ChatGPT Sites and GDPR: Where Your Site's Data Lives

ChatGPT Sites lets Plus, Pro and workspace accounts publish what ChatGPT builds without leaving the conversation. For a European business, the question that follows "how do I publish?" is "where does the data go?" OpenAI answers part of it in its own help center:

ChatGPT Sites does not support data residency or inference residency at launch. This includes deployed Sites, Site code, D1/R2 data and file storage, artifacts, and logs.

This article works through what that sentence means for a site that collects personal data from visitors, which questions to put to OpenAI before you publish, and what the same site looks like on an EU host. Everything below comes from OpenAI's own help center, developer guide, release notes and developer announcements, checked on 13 October 2026, and from our own published facts. It is a map, not legal advice: where a judgement is needed, it belongs to you or your data protection officer.

What a Site holds, piece by piece

OpenAI lists the ingredients itself. "Depending on how you create and publish a Site, Sites can include information from ChatGPT prompts, instructions, conversation context, uploaded or referenced files, site code, generated artifacts, hosted URLs, access settings, storage, metadata, logs, and operational data needed to host, secure, debug, or enforce policy on the Site."

For a GDPR review it helps to sort that list into three groups, because each one has a different data subject:

• What you put in. Prompts, instructions, conversation context and the files you uploaded. Usually your own business information, but it becomes personal data the moment you paste a client list, a CV or a support thread into the chat.
• What runs. The site code, the generated artifacts, the hosted URL and the access settings. Mostly not personal data, unless names or contact details are written into the pages.
• What visitors leave behind. This is the group the GDPR cares about most. OpenAI's developer guide describes two storage types a Site can use: "D1, a relational database for durable structured data" and "R2, object storage for files". Form entries, accounts, uploads and scores end up there. On top of that come the logs, and traffic statistics: "Sites records traffic automatically, so you can see how people use a deployed Site without adding an analytics SDK."

One more detail matters for access control inside your own team: "Editors can read the Site's live database data." Anyone you give edit rights also sees what your visitors submitted.

What "no data residency" means for a Site

Data residency is a promise that stored data stays in a region you chose. OpenAI does sell it for parts of ChatGPT: its residency article lists "Europe (EEA + Switzerland)" among the regions, for "Eligible API customers and new ChatGPT Enterprise/Edu customers". But the same article puts Sites on the list of unsupported features: "ChatGPT Sites, including deployed Sites, site code, D1/R2 data or file storage, generated site artifacts, and related logs, at launch." The workspace admin guide says it more briefly: "Sites is currently in beta, and is not eligible for data or inference residency."

So even an Enterprise workspace whose conversations are stored in Europe cannot extend that promise to its Sites. And OpenAI does not name the country where Sites data sits instead. Its data protection guidance tells site owners: "Do not assume a particular hosting country or transfer mechanism without checking the terms that apply to you." Its privacy policy guide goes further: "Do not copy a hosting country or transfer mechanism into your Privacy Policy unless it is accurate for your Site and agreement."

In practice that means the "where is it stored" line in your privacy notice and your record of processing has to come from your agreement with OpenAI, not from a blog post, ours included.

Who is responsible for what

OpenAI's article on data protection sets out the roles plainly. "Under the ChatGPT Sites Terms, you are the data controller of End User Data collected through your Site." OpenAI's role: "By using Sites, you instruct OpenAI to process End User Data as needed to provide and operate your Site. Where applicable, OpenAI processes that data under the OpenAI Data Processing Addendum or the Data Processing Addendum signed by your organization and OpenAI."

As controller, the duties OpenAI itself lists for you are the familiar ones: a visible privacy policy, choices and opt-in consent where needed, collecting only what the Site needs, security, answering access and deletion requests within the legal deadlines, checking transfers, consent for non-essential cookies, and extra care with children's data. On incidents: "OpenAI's breach-notification obligations are governed by the agreement and Data Processing Addendum that apply to your account. You are responsible for determining whether you must notify affected individuals, regulators, or other authorities."

One step is less clear. When someone signs in to view a restricted Site, they do so with an OpenAI account. OpenAI's Sites articles do not say which party is controller for that sign-in step, so ask (question 10 below), and make your privacy notice clear about which part of the experience is yours and which is OpenAI's.

Training settings depend on your plan

The conversations in which you build and edit a Site follow the usual ChatGPT data controls. "For ChatGPT Business and Enterprise/Edu customers: OpenAI does not use these conversations or the Site information included in them to train its models by default." For individual plans the default differs: OpenAI "may use your conversations with ChatGPT to train our models if your 'Improve the model for everyone' setting is on, which includes conversations where you use ChatGPT to create, design, edit, or manage a Site."

The part to read twice is the example that follows: "if you ask ChatGPT to edit a published Site or retrieve information from it, ChatGPT may bring relevant information from the Site into your conversation". If your Site stores visitor submissions and you work with them through ChatGPT on a Plus or Pro account, check that setting first. OpenAI's own advice: "You should only ask ChatGPT to use information that you have permission to provide."

Data a Site must not handle

OpenAI rules out some categories outright. "Sites must not process sensitive data such as Protected Health Information or payment-card data (unless solely through a third-party payment processor); target children under 13 or the applicable age of digital consent", among other uses.

Payments deserve a second look. The help article says that to "sell goods or services or collect payments through your ChatGPT Site, you can do so by using a third-party payment processor". OpenAI's responsibilities article and its developer guide, however, list "enable financial transactions" among the things Sites must not do. Until OpenAI aligns its pages, ask which rule applies to your account before you add a checkout.

Sign in with ChatGPT and your privacy notice

A Site can offer Sign in with ChatGPT. According to OpenAI's privacy policy guide, "OpenAI may share the visitor's name, email address, and profile photo, if available, with the Site after the visitor approves the sign-in. Explain this in your Privacy Policy and describe how your Site uses that information." The developer guide shows how it reaches your code: after sign-in, Sites passes the visitor's email address and, when present, full name to your server in request headers.

Restricted Sites work the same way. Invited visitors "must sign in with the account that received access", and OpenAI notes that a private Site can be opened by people outside your workspace "when their invited email address is verified, including people who did not already have an OpenAI account." For a client portal, that means each client signs in through OpenAI to see your pages. Say so in your notice, and decide whether that suits your clients.

Is Sites available in the EEA?

For Plus and Pro accounts, yes. When Sites launched on 9 July 2026, OpenAI's release notes said public publishing and the expanded beta rollout were "not available in the EEA, Switzerland, or the United Kingdom at launch". On 20 July 2026 OpenAI Developers announced: "Sites is now available in the UK, EEA, and Switzerland for Plus and Pro accounts." The help center still adds that Sites "is available in public beta for ChatGPT workspaces, Plus, and Pro accounts", "If you do not see Sites, it may still be rolling out to your account", and "Availability varies by plan and region." It is not available on Free or Go. The announcement does not mention workspaces, and we have not tested from an account in Belgium, so check in your own account. And treat availability and residency as separate questions: a feature can be offered in Belgium without its data being stored in Europe.

DPA questions to ask before you publish

If your Site will collect anything about real people, put these questions to OpenAI, or to whoever manages your OpenAI agreement, and keep the written answers with your records:

1. Which agreement and which data processing addendum cover my account: OpenAI's standard addendum for my plan, or one my organisation signed?
2. In which countries are my Site's code, its D1 database, its R2 files, its logs and its traffic statistics stored and processed?
3. Which transfer mechanism covers visitor data that leaves the EEA, and where is it written down?
4. Which sub-processors host and store Site data, and how will I hear about changes?
5. How long are Site logs and traffic statistics kept, and does the built-in traffic recording place cookies or similar identifiers on visitors' devices?
6. How do I find, export or delete one visitor's records in D1 or R2 when that visitor exercises their rights?
7. When I delete a Site, OpenAI says it is removed permanently. Are copies kept in backups or logs afterwards, and for how long?
8. What are the breach-notification commitments to me, and within what time?
9. Who in my workspace can read live visitor data, given that editors can?
10. When visitors sign in, to open a restricted Site or through Sign in with ChatGPT, exactly which fields reach my Site, and which party is controller for each step?

None of these questions implies a wrong answer. They are the questions a European customer, auditor or works council will ask you, so it is better to have the replies before launch than after.

The EU-hosted route for the same site

If the answers do not fit your requirements, the site itself is not locked in. The pages ChatGPT wrote are standard web code that any host serving files can run; only the parts that depend on the Sites platform need a replacement. The move takes four steps.

1. Keep a copy of the files first. "Deleting a site permanently removes it, and you cannot restore deleted sites." If the Site was built from a local project, the source is already on your machine: the developer guide describes Sites linking a local project and tying each saved version to the Git commit it was built from. If you built it on the web, ask ChatGPT in the same conversation for the complete files, or look for a download option on your Site; we have not confirmed that every account offers an export.
2. Replace what only Sites provides. D1 and R2 storage, Sign in with ChatGPT, connected apps and scheduled automations are platform features, so they do not move with the files. On VibeDeploy, a contact form can go through the built-in forms relay, which emails submissions to you without a form service. If the site needs a real database, your AI can run Postgres, MySQL or Redis in a container next to the site with the MCP tool deploy_app (up to 6 services with 2 CPU and 2 GiB in total). Container data is not yet included in snapshots.
3. Upload the site. Drag the folder or a ZIP into the dashboard, or paste the deploy guide from your dashboard into ChatGPT: it either sends the files itself or gives you a deploy page to open in your browser. If the project has a package.json but no build output, run npm install and npm run build first and upload the dist/ or build/ folder.
4. Move the domain. At your domain provider, replace the records Sites gave you with the records the VibeDeploy dashboard shows; SSL is issued automatically. Then remove the domain from the Site's settings and restrict or delete the old Site.

What changes when the site runs in Belgium

On VibeDeploy the contracting party is Tnet Comm.V, a Belgian company, and sites run on its own Kubernetes cluster in Belgium. A DPA is included on every paid plan, and the eight sub-processors are listed publicly. Customer site content does not pass a US processor unless you use the AI site-builder features. Every deploy is kept as a version with a restore point before it, which you can restore from the dashboard within your plan's window: 3 days on the trial, Solo and Maker, 7 on Studio, 30 on Business.

What it does not change: you are still the controller, you still need a privacy notice, a lawful basis and consent for any analytics or cookies you add, and you still answer your visitors' requests. Hosting in the EU removes a transfer question from your map. It does not make a site compliant on its own. And our own limits apply: one region in Belgium with no global edge network, and container data not yet in snapshots.

If you are comparing hosts, the list of European alternatives to Vercel and Netlify covers the options with their sources, and data sovereignty for AI-built websites explains which layers of a site carry a jurisdiction. Plans start at EUR 4 a month on the pricing page, and prices include VAT where it applies.

Sources

OpenAI pages and our own, all checked on 13 October 2026:

• OpenAI Help Center: Creating and using ChatGPT Sites
• OpenAI Help Center: ChatGPT Sites: Complying with data protection laws
• OpenAI Help Center: ChatGPT Sites: How to Prepare a Privacy Policy
• OpenAI Help Center: Understanding responsibilities for your ChatGPT Sites
• OpenAI Help Center: Managing ChatGPT Sites for your workspace
• OpenAI Help Center: Data residency and inference residency for ChatGPT
• OpenAI: ChatGPT Sites developer guide
• OpenAI Help Center: ChatGPT release notes, entry of 9 July 2026
• OpenAI Developers on X: Sites in the UK, EEA and Switzerland, 20 July 2026
• VibeDeploy: sub-processors, DPA, pricing