FAQ

Only what you give it. An AI tool sees what you type or say, the files you attach, what’s in the project folder you open for it, and anything you connect it to. Nothing else. What happens next depends on your plan. On free and individual paid plans, chats can be used to train models if the setting is on. Check yours and turn it off (in Claude, Settings > Privacy). Business plans generally don’t use your chats for training. Either way, red information never goes in. Will They Train on Our Data? has the details for each major tool, and the Safe Use Framework explains the colors.

Whether you can use AI for office work is mostly a question for your office’s policy. On October 2, 2026, we checked Online Sunshine and the list of 2026 Laws of Florida and found no Florida law that sets rules for how government offices use AI. The laws you already follow still apply, like public records and the confidentiality of voter information. (Safe Use has the details.) Your Supervisor decides which tools are approved and how they’re used, and your office’s attorney can tell you whether any state or county rules apply to a specific use. If your office doesn’t have a policy yet, the sample policy on For Leadership is a starting point. Until there is one, stick to green (public) information and ask before you use AI for anything official.

They can be. Florida law defines public records to include material made or received in connection with official business, “regardless of the physical form” (s. 119.011(12), F.S.). So a chat you used to draft office work may be a public record, just like an email or a draft memo. How long to keep it depends on what it is, not on the fact that AI was involved, so ask your records management liaison officer (RMLO). More in AI conversations can be public records.

It will be, sometimes. AI tools can state a wrong date, a made-up fact, or another state’s rule in the same confident tone as a right answer. The fix is you. If you could do the job by hand, you’re the best person to judge what the AI made, so read every page it makes, the way you’d read a new coworker’s first draft. The mistakes stand out to someone who knows the work. If it’s outside what you know, hand it to someone who does know it. Asking for a list of sources helps you look things up, but it doesn’t replace your read. If a mistake does get out, correct it the way you’d correct any other error. More: you’re the expert, that’s the check.

No. AI drafts; people decide. It doesn’t know your county, your voters, or your procedures the way your staff do, and it can’t be held accountable for anything. What it can do is take hours off routine work: two branded poll worker decks built from the state manual take about 4 to 6 hours by hand, and in a test run the AI’s part took about 20 minutes. That time goes back to the work only people can do.

Sometimes. Everything on this site is a folder project. In Claude, folder projects happen in the desktop app’s Code tab, which needs Pro or higher. OpenAI includes Codex on the free ChatGPT plan, with lower limits, but working in a local folder depends on what’s turned on for your account.

Free plans aren’t for yellow information either: free and other consumer plans can use your chats to train models if the setting is on (check yours and turn it off), and they don’t give your office any admin controls. They also run out of usage fast. See Pricing & Plans and Will They Train on Our Data?

Avoid it. Office business done in a personal account creates public records your office can’t easily search, keep, or produce, and the privacy settings are yours to manage, not your office’s. Ask your Supervisor or your IT staff for an approved work account. Learning on your own time with public information is a different thing; just don’t do office work there.

If your county uses Microsoft 365, you may already have Microsoft Copilot Chat at no extra cost. Look for the green shield. It means enterprise data protection is on: your prompts stay inside your county’s Microsoft 365 service and aren’t used to train Microsoft’s AI. The paid Microsoft Copilot (formerly Microsoft 365 Copilot) can also search your email and files, so your SharePoint permissions matter. (Your county’s Microsoft 365 account is called a tenant.) Ask IT which one you have, and see Microsoft Copilot.

No, and it shouldn’t. Exports from your voter registration system can include information that’s confidential and exempt under s. 97.0585(1), F.S. (Social Security numbers, driver license and Florida ID numbers, where someone registered, and everything about preregistered 16- and 17-year-olds), plus signatures and some voters’ addresses that the law also protects. Even a list with only public fields is yellow for AI purposes, because a filter can fail and a voter list in an AI project folder or chat is hard to explain later. The red list covers the rest. If you do need AI’s help with a file from any system, first delete every column that holds personal information (addresses, dates of birth, phone numbers, emails, ID numbers, and any name the job does not need), then put only the cleaned file in the project folder. The full export stays out. (Clean the file first.)

Yes, it can. AI tools sometimes cite statute sections or subsections that don’t exist, quote an old version, or mix in another state’s law. Always open the statute on Online Sunshine, the Legislature’s official site, and read the current text yourself. Every statute on this site links there. For anything that affects a legal decision, ask your office’s attorney.

For your office’s own communications, that’s mostly a decision for your office’s policy. One reasonable approach is to label AI-generated or AI-altered images, audio, and video, since people may take them for real, and to treat AI-assisted text like any other draft a staff member reviewed and approved. The sample policy has a labeling section you can adapt. Florida requires a disclaimer on certain AI-made political ads that show a real person doing something they didn’t do, when made to hurt a candidate or mislead about a ballot issue (s. 106.145, F.S.). On its face the law is about political advertising. If you are not sure whether it applies to something specific, ask your office’s attorney.

Save the evidence (the link, a screenshot, and the date). Don’t repost it, and don’t tag anyone. Take it to your Supervisor and your office’s attorney, who decide what to report and where. Florida has a disclaimer law for certain AI-made political ads (s. 106.145, F.S.) and a state rule on election fraud complaints (Rule 1S-2.025, F.A.C., revised this fall, so read the current text for contact details). Complaints about a missing disclaimer go to the Florida Elections Commission (s. 106.145(4)(b), F.S.). More in Misinformation response.

Yes. The U.S. Election Assistance Commission (EAC) keeps a page of AI resources for election officials, including an AI toolkit and case studies from offices that use AI. Some examples use staff or poll worker information. On this site that is yellow, so clean the file first. CISA, the federal cybersecurity agency, offers no-cost election security services, such as tabletop exercises (practice drills) and a security toolkit.

Install the Claude or ChatGPT desktop app (check with IT first on a county PC), make a folder for your first project, and try the two poll worker decks prompt on Getting Started. It uses only public information, and in a few minutes you’ll have two decks to click through. Folder Projects explains how the folder part works, and the Small Office Quick Start has more to try in your first week. Read the Safe Use Framework before you put anything in a folder. Supervisors and managers should also read For Leadership and its one-page decision brief.

A systems administrator with 21 years at a Florida county Supervisor of Elections office built it as a personal project. It’s not an official county or state website, and it doesn’t speak for any Supervisor of Elections office. It isn’t sponsored by any AI company, and everything on it uses public information only. More on the About page.