Your privacy on Georgia Commons

Georgia Commons is run by Georgia Civic Data, an independent, nonpartisan, public-interest organization. This policy covers our website at georgiacommons.org (including the Spanish pages under /es), the chat panels on bill and opinion pages, our public API at api.georgiacommons.org, and the MCP connector at mcp.georgiacommons.org/mcp.

We are not affiliated with, endorsed by, or operated by the State of Georgia, the Georgia General Assembly, the Georgia courts, or any government agency. We gather public documents and make them easier to use; we are not a government service.

Questions about this policy: shane@georgiacivicdata.org.

• No accounts, no logins. You never create an account or tell us who you are to use the site, the API, or the connector.
• We do not sell your data. We do not sell or rent your information, we do not show ads, and we do not build a profile of you.
• Chat questions go to our AI provider. A question you ask in a bill or opinion chat panel is sent, with the document, to Anthropic to write the answer. We do not save the conversation.
• We count usage, not users. We keep a permanent tally of answers given, documents served, and connector tool calls, so we can report honestly how a public-interest service is used. The tally holds no question text, no IP address, and nothing that points back to you.
• Logs are short-lived. Our hosting providers keep standard request logs, which include IP addresses, for 7 to about 30 days.
• A few providers help run the service. Vercel hosts the website and its cookieless analytics, Railway hosts the API and the connector, Anthropic writes chat answers, and Vimeo plays the meeting videos. They are listed below.

Because there are no accounts, the only information tied to you is what you choose to send plus the standard technical information every website receives.

• Chat panels. When you ask a question on a bill or opinion page, your browser sends the question, the earlier turns of that conversation, and the page's language to our server, which sends them with the document to Anthropic to write the answer. We do not ask for your name, email address, or anything else about you; the only personal information we receive is whatever you choose to type.
• Browsing, search, and the API. When your browser or a program requests a page or data, our servers receive the request itself, such as the address of the page, a citation, or a bill number, plus standard technical information such as your IP address and browser type. A search puts the words you search for in the page's address, so they are part of that request.
• The MCP connector. When an AI assistant calls one of the connector's tools, it sends the tool's name and its arguments, such as a citation, a bill number, or search words. We use the arguments only to answer that call and do not store them.
• Analytics. We use cookieless analytics to count visits, as described below.
• Usage counts. Each answer, data request, and connector tool call adds one short entry to our usage counts, described below.

Bill and opinion pages have a chat panel that answers questions about that one document. To answer, our server sends Anthropic your question, the earlier turns of the conversation, and the full text of the document. The answer streams back to your browser.

The conversation lives in your browser tab's memory. It is not saved in your browser's storage or in our database, and it is gone when you reload or close the page. Our server holds a question and its answer only while it answers. It keeps the answer text in memory just long enough to count how many of its quotations appear word for word in the document; only those two counts are kept, in the usage counts below.

Two exceptions are written to our server logs. If a question looks like an attempt to make the AI ignore its instructions, it is refused, and the log line records your IP address and the question's first 100 characters. If you send questions faster than the limit allows, the log line records your IP address. These lines are kept with our other server logs (see How long we keep information).

Please do not type anyone's personal information into the chat. The documents here are public laws, bills, court opinions, and meetings; a question never needs it.

We use Vercel Web Analytics to understand which pages people visit, so we can keep the site useful. It is privacy-friendly by design: it does not use cookies, it does not identify you personally, and it does not track you across other websites. It records aggregate information such as page views, referring sites, the general type of device and browser, and an approximate country derived from your network connection.

The site also sends two kinds of anonymous events to the same analytics, each a name and nothing else. When a bill, opinion, Code section, constitution provision, or meeting page has been visible on your screen for a total of ten seconds, it counts one read, named only by the kind of document and the language of the page. When you arrive from a link in an AI assistant such as ChatGPT or Claude, it counts one referral, named only by the assistant. Neither carries an identifier, the page address, or anything else about you.

Like most online services, our servers and our hosting providers keep short-term operational logs so we can run, secure, and debug the service.

• Our backend (the API and the connector) writes one log line per request with the address requested (which for a search includes the words searched), the response status, and the network address the request came from. For a connector tool call it records the tool's name, how long it took, and whether it succeeded, but not the tool's arguments. For a chat answer it records the number of AI tokens used, not the question or the answer. The two chat exceptions above are the only log lines that contain any part of a question.
• Our hosting providers separately keep standard infrastructure logs that include IP addresses and basic connection information, which they use to route and secure traffic.

We use these logs only to operate and improve the service. Our hosting providers keep them for up to 30 days, then delete them. (Our backend host keeps them for 7 days; our website host keeps request logs for about 30 days.)

Georgia Commons is a public-interest project, so we need to be able to say honestly how much it is used: how many questions the chat panels answered, how many documents the site, the API, and the connector served, and how many connector tool calls were made. Each time one of those happens, our server records one short entry.

Each entry holds the date and time; which part of the service was used (for example, the bills chat, the Code API, or the connector); what kind of event it was (a question answered, a data request, or a connector tool call with the tool's name); whether it succeeded, with a short error code if it did not; whether the reply was in English or Spanish; and a few one-word categories: the broad kind of client (our website, a program using the API, or an AI assistant using the connector), a client family and assistant name read from the browser or program's self-description (such as browser, crawler, or AI assistant, and Claude or ChatGPT), which of our web addresses the request came to, and, for a Markdown or plain-text copy of a page, which format was read. The self-description itself is discarded after that one word is taken from it.

A chat answer also records the number of AI tokens it used, how many calls to the AI model it took, its position in the conversation (first question, second question), and how many of the passages it quoted were checked and found word for word in the document.

What these entries never contain: the text of a question or an answer, a connector tool's arguments, search words, your IP address, your browser or user agent, the page you came from, or any cookie, session, or account identifier. Nothing in them points back to you, and a conversation cannot be put back together from them. Because they are counts about the service rather than records about people, we keep them indefinitely. Georgia Civic Data's analytics system copies them each day into its private records, together with daily traffic totals from our website host (counts of visits and requests by page, referring site, country, and type of visitor, with no IP address), also kept indefinitely.

The public API at api.georgiacommons.org and the MCP connector at mcp.georgiacommons.org/mcp are open to everyone and need no sign-in or account. They are read-only: they answer requests for Georgia's public documents and do not write, change, or store anything about you.

The MCP connector lets you connect an AI assistant you already use, such as Claude, ChatGPT, or Gemini, to the documents here. When your assistant calls a tool, the connector answers from the stored documents. Each call adds one entry to the usage counts above, with the tool's name and not its arguments, and one line to our server logs, also without its arguments.

One important point about your assistant. The AI assistant you choose is operated by its own provider under that provider's terms, not by us. Whatever you type into your assistant, and the documents it retrieves through the connector, are handled by that provider, for example Anthropic, OpenAI, or Google. We do not control or receive your conversation with your assistant.

To keep the service available, our server counts requests per IP address in its memory: every request to the MCP connector, every request from a program calling the API directly, and every chat question asked on the site. For the connector, the address is usually your AI assistant provider's, not yours, because hosted assistants such as Claude and ChatGPT call the connector from their own servers. The counts cover only the last hour of requests, are never written to disk or to our logs, and are cleared from memory periodically and whenever the server restarts. A request over the limit is refused with a message saying how long to wait.

Meeting pages show the House's or the Senate's own video, played by Vimeo's embedded player. When a meeting page loads, your browser connects to Vimeo, which receives your IP address and browser information. We load the player with Vimeo's do-not-track setting on, which tells it not to track your viewing session or set cookies. See Vimeo's privacy policy.

Document pages have buttons that open Claude or ChatGPT with a prompt naming the page's address. If you use one, your browser goes to that provider's site, and what happens there is under that provider's terms and privacy policy.

Georgia Commons sets no cookies, and our analytics are cookieless. Two small notes are kept in your browser's session storage, a temporary storage area on your own device that your browser clears when you close the tab: which pages your browser has already counted as read, so a page is counted once per visit, and whether this visit already counted an arrival from an AI assistant. Neither holds personal information. Chat conversations are not stored in your browser.

• Answer your questions. In a chat panel, your question, the earlier turns, and the document are sent to Anthropic to write the answer.
• Serve the documents you ask for. The site, the API, and the connector return the public documents a request names.
• Operate, secure, and debug the service. We use logs, per-IP rate limits, and error information to keep the service running and to fix problems.
• Understand overall usage. We use the usage counts and aggregate analytics to see what is useful and report how the service is used.

We do not sell or rent your information, we do not use it for advertising, we do not use it to build a profile of you, and we do not use your questions to train any AI model of our own.

We do not sell your data. To run the service, we rely on a small number of providers that process data on our behalf:

• Anthropic (Claude). Writes the chat panels' answers. When you send a chat question, its text, the earlier turns of the conversation, and the document are sent to Anthropic's API. Anthropic processes this as a service provider under its own privacy policy, and under its commercial terms for the API, it does not use data sent through the API to train its models. The API and the connector do not use Anthropic. See Anthropic's privacy policy.
• Vercel. Hosts the website and provides the cookieless analytics described above. The website's server, on Vercel, passes chat questions and data requests on to our backend without storing them. See Vercel's privacy notice.
• Railway. Hosts our backend: the API, the connector, the chat service, the usage counts, and their logs. See Railway's privacy policy.
• Vimeo. Plays the meeting videos on meeting pages, as described above.

Some of our providers never receive anything about you. The meeting transcripts are made by ElevenLabs from the General Assembly's public videos, and the meeting summaries are written by OpenAI from those transcripts; bill and opinion summaries are written by Anthropic from the documents. We read the documents from the General Assembly's website, CourtListener, the Supreme Court of Georgia's website, and the State's published Code. None of them receives information about visitors.

We may also disclose information if the law requires it, or where we believe in good faith it is necessary to protect the safety, rights, or security of our users, the public, or the service.

Chat conversations are held only in your browser tab and, while an answer is written, in our server's memory. The usage counts are the one thing we keep long term, and they contain no conversation text and no identifier for you. We host the service with established providers that encrypt data in transit over HTTPS.

No method of transmitting or storing data is completely secure. We work to protect the little information we handle, but we cannot guarantee absolute security, so please do not send sensitive personal details through the service.

• Chat conversations. Not kept. A conversation exists in your browser tab until you reload or close it, and on our server only while an answer is written.
• Connector tool arguments and search words. Not kept by our application. Search words appear in the addresses recorded in request logs, which follow the log periods below.
• Server and request logs. Up to 30 days: 7 days on our backend host, about 30 days on our website host.
• Rate-limit counts. In memory only, covering the last hour, and gone when the server restarts. This covers the chat, the API, and the connector.
• Usage counts. Kept indefinitely. They contain no question text, no IP address, and no identifier for you or your session.
• Daily traffic totals. We keep daily counts of visits and requests, by page and by type of visitor, indefinitely. They contain no IP address and nothing that identifies you.
• Session storage in your browser. Cleared by your browser when you close the tab.

You can use all of Georgia Commons without giving any identifying information, and you control what you type into a chat panel.

• Clear a chat. Reload or close the page.
• Limit analytics. The site works fully without analytics. You can use your browser's privacy settings or a content blocker to block them.
• Ask us. Depending on where you live, you may have rights to access, correct, or delete personal information, or to object to how it is processed. Because we keep no accounts and no conversations, we usually hold no personal information tied to you. For a request or a question, email shane@georgiacivicdata.org and we will do our best to help.

We do not sell personal information, and we do not use it for targeted advertising.

Georgia Commons is a general-audience resource for public government documents. It is not directed to children, and it is not intended for use by children under 13. We do not knowingly collect personal information from children, and no personal information is needed to use the site.

If you believe a child has sent us personal information, contact us at shane@georgiacivicdata.org. Because chat conversations are not stored, such information is typically gone already, and we will promptly delete anything brought to our attention.

We may update this policy, for example if we add a feature or change a service provider. When we do, we will change the effective date at the top of this page and post the new version here, at georgiacommons.org/privacy (and /es/privacy in Spanish). If a change is significant, we will make it clear on the site.

Questions about this policy or how your information is handled: shane@georgiacivicdata.org. See also the terms of service and the support page.