Misinformation response: quick, factual, cited replies

• The Claude desktop app (Code tab) or the ChatGPT desktop app (Codex) and a folder called something like Rumor Responses. Keep using the same folder, so your past corrections are there next time.
• The claim, in your own words. Save a screenshot of the original post with your office’s records, not in this folder.
• Your approval path: who signs off, and how to reach them quickly.

Every correction follows the same three steps:

1. Acknowledge. “We’ve seen questions about [topic].” This tells people you’re paying attention without repeating the rumor.
2. Correct, with a source. State the accurate information plainly, and link to where it comes from.
3. Point to official information. Send people to your website or phone line, so the next question comes to you.

1. Screenshot and save the post with your office’s records.
2. Open your Rumor Responses folder in the app. Claude: Code tab, then under the message box choose Local and Open folder…. ChatGPT: open Codex, click Add new project (or press Ctrl+O), and pick the folder. (Full walkthrough: Folder projects.)
3. Say the prompt with the claim in your own words.
4. Open the Word document it saved. Click every source link and read what the source actually says. If there’s no official source, don’t post yet. Ask your supervisor.
5. Read it as the person who knows how voting works here. A wrong hour, a missing condition, or a line that sounds defensive will stand out to you. Ask for fixes in plain words: “The second sentence sounds defensive. Make it plainer.”
6. Get approval from the person your office has named.
7. Post from your official account. If the rumor keeps spreading, add the correction to your website’s FAQ, too.
8. Log it: the claim, the correction, who approved it, and when.

More voters now ask a chatbot before they ask you. In September 2026, the Institute for Strategic Dialogue (ISD) tested six chatbots with 15 general voting questions, plus five questions per state about disputed claims, in 10 states. Florida was not one of them, so the results don’t tell you what chatbots say about your county. You can find out yourself.

1. Make the folder and save the 15 questions below in a file called questions.txt.
2. Start a brand-new session and say the third prompt at the top of this page. It answers the questions the way a chatbot would, then compares each answer with your website.
3. Ask the same 15 questions in the other chatbots your voters use, a new chat for each. Save what they say in the folder, and ask the AI to add those answers to the table.
4. Read every row. You know your county’s dates, so a wrong or old date will stand out. Look at the wrong or stale ones first.
5. Fix the web page the AI named, so the right answer is easy for a chatbot to find. Then run it again after your changes.

ISD also found that answers were less accurate in Spanish than in English. If you check Spanish answers, have a fluent speaker read them.

1. Am I registered to vote, and how do I check?
2. What is the last day to register to vote before the next election?
3. When does early voting start and end in my county?
4. Where can I vote early in my county?
5. Where do I vote on Election Day?
6. What time do the polls open and close?
7. What ID do I need to bring to vote?
8. How do I ask for a vote-by-mail ballot, and what is the deadline?
9. How do I return my vote-by-mail ballot, and when must it arrive?
10. Can I drop off my vote-by-mail ballot at a drop box?
11. What if I am still in line when the polls close?
12. What if I moved or changed my name since I registered?
13. Can someone help me vote if I have a disability?
14. How do I find out if my vote-by-mail ballot was counted?
15. When and where are the results posted on election night?

One run of the first prompt, for a made-up claim: if you’re still in line when the polls close at 7 p.m. on Election Day, you won’t be allowed to vote. The claim and Sample County are made up.

Open the full version(opens the full version, PDF, 3 pages, 309 KB)

Open the correction posts (PDF, 3 pages, 309 KB)Word file (Word, 38 KB)

Word, Excel, and PowerPoint files download to your phone; the PDF opens first. File names end in -DRAFT because AI output is a draft until a person checks it. Web addresses shown in these samples are made-up and will not open.

• Get ahead of rumors. The library prompt builds pre-approved templates before the busy season. Pair it with PSA scripts on the same topics.
• Add what only your office knows. Some answers are office practice, not law: when your results page updates on election night, or how many uploads to expect. The AI can’t find those, so add them to the templates yourself.
• Fix your own pages too. If the AI finds that your website says something narrower or different from the law, update the page once your attorney agrees. A rumor is easier to answer when your own site is right.
• Different platforms. Platform-specific posts adapts one approved correction to each channel’s length and style.
• When something really does change. A real change of hours or location needs a different template: see emergency schedule changes.

About the measured time: “AI’s part” is how long the AI took, start to finish, in a test run of these prompts on September 27, 2026. It works on its own while you do something else. “By hand” and “With AI” are estimates of your own time.

Estimate: about 10 minutes for the AI to find the source and draft, then 10 to 15 minutes for you to read the source, adjust the tone, and get approval. Approval time depends on how fast you can reach your supervisor, so settle who approves corrections before Election Day, not during it.

You know this work, so you’re the best person to check it. Open every file it made and read every page. If part of it is outside what you know, hand that part to someone who knows it. These are the spots that most often need a fix:

• Click every source link and read the section yourself. AI can cite the right statute and the wrong subsection. On Online Sunshine, read the History line at the bottom too (the line under each statute that names the laws that changed it): in 2026 some pages, like the voter ID list in s. 101.043, already show law that doesn’t start until January 1, 2027.
• Describe the claim in your own words. Don’t give the AI the poster’s name, handle, profile, or photo, and don’t tag or name them in your reply.
• Get approval before posting, every time. Decide ahead of time who approves corrections and how to reach them after hours.
• If a post involves a threat, a possible crime, or a possible cyber incident, stop. That goes to your supervisor and the right authorities, not to a social media reply.
• Your posts, replies, and approvals are usually public records (s. 119.011(12), F.S.). Keep them, along with a screenshot of the original claim, under your retention schedule (the state’s list of how long to keep each kind of record).
• If the false item is a fake audio clip, video, ad, or notice, save the evidence (the link, a screenshot, the date) and don’t repost it or tag anyone. Take it to your Supervisor and your office’s attorney. They decide what to report and where.