GPT Store

The GPT Store is a discovery and distribution area within ChatGPT for public Custom GPTs. These GPTs are configured versions of ChatGPT that can combine instructions, uploaded knowledge, built-in capabilities, and connections to external services. They are used inside ChatGPT, rather than installed as standalone software or embedded in another website.^[1]

OpenAI announced GPTs and the planned store on November 6, 2023. The GPT Store began rolling out on January 10, 2024. OpenAI said at launch that users had created more than 3 million GPTs. That figure covered custom GPTs created on the platform, not the number of public store listings.^[2][3]

History

OpenAI introduced GPTs on November 6, 2023 as no-code, customized versions of ChatGPT. A GPT could be kept for personal use, shared within a company, or shared publicly. OpenAI initially said that the GPT Store would open later that month, with search, leaderboards, featured selections, and future payments to builders based on usage.^[2]

The November launch target was not met. On December 1, Axios reported that an OpenAI memo to developers moved the store to early 2024. The memo said that "unexpected things" had delayed it, without giving a more specific cause.^[4]

The store began rolling out on January 10, 2024. The launch post named Plus, Team, and Enterprise users in its rollout announcement, while its enterprise section said Enterprise availability would follow soon. Its initial discovery features included a community leaderboard, featured GPTs, and categories such as writing, research, programming, education, lifestyle, and image generation. Publishing initially required a GPT to be shared with everyone and a builder profile to be verified. OpenAI said that review combined automated and human checks and that users could report GPTs.^[3]

On May 13, 2024, OpenAI announced that access to discover and use GPTs would roll out to ChatGPT Free users, subject to usage limits.^[5] OpenAI's later help documentation states that GPTs are available to signed-in ChatGPT users, while creating or editing a GPT requires an eligible paid subscription and may also depend on workspace permissions.^[1]

What the store distributes

A store listing points to a configured GPT that runs within ChatGPT. It is not a downloadable model, a package of model weights, or an application deployed on the builder's own server. OpenAI distinguishes GPTs from developer-built assistants that use an API in an external product, and states that a GPT cannot be embedded in an outside website or application.^[1]

The configuration available to a builder can include the following elements:^[1][6]

Scroll sideways for more →

OpenAI's help documentation states that one GPT can use apps or actions, but not both at the same time. It also states that GPTs do not use a user's saved memory, custom instructions, or previous conversations, so each GPT conversation starts without those sources of prior context. Available models can change; when a model is retired, OpenAI may move affected GPTs to a similar current model.^[1] These product details can therefore change without the builder changing the public listing.

Discovery and access

Users can reach a GPT through the Explore GPTs interface, a direct link, or a managed workspace. Availability depends on how the builder shared it and, in an organization, on workspace policy.^[1][7]

OpenAI documents several sharing levels: invite-only access, access for people in a workspace who have the link, workspace-wide access, anyone with the link, and publication in the GPT Store. The exact options depend on the account plan and workspace settings. Public-link sharing and GPT Store publication allow users to chat with the GPT but do not give them permission to view its configuration.^[7]

A public listing may display the GPT's name, icon, description, category, capabilities, conversation starters, ratings when available, and builder-profile details. The store is therefore a discovery surface for public GPTs, while direct-link and workspace sharing remain separate distribution methods.^[7]

Publishing and review

Publishing eligibility depends on the builder's plan and, for managed accounts, workspace permissions. During publication, a builder may be asked to choose a category, review the public builder name or website, and confirm compliance with product and policy requirements.^[7]

A builder profile can include a name, a website from a verified domain, and supported verified social profiles. If a public GPT uses actions, every public action must include a valid privacy-policy URL. OpenAI says that a GPT may be checked automatically against sharing and policy requirements before broader sharing or store publication. If publication is blocked, the builder may be able to revise the GPT, keep it private, or appeal when an appeal option is available.^[7]

After publication, edits are saved as a draft until the builder updates the live GPT. Builders can also change who can access it. These controls distinguish the editable GPT configuration from the public version that users can open in the store.^[7]

Actions, privacy, and data handling

GPT builders cannot view individual conversations that users have with their GPTs. Whether OpenAI may use a conversation to improve its models depends on the user's plan and data controls. OpenAI says that data from Business, Enterprise, and Edu plans is not used for training by default; for consumer plans, training use depends on whether the user has opted out.^[1]

The data path changes when a GPT uses an app or external API. OpenAI states that relevant parts of a user's input may be sent to the third-party service and that ChatGPT may ask for approval before the request is sent or an action runs. OpenAI also states that it does not audit or control how those outside services use or retain the data they receive.^[1]

For an action, the builder supplies an OpenAPI specification in JSON or YAML that defines the server, endpoints, parameters, and operation identifiers. OpenAI documents three authentication options: no authentication, an API key, or OAuth for account-based access. Public GPTs with actions require a valid privacy-policy URL, and users may be asked to approve an action before it runs.^[8]

These mechanisms do not guarantee that a GPT's output is correct or that a third party's data practices are safe. A store listing describes the configured product and its declared connections; it is not an independent certification of every answer, API, or privacy policy.

Managed workspaces

Enterprise and Edu workspaces have controls that are separate from public-store publication. Administrators can restrict who may create or edit GPTs, which sharing levels are available, and whether users may access third-party GPTs. They can also disable apps for GPTs created in the workspace or restrict actions to approved domains.^[9]

Managed workspaces support chat, view-settings, and edit permissions for collaborators. If a GPT owner is removed through the normal workspace-removal process, ownership transfers to a workspace owner until it is reassigned. OpenAI also documents GPT conversation access through its Compliance Platform and GPT adoption metrics in workspace analytics.^[9]

Builder earnings announcement

At the January 2024 launch, OpenAI said that it planned to start a builder revenue program in the first quarter of 2024, beginning with payments to United States builders based on engagement. The announcement did not publish a payment formula or an eligibility threshold.^[3]

On March 27, 2024, OpenAI said that it was working with a small group of United States builders to test usage-based GPT earnings.^[10] This was a limited test announcement, not evidence that a general revenue-sharing program was available to all builders. Public claims about standard earnings tiers, fixed per-conversation rates, or a universal minimum-usage threshold are not supported by those two OpenAI announcements.

Independent research

The public store does not expose a comprehensive index of every GPT, which complicates attempts to measure its size. A 2025 Internetware paper combined third-party indexes with searches and recommendation traversal, then validated and deduplicated the results. Its January 2025 crawl found 722,349 accessible GPTs attributed to 292,403 developers.^[11] Those values describe the researchers' discovered dataset, not an official total or a complete census of every public or private GPT.

A 2025 CHI study used semi-structured interviews with 23 GPT users who had different levels of creation experience. The researchers found that participants often had incomplete or uncertain mental models of GPT data flows, particularly about what builders and third parties could receive. The authors explicitly described the study as qualitative and limited by its small sample, so its findings should not be treated as population estimates.^[12]

A 2026 Proceedings on Privacy Enhancing Technologies study examined 5,286 GPTs with extractable action schemas. Across 44,102 API parameters, the researchers classified about 35% as sensitive or personally identifying and found complete privacy-policy traceability for 792 GPTs, about 15% of the analyzed set.^[13] The study covered data declared in action parameters, not later retention or downstream use. It also relied on automated policy analysis limited to English-language policies, limitations the authors reported.^[13]

A 2025 IEEE Symposium on Security and Privacy study collected 786,036 records across six LLM app stores, including 663,119 GPT Store records. The authors reported misleading descriptions, privacy-policy conflicts, harmful-content behavior, and applications capable of facilitating malicious activity across the six-store dataset. They also reported that 1,643 GPTs were removed after they submitted flagged cases to the GPT Store.^[14] The authors cautioned that their dataset might not represent the full ecosystem, that platforms exposed uneven metadata, and that automated detection could miss or misclassify behavior.^[14]

Together, these studies show why store-size, usage, privacy, and security figures must be tied to a collection date and method. None provides a live, exhaustive count of public GPT Store listings.

See also

• ChatGPT Plus
• ChatGPT Enterprise
• ChatGPT plugins
• AI Agents
• Prompt injection

References

1. ^OpenAI Help Center. "GPTs in ChatGPT." help.openai.com/...8554407-gpts-faq
2. ^OpenAI. "Introducing GPTs." November 6, 2023. openai.com/...introducing-gpts
3. ^OpenAI. "Introducing the GPT Store." January 10, 2024. openai.com/...introducing-the-gpt-store
4. ^Ina Fried. "OpenAI delays launch of custom GPT store until early 2024." Axios. December 1, 2023. axios.com/...i-delays-launch-custom-gpt-store-2024
5. ^OpenAI. "Introducing GPT-4o and more tools to ChatGPT free users." May 13, 2024. openai.com/...gpt-4o-and-more-tools-to-chatgpt-free
6. ^OpenAI Academy. "Using custom GPTs." April 10, 2026. openai.com/...custom-gpts
7. ^OpenAI Help Center. "Sharing and publishing GPTs." help.openai.com/...878-sharing-and-publishing-gpts
8. ^OpenAI Help Center. "Configuring actions in GPTs." help.openai.com/...513-configuring-actions-in-gpts
9. ^OpenAI Help Center. "Managing GPT access in Enterprise and Edu workspaces." help.openai.com/...n-enterprise-and-edu-workspaces
10. ^OpenAI (@OpenAI). "We're partnering with a small group of US builders to test usage-based GPT earnings." March 27, 2024. twitter.com/...1773032605002203559
11. ^Dongxun Su, Yanjie Zhao, Xinyi Hou, Shenao Wang, and Haoyu Wang. "GPT Store Mining and Analysis." Proceedings of the 16th Asia-Pacific Symposium on Internetware. 2025. doi.org/...3755881.3755900
12. ^Rongjun Ma, Caterina Maidhof, Juan Carlos Carrillo, Janne Lindqvist, and Jose Such. "Privacy Perceptions of Custom GPTs by Users and Creators." Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. 2025. doi.org/...3706598.3713540
13. ^Juan-Carlos Carrillo, Jose Luis Martin-Navarro, Rongjun Ma, and Jose Such. "Personal Data Flows and Privacy Policy Traceability in Third-party LLM Apps in the GPT Ecosystem." Proceedings on Privacy Enhancing Technologies 2026(1): 273-295. doi.org/...popets-2026-0015
14. ^Xinyi Hou, Yanjie Zhao, and Haoyu Wang. "On the (In)Security of LLM App Stores." 2025 IEEE Symposium on Security and Privacy, 317-335. doi.org/...SP61157.2025.00117