Insights · Elections ·

The polls closed on Tuesday. The fight over the count did not.

Nearly half of the false claims tracked around the US midterms were about election fraud, and the list kept growing into the count. What a team can do before, during, and after a count, including when the false claim comes from its own side.

The US midterm elections were held on November 3, 2026. Counting is still going on in many places, and so is a second contest: over whether the count can be trusted.

NewsGuard, a company that sells misinformation tracking and news reliability ratings, keeps a public tracker of false claims about the midterms. As of its November 5 update, it had debunked 55 false claims that went viral, and confirmed 611 appearances of them across 12 platforms, more than 392 social accounts, 71 websites, and 24 segments on national cable news. Twenty-six of the 55 were about supposed election fraud. On October 23, the same team counted 30 claims, 13 of them about fraud. In under two weeks, the list nearly doubled.

On Election Day itself, Snopes checked a claim that the FBI had found more than 20,000 fraudulent Democratic ballots in a raid on a Houston storage facility. That didn't happen. As results came in, Snopes also reminded readers that sudden jumps in the count are not, on their own, a sign of fraud. They come from state rules on when election workers may start processing ballots.

Why this matters for your team

Most of the attention goes to the weeks before a vote. This tracker shows that the pressure doesn't stop on Election Day. During the count, people are tired, results move, and every jump in the numbers can be framed as theft. A union, party, or NGO that campaigned for weeks often goes quiet the night results come in. That is the night its members need a clear, calm voice the most.

Two more findings from NewsGuard's October 26 briefing are worth knowing.

First, false claims don't come from one side only. In early October, NewsGuard found liberal users on Threads, Facebook, and X spreading the false claim that any ballot with written markings from a poll worker is automatically disqualified. A claim like that can make your own supporters doubt whether their vote counted.

Second, AI tools repeat what they find. A Russian influence operation known as Storm-1516 produced AI deepfake videos in which a supposed former student accused James Talarico, the Democratic US Senate candidate in Texas, of sexual abuse. The real former student told the fact-checker Lead Stories: "It's not me. The video is very much AI, very bad AI." Even so, when NewsGuard asked Google's AI Overview, Gemini, and Microsoft Copilot about the claim, all three repeated it.

Four habits for the next count

  1. Explain the count before it starts. Tell your members, in plain words, when results will come in, why some ballots are counted later, and why the numbers may jump. A supporter who knows that in advance is harder to scare on the night.
  2. Check before your side shares. A claim that flatters your side is the easiest one to share without checking. Ask staff and supporters to wait for an official source or a fact-check before they pass on anything about ballots, results, or an opponent, even when it would be good news for you.
  3. Answer once, on the record, and in writing. When an AI-generated image of Rob Sand, the Democratic candidate for governor of Iowa, circulated, his campaign told NewsGuard in a September 9 email that the image was AI-generated and "not a real image of Rob Sand." When the Talarico video spread, a campaign spokesperson called the claim "totally baseless and false." Short written statements like these give fact-checkers and reporters something to cite. Post the same statement on your own website, so people who search for the claim have a better chance of finding your answer.
  4. Don't repeat the claim to knock it down. Quoting a fraud rumor to mock it still shows it to people who hadn't seen it. Lead with what is true and how to check it, and link to the official election office.

Where Ostracon fits

Ostracon helps a team decide, in the first hour, whether a claim needs an answer at all. Its briefing recommends ignore, mirror, or push your own agenda, with ignore as the default on bait, and a person on your team makes the call. Ostracon doesn't judge what is true and doesn't detect fake accounts. When the answer is to act, it sends a mission to supporters who chose to receive them, such as "share the election office's results page and add one sentence in your own words," and every mission stays in a permanent archive.

Sources

Insights

Read next.

Organizing ·

What organizers can learn from Serbia's student vote

A movement with no leader and a list founded a month before the vote won a majority on the highest turnout in a decade. Five lessons for any organization that coordinates many groups at once.

Read

Platforms ·

The replies under your post are not the public

A new report on six US midterm races found that up to one in three comments on a candidate's own X account came from inauthentic accounts. If your team reads its replies as a mood check, it is reading the most distorted sample there is.

Read

Regulation ·

Europe's platform rules have teeth now. Your response still can't wait for them.

The EU has fined X and opened cases against Meta and TikTok under the Digital Services Act. Enforcement runs in months and years. A false claim about your organization runs in hours.

Read

Ostracon

Act within the hour. Or know when not to.