ChatGPT Sites Explained: Features, Hosting, SEO Limits, and the Best Use Cases


ChatGPT Sites is a managed website creation and hosting service built directly into ChatGPT Work and Codex. It allows a user to describe a desired website, attach supporting files or visual references, review the generated application, request changes, save versions, and publish the result without configuring a separate hosting provider.
A typical AI website builder generates code and sends the user to another platform for deployment. ChatGPT Sites attempts to cover the entire workflow:
This makes Sites closer to an integrated application platform than a conventional no-code page builder.
ChatGPT Sites first appeared as a preview in the Codex desktop application on June 2, 2026. The initial release focused on creating, saving, deploying, and inspecting websites hosted by OpenAI.
On July 9, 2026, OpenAI expanded Sites through ChatGPT Work and introduced broader public publishing. This made it possible to share a Site publicly through a URL rather than limiting it to members of a business workspace.
The public beta rollout included Pro, Pro Lite, Business, Enterprise, Edu, and later Plus accounts. Free and Go accounts were excluded from the initial public beta.
Availability can still vary because of:
If @Sites appears in the ChatGPT composer, the account has access to the native Sites workflow.
ChatGPT Sites supports substantially more than static landing pages. The platform can create both content-driven websites and stateful applications.
Common website types include:
OpenAI showcase projects demonstrate React, TypeScript, Three.js, React Three Fiber, WebGPU, physics engines, responsive interfaces, persistent game progress, and generated visual assets. The practical ceiling is therefore much higher than that of template-based AI page generators.
| Feature | Current capability | Important limitation |
|---|---|---|
| AI website generation | Build from prompts, files, screenshots, links, and data | Output still requires human review |
| Managed hosting | OpenAI hosts the generated application | Public beta limits apply |
| Private preview | Review the Site before sharing it | A preview should not be treated as final QA |
| Production deployment | Publish a saved version to a live URL | Every deployed URL is a production deployment |
| Version management | Save and inspect deployable versions | Permanent deletion cannot be reversed |
| Existing projects | Deploy compatible local projects | Not every framework or hosting pattern is supported |
| Custom domains | Connect an owned apex domain or subdomain | Not available to every plan or workspace |
| Environment variables | Configure runtime values in Site settings | Changes normally require redeployment |
| Secrets | Store API keys and private configuration | Secrets must not be placed in prompts or source files |
| Relational database | Store structured records in D1 | D1 storage is limited to 10GB per Site |
| Object storage | Store uploaded files in R2 | No fixed R2 cap does not mean unlimited usage |
| Authentication | Add Sign in with ChatGPT | Authorization must be enforced server-side |
| Access controls | Owner, selected users, workspace, or public | Options depend on workspace settings |
| Analytics | Track visitors and page views | More limited than GA4 or product analytics platforms |
A Site can be started by mentioning @Sites or explicitly asking ChatGPT to build a website. A strong request should define the audience, core task, content, behavior, storage requirements, and publication constraints.
The recommended workflow is:
Specify:
Test more than the visual design. Review:
Broad prompts such as make it better produce unpredictable changes. Better instructions identify a component, breakpoint, user action, and expected result.
Saving creates a deployable candidate without necessarily updating the live production website. For local source projects, the saved version can be associated with the Git commit used during the build.
Deployment publishes the saved version to the selected audience. Because deployment URLs are production URLs, changes should be saved and reviewed before deployment.
ChatGPT Sites uses D1 for structured data that must remain available between visits. D1 is described as a relational, SQLite-compatible database.
Appropriate D1 use cases include:
Each Site currently has a documented D1 storage limit of 10GB.
Temporary interface preferences do not always need a database. A dismissed banner, selected theme, or temporary filter can often remain in browser storage. Persistent business records, user progress, and shared application state should use server-side storage.
R2 is intended for larger file objects, including:
Applications that need searchable uploads can combine the two storage systems:
OpenAI currently describes R2 as having no fixed per-Site storage limit. However, public beta account limits can still prevent additional storage from being added. No fixed limit should therefore be interpreted as elastic beta capacity, not guaranteed unlimited hosting.
A public Site can remain accessible to anonymous visitors while offering optional Sign in with ChatGPT.
This enables:
After successful authentication, Sites can forward the verified email address and, when available, the profile name to server-side application code.
Authentication and authorization must remain separate. Knowing a visitor's identity does not automatically mean that the visitor should be allowed to read or modify a record. Every protected database operation should verify ownership or role permissions on the server.
A published Site receives an OpenAI-managed production URL, commonly using a subdomain associated with chatgpt.site.
Where custom domains are available, users can connect:
example.comapp.example.comThe domain must already be registered. Sites provides DNS records that must be added through the domain registrar or DNS provider.
Important custom-domain limitations include:
Sites automatically records basic traffic information without requiring an analytics SDK.
The current analytics interface includes:
This is useful for answering whether anyone is visiting and whether usage is increasing. It is not a complete replacement for advanced analytics.
Missing or undocumented capabilities include:
Enterprise-owned Sites may not have access to the same built-in Site analytics view.
ChatGPT Sites can publish public websites and connect custom domains, but its long-term SEO suitability remains uncertain.
The official documentation does not currently establish comprehensive support for:
ads.txtA small public landing page, event guide, calculator, or interactive tool may still rank if it is crawlable, useful, fast, and connected to a custom domain. The uncertainty becomes more serious for directories or programmatic SEO projects containing thousands of URLs.
Before using Sites for an SEO-dependent project, verify:
For large SEO websites, an established stack such as Astro, Next.js, Cloudflare, Vercel, or a conventional server remains more predictable.
| Platform | Strongest use case | Main advantage | Main tradeoff |
|---|---|---|---|
| ChatGPT Sites | Rapid apps and interactive deliverables inside ChatGPT | Integrated creation, hosting, storage, authentication, and sharing | Beta limits and incomplete SEO documentation |
| Base44 | AI-generated business applications | Fast application scaffolding and managed services | Third-party account and platform dependency |
| Lovable | Visual web app prototyping | Accessible prompt-driven interface and fast front-end iteration | Production architecture may need additional review |
| Vercel | Production React and Next.js applications | Mature deployment workflow, previews, domains, and framework support | Database and application services are separate decisions |
| Cloudflare | Edge applications and globally distributed websites | Workers, Pages, D1, R2, caching, and infrastructure control | Requires more technical configuration |
ChatGPT Sites offers the shortest path from a ChatGPT conversation to a hosted interactive result. Vercel and Cloudflare still provide more infrastructure visibility, automation, portability, and control for mature production systems.
Sites is well suited to operational dashboards, weekly business reviews, project status pages, and searchable internal resources. Access can remain limited to selected users or the workspace.
A daily guessing game can combine:
This is particularly useful for validating whether a game loop is engaging before investing in independent infrastructure.
A static document can be converted into filters, charts, calculators, scenario comparisons, and drill-down views. The result is easier to explore than a long PDF or spreadsheet.
Sites can produce event calendars with date filters, transportation information, ticket details, weather alternatives, accessibility labels, maps, and saved favorites.
A founder can test a calculator, AI directory, comparison tool, submission workflow, content portal, or lightweight SaaS concept before building the final architecture.
A production-oriented prompt should request requirements, testing, review, and controlled deployment rather than only visual generation.
@Sites
Build a public, mobile-first website for the specified audience and use case.
Requirements:
- Define the primary user journey before implementation
- Create responsive desktop, tablet, and mobile layouts
- Include accessible labels, keyboard navigation, focus states, and sufficient contrast
- Use D1 only for data that must persist between visits
- Use R2 for uploaded file contents and D1 for searchable file metadata
- Keep authentication and authorization checks in server-side code
- Add loading, empty, validation, failure, and success states
- Configure page titles, descriptions, canonical URLs, social metadata, and structured data
- Test navigation, forms, database writes, uploads, authentication, mobile layout, and console errors
- Keep the Site private during review
- Save a version before deployment
- Do not publish until the final version has been reviewed
After implementation, report the supported routes, storage bindings, environment variables, test results, limitations, and publication checklist.AI-generated applications can contain broken links, weak error handling, inaccessible controls, or incorrect assumptions. The first version should be treated as a prototype.
A saved version is a reviewable candidate. A deployed version is live production output. Publishing should remain a deliberate step.
API keys and tokens should be stored in the Site settings as hosted secrets. They should not appear in prompts, attachments, source files, screenshots, or .openai/hosting.json.
Hiding a button does not protect data. Every protected read and write must be checked by server-side code.
R2 may have no fixed documented storage cap, but account-level beta limits still apply. A high-traffic Site may be restricted if it exceeds the plan's current allowance.
Some private networks, databases, background services, frameworks, and hosting patterns are not supported. Compatibility should be confirmed before migrating an existing production application.
A Site that collects identity information, uploads, comments, or form submissions may need a privacy policy, consent controls, moderation procedures, and a deletion process.
The Site creator remains responsible for the website, its functionality, uploaded content, visitor submissions, and legal compliance.
Important restrictions include:
Sites did not support data residency or inference residency at launch. Deployed code, database records, stored files, generated artifacts, and logs may therefore be processed outside a preferred residency region.
OpenAI may also display platform attribution in a Site footer or toolbar. A public Site should not imply that OpenAI endorses the website or operates the underlying business.
The answer depends on the consequences of failure and the importance of infrastructure control.
Good production candidates include:
Projects requiring more caution include:
A practical strategy is to use Sites for validation and early users, then migrate proven products to dedicated infrastructure when traffic, compliance, SEO, observability, or operational requirements justify the transition.
ChatGPT Sites turns ChatGPT into a complete lightweight website and application platform. Its most important advantage is not simply faster code generation; it is the removal of multiple deployment decisions that normally separate an idea from a usable public application.
The combination of AI development, managed hosting, D1, R2, authentication, version management, analytics, access controls, and custom domains makes Sites highly competitive for prototypes, internal tools, interactive reports, and browser games.
However, public beta limits, incomplete SEO controls, unclear high-traffic allowances, runtime restrictions, and uncertain portability prevent it from replacing established production infrastructure in every scenario.
The best next step is to build one focused Site with a real user journey, test storage and authentication, connect a custom domain if available, measure public traffic, and evaluate migration options before committing a large or SEO-dependent project to the platform.
More articles connected to the same themes, protocols, and tools.
Browse entries that are adjacent to the topics covered in this article.