesses of named public figures or images in specific living artists' styles.
Practice DocumentedView practice →Aug 10, 2022
ModerationPractice documented: OpenAI offers a free Moderation API that developers can use to automatically check whether text or images contain harmful content — such as hate speech, graphic violence, or sexual material — before allowing it into their apps. It is used both inside OpenAI's own products and by third-party developers building on the OpenAI API.
Practice DocumentedView practice →Sep 21, 2022
ProductivityPractice documented: Whisper is OpenAI's AI model for converting spoken words into written text. It works in more than 99 languages and approaches human-level accuracy. Released as open-source software in September 2022, it is widely used for transcribing meetings, interviews, podcasts, and videos.
Practice DocumentedView practice →Sep 19, 2023
ModerationPractice documented: Before releasing new AI models, OpenAI runs a structured safety-testing process. This includes a formal risk framework with defined thresholds (the Preparedness Framework), a network of outside experts who try to find dangerous capabilities (red teamers), and an open-source benchmarking tool called Evals. Safety evaluations can block a model from being released if risks are deemed too high.
Practice DocumentedView practice →Oct 19, 2023
Creative GenPractice documented: OpenAI's DALL-E models let users type a text description and receive an original image in seconds. Artists, designers, and everyday users can generate artwork, illustrations, and photos without drawing or photography skills. DALL-E 3 launched in September 2023 and was integrated directly into ChatGPT.
Practice DocumentedView practice →Oct 19, 2023
ModerationNew evidence: DALL·E 3 is now available in ChatGPT Plus and Enterprise
Evidence AddedView practice →Nov 6, 2023
ProductivityPractice documented: OpenAI lets anyone build a personalized version of ChatGPT for a specific job — like a cooking helper, a coding tutor, or a customer service bot — without writing any code. These are called Custom GPTs. Since January 2024, they can also be shared or sold through the GPT Store marketplace.
Practice DocumentedView practice →Feb 13, 2024
RecommendationPractice documented: ChatGPT can remember things about users across conversations — like their preferences, job, and communication style — and use that information to give more relevant responses over time. This feature, called Memory, launched in early 2024. Users can view, edit, or delete what ChatGPT has stored about them.
Practice DocumentedView practice →Feb 29, 2024
OtherPractice documented: OpenAI has invested in companies building humanoid robots and relaunched an internal robotics team to develop its own AI-powered robots. The goal is to create physical robots that can understand and respond to spoken instructions, eventually capable of performing work in factories and other physical settings.
Practice DocumentedView practice →Sep 24, 2024
Creative GenPractice documented: ChatGPT can now hold spoken conversations using a synthetic voice — users speak to it and it speaks back, in real time, in one of several AI-generated voices. This feature, called Advanced Voice Mode, launched in September 2024. Separately, developers can use the text-to-speech API to have any text read aloud in a lifelike voice for their own apps.
Practice DocumentedView practice →Oct 9, 2024
ModerationPractice documented: OpenAI runs an ongoing program to detect and disrupt people using its tools for harmful purposes — including running disinformation campaigns, cyberattacks, and scams. It publishes regular public reports on what it finds, making it one of the first AI companies to do so. Since the beginning of 2024 it has disrupted more than 20 operations and deceptive networks worldwide.
Practice DocumentedView practice →Oct 31, 2024
OtherPractice documented: ChatGPT can search the web in real time and give users up-to-date answers with links to original sources — similar to a traditional search engine, but in a conversational format that synthesizes information instead of just listing links. The feature launched for all users in late 2024.
Practice DocumentedView practice →Nov 21, 2024
Creative GenPractice documented: ChatGPT can write and edit stories, screenplays, poems, song lyrics, marketing copy, and other creative content on request. Writers can use it as a drafting partner or to get new ideas. This capability has been available since ChatGPT launched in late 2022 and improved significantly with GPT-4 in 2023.
Practice DocumentedView practice →Dec 9, 2024
Creative GenPractice documented: Sora is OpenAI's tool for generating short video clips from a text description or an uploaded image. Users can type a scene — "a fox walking through a snowy forest" — and Sora produces a video clip, including sound. It launched publicly in December 2024 through sora.com and a mobile app.
Practice DocumentedView practice →Jan 23, 2025
ProductivityPractice documented: OpenAI deployed Operator, an AI agent that browses the web and completes online tasks on a user's behalf by clicking buttons, filling out forms, and navigating websites. Think of it as a digital assistant that can handle online errands. Operator launched as a research preview for US ChatGPT Pro subscribers in January 2025 and was shut down on August 31, 2025, after OpenAI replaced it with the unified ChatGPT agent on July 17, 2025.
Practice DocumentedView practice →Feb 2, 2025
OtherPractice documented: Deep Research is a ChatGPT feature that works like a tireless research assistant — give it a question or topic, and it independently browses hundreds of web pages over 5 to 30 minutes, then delivers a structured, cited report summarizing what it found. It launched in February 2025 and is available to all ChatGPT users.
Practice DocumentedView practice →Mar 11, 2025
ProductivityPractice documented: The OpenAI API lets businesses and developers access all of OpenAI's AI models to build their own products. In 2025, OpenAI also launched AgentKit — a toolkit for enterprises to visually design, deploy, and manage multi-step AI agent workflows. More than 80% of Fortune 500 companies use ChatGPT Enterprise.
Practice DocumentedView practice →Mar 20, 2025
Creative GenNew evidence: Introducing next-generation audio models in the API
Evidence AddedView practice →Apr 18, 2025
RecommendationNew evidence: ChatGPT will now use its ‘memory’ to personalize web searches
Evidence AddedView practice →Jun 5, 2025
ModerationNew evidence: Disrupting malicious uses of AI: June 2025
Evidence AddedView practice →Jun 16, 2025
OtherPractice documented: In February 2026, OpenAI signed a contract with the U.S. Department of Defense to deploy its AI models in classified military environments. The deal — worth up to $200 million — covers uses like military healthcare, program data analysis, and cyber defense. It came days after the Pentagon blacklisted rival Anthropic for refusing to drop restrictions on domestic surveillance and autonomous weapons use.
Practice DocumentedView practice →Jul 17, 2025
ProductivityNew evidence: Introducing ChatGPT agent: bridging research and action
Evidence AddedView practice →Aug 27, 2025
ModerationNew evidence: Findings from a pilot Anthropic–OpenAI alignment evaluation exercise: OpenAI Safety Tests
Evidence AddedView practice →Oct 13, 2025
ProductivityPractice documented: ChatGPT can connect to dozens of external tools — like Google Drive, Outlook, GitHub, and Jira — allowing users to search documents, draft emails, create tasks, and take other actions across those services without leaving the chat. These connections are called Apps and are built on a standard called MCP (Model Context Protocol).
Practice DocumentedView practice →Oct 29, 2025
ModerationPractice documented: OpenAI uses automated detection systems, including classifiers, reasoning models, hash-matching, and blocklists, combined with human review, to enforce its usage policies across ChatGPT and its API. The policies prohibit activities such as generating child sexual abuse material, facilitating real-world violence, automating high-stakes decisions without human oversight, and violating personal privacy. Users found to have violated the policies may have their accounts restricted or terminated; in cases involving an imminent threat of serious physical harm, OpenAI may refer the matter to law enforcement after human review.
Practice DocumentedView practice →Nov 24, 2025
RecommendationPractice documented: ChatGPT includes a shopping tool that acts like a personal shopping assistant — users describe what they're looking for, answer a few preference questions, and ChatGPT recommends specific products with current prices, reviews, and images. It launched in November 2025. Recommendations are not paid placements; products are ranked by relevance only.
Practice DocumentedView practice →Jan 8, 2026
OtherPractice documented: OpenAI offers a set of AI tools built specifically for hospitals and healthcare organizations, including a version of ChatGPT designed to meet U.S. patient privacy law (HIPAA). These tools help with tasks like summarizing patient charts, drafting clinical documents, and adapting health education materials. Major institutions including Stanford Medicine, Cedars-Sinai, and Boston Children's Hospital are users.
Practice DocumentedView practice →Jan 19, 2026
RecommendationPractice documented: Since February 2026, ChatGPT shows sponsored ads to users on its free and low-cost Go tiers, matched to the topic of the current conversation. Someone asking about recipes might see an ad for a grocery delivery service. Ads are clearly labeled as sponsored and appear separately from ChatGPT's answers, which OpenAI says are never influenced by advertisers.
Practice DocumentedView practice →Jan 24, 2026
Data AnalysisPractice documented: ChatGPT has a built-in tool that can analyze spreadsheets, CSV files, and other data files that users upload, then produce charts, statistics, and written summaries — all without the user needing to know how to code. Originally called Code Interpreter when it launched in mid-2023, it was renamed Advanced Data Analysis in August 2023.
Practice DocumentedView practice →Feb 9, 2026
ProductivityPractice documented: OpenAI's Codex is an AI model that writes, debugs, and manages computer code. It powers GitHub Copilot, a coding tool used by more than 20 million software developers. Codex can be assigned a programming task and will independently write code, create a pull request, and respond to feedback — without a human writing a single line.
Practice DocumentedView practice →Feb 13, 2026
RecommendationNew evidence: OpenAI kicks off test phase for ChatGPT ads
Evidence AddedView practice →Feb 26, 2026
ProductivityNew evidence: Claude and Codex now available for Copilot Business & Pro users
Evidence AddedView practice →Feb 28, 2026
OtherNew evidence: OpenAI announces Pentagon deal after Trump bans Anthropic
Evidence AddedView practice →Mar 18, 2026
ProductivityNew evidence: GPT-5.3-Codex long-term support in GitHub Copilot
Evidence AddedView practice →Mar 19, 2026
Customer SvcPractice documented: OpenAI uses an automated chatbot to handle initial customer support requests at its own help center (help.openai.com). The bot can resolve common account and billing issues automatically. For unresolved problems, users submit a support ticket by form. There is no human live chat or phone support available to most users.
Practice DocumentedView practice →Mar 24, 2026
Creative GenPractice documented: Sora was OpenAI's tool for generating short video clips from a text description or uploaded image. OpenAI announced on March 24, 2026 that it is shutting down Sora entirely — including the app and API — citing the need to reallocate computing resources toward higher-priority products. The shutdown comes just three months after a landmark deal with Disney, which is also now cancelled.
Practice DocumentedView practice →Sora was OpenAI's tool for generating short video clips from a text description or uploaded image. OpenAI announced on March 24, 2026 that it is shutting down Sora entirely — including the app and API — citing the need to reallocate computing resources toward higher-priority products. The shutdown comes just three months after a landmark deal with Disney, which is also now cancelled.
Sora launched publicly in December 2024, with a standalone app following in September 2025 and a second-generation Sora 2 model that added synchronized audio. The app reached the top of Apple's App Store on launch day. It was discontinued on March 24, 2026, with OpenAI citing compute costs and a strategic pivot toward coding, reasoning, and business-focused products ahead of an anticipated IPO. OpenAI stated that the Sora research team will continue work on "world simulation research to advance robotics" rather than consumer video generation. Disney's $1 billion investment in OpenAI — which had centered entirely on Sora — was confirmed cancelled; according to Deadline, no money had actually changed hands as the deal was never finalized. OpenAI said it will share timelines for app and API shutdown and details on preserving user-created content.
Sora was OpenAI's tool for generating short video clips from a text description or uploaded image. OpenAI announced on March 24, 2026 that it is shutting down Sora entirely — including the app and API — citing the need to reallocate computing resources toward higher-priority products. The shutdown comes just three months after a landmark deal with Disney, which is also now cancelled.
Jukebox is an OpenAI research project that generates music — including singing — in a wide range of genres and styles, entirely from scratch using AI. It was published in 2020 as an open-source demonstration but was never turned into a product. It is likely no longer actively maintained.
ChatGPT can write and edit stories, screenplays, poems, song lyrics, marketing copy, and other creative content on request. Writers can use it as a drafting partner or to get new ideas. This capability has been available since ChatGPT launched in late 2022 and improved significantly with GPT-4 in 2023.
ChatGPT can now hold spoken conversations using a synthetic voice — users speak to it and it speaks back, in real time, in one of several AI-generated voices. This feature, called Advanced Voice Mode, launched in September 2024. Separately, developers can use the text-to-speech API to have any text read aloud in a lifelike voice for their own apps.
Sora is OpenAI's tool for generating short video clips from a text description or an uploaded image. Users can type a scene — "a fox walking through a snowy forest" — and Sora produces a video clip, including sound. It launched publicly in December 2024 through sora.com and a mobile app.
OpenAI's DALL-E models let users type a text description and receive an original image in seconds. Artists, designers, and everyday users can generate artwork, illustrations, and photos without drawing or photography skills. DALL-E 3 launched in September 2023 and was integrated directly into ChatGPT.
The OpenAI API lets businesses and developers access all of OpenAI's AI models to build their own products. In 2025, OpenAI also launched AgentKit — a toolkit for enterprises to visually design, deploy, and manage multi-step AI agent workflows. More than 80% of Fortune 500 companies use ChatGPT Enterprise.
ChatGPT can connect to dozens of external tools — like Google Drive, Outlook, GitHub, and Jira — allowing users to search documents, draft emails, create tasks, and take other actions across those services without leaving the chat. These connections are called Apps and are built on a standard called MCP (Model Context Protocol).
OpenAI lets anyone build a personalized version of ChatGPT for a specific job — like a cooking helper, a coding tutor, or a customer service bot — without writing any code. These are called Custom GPTs. Since January 2024, they can also be shared or sold through the GPT Store marketplace.
Whisper is OpenAI's AI model for converting spoken words into written text. It works in more than 99 languages and approaches human-level accuracy. Released as open-source software in September 2022, it is widely used for transcribing meetings, interviews, podcasts, and videos.
OpenAI deployed Operator, an AI agent that browses the web and completes online tasks on a user's behalf by clicking buttons, filling out forms, and navigating websites. Think of it as a digital assistant that can handle online errands. Operator launched as a research preview for US ChatGPT Pro subscribers in January 2025 and was shut down on August 31, 2025, after OpenAI replaced it with the unified ChatGPT agent on July 17, 2025.
OpenAI's Codex is an AI model that writes, debugs, and manages computer code. It powers GitHub Copilot, a coding tool used by more than 20 million software developers. Codex can be assigned a programming task and will independently write code, create a pull request, and respond to feedback — without a human writing a single line.
In February 2026, OpenAI signed a contract with the U.S. Department of Defense to deploy its AI models in classified military environments. The deal — worth up to $200 million — covers uses like military healthcare, program data analysis, and cyber defense. It came days after the Pentagon blacklisted rival Anthropic for refusing to drop restrictions on domestic surveillance and autonomous weapons use.
OpenAI has invested in companies building humanoid robots and relaunched an internal robotics team to develop its own AI-powered robots. The goal is to create physical robots that can understand and respond to spoken instructions, eventually capable of performing work in factories and other physical settings.
OpenAI offers a set of AI tools built specifically for hospitals and healthcare organizations, including a version of ChatGPT designed to meet U.S. patient privacy law (HIPAA). These tools help with tasks like summarizing patient charts, drafting clinical documents, and adapting health education materials. Major institutions including Stanford Medicine, Cedars-Sinai, and Boston Children's Hospital are users.
Deep Research is a ChatGPT feature that works like a tireless research assistant — give it a question or topic, and it independently browses hundreds of web pages over 5 to 30 minutes, then delivers a structured, cited report summarizing what it found. It launched in February 2025 and is available to all ChatGPT users.
ChatGPT can search the web in real time and give users up-to-date answers with links to original sources — similar to a traditional search engine, but in a conversational format that synthesizes information instead of just listing links. The feature launched for all users in late 2024.
OpenAI runs an ongoing program to detect and disrupt people using its tools for harmful purposes — including running disinformation campaigns, cyberattacks, and scams. It publishes regular public reports on what it finds, making it one of the first AI companies to do so. Since the beginning of 2024 it has disrupted more than 20 operations and deceptive networks worldwide.
Before releasing new AI models, OpenAI runs a structured safety-testing process. This includes a formal risk framework with defined thresholds (the Preparedness Framework), a network of outside experts who try to find dangerous capabilities (red teamers), and an open-source benchmarking tool called Evals. Safety evaluations can block a model from being released if risks are deemed too high.
When users ask OpenAI's image generation tools to create a picture, the system runs multiple safety checks — on the request and on the resulting image — before anything is shown. Requests for violent, explicit, or hateful images are declined automatically. The filters also block attempts to generate realistic likenesses of named public figures or images in specific living artists' styles.
OpenAI uses automated detection systems, including classifiers, reasoning models, hash-matching, and blocklists, combined with human review, to enforce its usage policies across ChatGPT and its API. The policies prohibit activities such as generating child sexual abuse material, facilitating real-world violence, automating high-stakes decisions without human oversight, and violating personal privacy. Users found to have violated the policies may have their accounts restricted or terminated; in cases involving an imminent threat of serious physical harm, OpenAI may refer the matter to law enforcement after human review.
OpenAI offers a free Moderation API that developers can use to automatically check whether text or images contain harmful content — such as hate speech, graphic violence, or sexual material — before allowing it into their apps. It is used both inside OpenAI's own products and by third-party developers building on the OpenAI API.
Since February 2026, ChatGPT shows sponsored ads to users on its free and low-cost Go tiers, matched to the topic of the current conversation. Someone asking about recipes might see an ad for a grocery delivery service. Ads are clearly labeled as sponsored and appear separately from ChatGPT's answers, which OpenAI says are never influenced by advertisers.
ChatGPT includes a shopping tool that acts like a personal shopping assistant — users describe what they're looking for, answer a few preference questions, and ChatGPT recommends specific products with current prices, reviews, and images. It launched in November 2025. Recommendations are not paid placements; products are ranked by relevance only.
ChatGPT can remember things about users across conversations — like their preferences, job, and communication style — and use that information to give more relevant responses over time. This feature, called Memory, launched in early 2024. Users can view, edit, or delete what ChatGPT has stored about them.
OpenAI uses an automated chatbot to handle initial customer support requests at its own help center (help.openai.com). The bot can resolve common account and billing issues automatically. For unresolved problems, users submit a support ticket by form. There is no human live chat or phone support available to most users.
Businesses around the world use OpenAI's API to build customer service chatbots powered by ChatGPT's underlying models. These bots handle customer questions, route complex issues to human agents, and operate 24/7 in multiple languages. In one documented case, a single AI assistant handled work equivalent to 700 full-time human agents.
Have evidence about OpenAI's AI practices? Submit a report.
Submit a report →AI Trace is free and nonprofit. Support our work
In February 2026, OpenAI signed a contract with the U.S. Department of Defense to deploy its AI models in classified military environments. The deal — worth up to $200 million — covers uses like military healthcare, program data analysis, and cyber defense. It came days after the Pentagon blacklisted rival Anthropic for refusing to drop restrictions on domestic surveillance and autonomous weapons use.
OpenAI announced the deal on February 27, 2026, framing it around three stated "red lines": no use for mass domestic surveillance, no use to direct autonomous weapons systems, and no use for high-stakes automated decisions without human oversight. The contract is structured so that OpenAI retains control over its safety stack, deploys via cloud with cleared OpenAI personnel involved, and excludes NSA access absent a separate agreement. However, the full contract text has not been made public, and legal experts, civil liberties organizations including the EFF, and some OpenAI employees publicly questioned whether the contract language contained meaningful loopholes. CEO Sam Altman later admitted the deal was "rushed" and amended the contract on March 2, 2026 to add clearer anti-surveillance language — though critics said unresolved ambiguities remained.
Since February 2026, ChatGPT shows sponsored ads to users on its free and low-cost Go tiers, matched to the topic of the current conversation. Someone asking about recipes might see an ad for a grocery delivery service. Ads are clearly labeled as sponsored and appear separately from ChatGPT's answers, which OpenAI says are never influenced by advertisers.
OpenAI announced the program on January 16, 2026 and began live testing on February 9, 2026, initially limited to logged-in adults in the United States. Ads are matched using conversation topic, past chat history, and previous ad interactions, but OpenAI states that individual chat logs are never shared with advertisers — only aggregated metrics like views and clicks. Ads will not be shown to users under 18, and they are blocked from appearing near sensitive topics including mental health, politics, and health. The paid tiers — Plus, Pro, Business, Enterprise, and Education — remain ad-free. Omnicom Media confirmed more than 30 client brands are participating in the initial pilot, including agencies across multiple consumer categories.
