About

What On Air is

On Air follows political talking points through America’s podcasts: which shows say them, how many people hear them, and what claims come with them.

This is issue No. 1. We counted every English-language podcast episode that used one of six phrases about election fraud from July 27 to October 28, 2026: how fast it grew, which side’s shows said it, who heard it and what the people speaking did with it.

Every On Air issue →

Data and window

We searched English-language podcast transcripts for six exact phrases (“voter fraud”, “election fraud”, “rigged election”, “stolen election”, “illegal voting”, “noncitizen voting”), one week at a time from July 27 to October 28, 2026, and counted each episode once even when it used several phrases. Episode counts and lean ratings come from Particle. Two feeds that re-post old episodes under new dates were left out (369 episodes).

Show size is the show’s rank on Apple Podcasts charts on October 28: near the top means the top tenth of charting shows. Categories come from each show’s own feed and were matched for 82 percent of shows: 56 percent of the episodes came from shows whose first category on Apple Podcasts is News, 14 percent from Commentary, and about one in six from something else, including Comedy, Education, Business and Religion. Every show is listed, with its episodes, in Shows.

Lean

Lean ratings come from Particle. We looked up the lean of the shows with the most episodes, every show near the top of the charts and the 150 further shows with the largest estimated audience, and count the rest as unrated. Most small shows have no rating.

Audience size

Estimated audience weighs each episode by its show’s size, using follower counts from a podcast app, scaled to listeners at 3 listeners per follower (we also tried 1.5 and 6), with a floor of 30 listeners per episode. Shows without a count get an estimate from their chart rank. These are rough estimates for comparison, and an exposure is not someone hearing the phrase.

Every audience finding in this report was checked under four other weightings, and we report only what held under all of them:

• follower counts with big shows weighed somewhat less;
• the square root of the main estimate, which weighs big shows much less;
• chart rank only, with no follower counts;
• plain episode counts, every episode weighed the same.

A mention is not an endorsement

Shows that rebut fraud claims use the same phrases: NPR’s hourly newscast is in these counts, and so are hosts rebutting the claims. The stance sample, where a show has one, says which.

What a count can’t tell us

A count of phrases is not a count of false claims. The same words come up when a host repeats a claim, when a reporter describes one and when a guest knocks one down.

The transcribed sample below says who made a claim and how they treated it. It does not say whether the claim is true: Polygraph has not yet checked the sampled claims against the record, and those results will follow.

The transcribed sample

For “Who’s saying it” we drew a random sample of 250 segments per side from the episodes found (2,860 right-leaning and 1,077 left-leaning), at most 4 episodes per show, with the re-posting feeds left out. Four draws turned out to come from an archive feed that re-uploads old episodes; we dropped them, as the count does. We transcribed each segment from the publisher’s audio and identified speakers by voice.

Some draws were lost before transcription: 22 on the right and 5 on the left, mostly short-lived news-bulletin feeds such as NTD and Fox’s hourly news, whose episodes had left their feeds. A few more had no fraud phrase in the audio. That leaves 210 right-leaning and 227 left-leaning segments and 1,353 statements.

Each statement was read twice, by independent automated readers. They agreed on the speaker 95 percent of the time and on the stance 79 percent. Stance shares count every voice on air, hosts, guests and played clips alike. A stance is what the speaker said, not a verdict on the claim. There is effectively no far-left tier: one far-left segment was in the pool. Claims were grouped by meaning. “First in the sample” means the earliest episode in our window, which starts July 27, not where a claim began.

The quotes are from the transcribed sample, verbatim in our own transcript of each show’s published audio, with each clip cut from that audio.

How Polygraph checks and scores

The score is the share of a post’s claims that hold up, starting from the benefit of the doubt: every post begins with two claims’ worth of credit, a mixed claim counts half, and a false claim counts nothing. How it’s told can lower it further.

The full method is on the methodology page. It is a work in progress, and we will get things wrong. Here is why we get things wrong.

Polygraph has not checked the sampled claims yet, so no rating is shown in this issue. Checks will be listed on the Claims tab as they finish.

Credits

Source: Particle (episodes, lean ratings, audience estimates).

Show size: Apple Podcasts charts, October 28, 2026. Quotes and clips: our own transcripts of each show’s published audio.

Polygraph is built and published by Pluribus AI.