The Newsletter Zeitgeist

US POLITICAL NEWSLETTER ANALYSIS BY AI  ·  DESIGNED BY MIKE FOURCHER

Who saw the blue wave coming?

Produced by Mike Fourcher - mike@fourcher.net
Word cloudsDid Pundits Notice 2026 Election Turning Points?

To test if political pundits could see if a blue wave was coming - and that it would be as sizable as the +14 popular vote margin it was - I ran an analysis of language used by 72 political newsletters and podcast transcripts, gathered by me between September 1 and Election Day, November 3, 2026. As detailed below, I looked for certain phrases to identify “wave talk”.

Conclusion: For the most part, Democrats were slightly more optimistic about a broad Democratic victory over time. But around October 20, anticipation of a wave built among all groups built. This coincides with the October 19 launch of Vivek Ramaswamy’s “mother” ad (see “Did Pundits Notice 2026 Election Turning Points”). But then, on October 27, Nate Silver posted a story suggesting the final NYT/Siena poll might be right, predicting a +11 Democratic wave. Even that size of a win was not fully accepted by all pundits - and nobody saw the magnitude of a +14 Democratic victory as not even in the realm of possibility.

What was measured. Every newsletter and podcast transcript was read sentence by sentence with rules fixed in advance. A sentence counts only if it is predictive (will, expect, likely, poised, polls show and similar). Two kinds of Democratic outlook were counted: wave language (blue wave, tsunami, landslide, blowout, sweep, rout, in a US election context, not negated, not a hypothetical) and broad-victory language (Democrats will flip the House, win the Senate, gain seats, are favored or poised). Hopeful but non-predictive sentences are left out. The actual result was a +14 Democratic popular-vote margin. Counts are sentences per 10,000 words, so long shows and short columns compare fairly.

When the language arrived

LeftCenterRight

Weeks start on the date shown. The final point is election day alone (Nov 3), so it is a single day and includes live-night talk; treat its height with care.

What the data says

Groups before and after

Each number is the count of sentences per 10,000 words that use that kind of language (before the vote is Sept 1 to Nov 3, after is Nov 4 to 14), so a higher number means the group used it more often and a lower number means it rarely did.

The size of the win versus the numbers quoted

Quoted generic-ballot and popular-vote figures in predictive sentences, with the actual result for scale.

DateSourceFigureContext

Most figures are polls quoted by the source, not the source's own forecast. Silver Bulletin's own House popular-vote forecast was D+7.9 on Sept 29 and again on Nov 2. The highest widely quoted polls (NYT/Siena, Emerson) were 11 to 12. No source quoted a figure near +14.

After the vote

Earliest wave sentences from the leaders

Source by source

All sources with text before Nov 4 (the rest of the roster had no documents in the window). We were unable to retrieve articles from Just the News from Oct 18 to Oct 31, so it appears here but is left out of the group totals and the timeline. Sources under 60,000 words are marked thin, and their rates are noisy. "Expected" is how many wave sentences the source would have if it matched the pooled rate of all sources. The q value is a Benjamini-Hochberg false discovery rate: when many sources are tested at once, a few will look unusual by luck alone, and this correction estimates what share of the sources flagged as unusual would be flukes. A q below 0.10 means roughly one in ten or fewer of them would be a chance result. Click a heading to sort.

SourceGroupWords (k)Wave sentencesExpectedPer 10kDem-win sentencesGOP-holdDoubting a waveq

Read with care

Wave talk is not a forecast of the size. Many wave sentences are conditional or hedged ("maybe the blue wave is not as big as the polls suggest"). The counts show who was discussing a wave most, which is a proxy for expecting one.

Quoting is not predicting. Dan Bongino's count is the clearest case. His samples quote Democratic wave talk in order to dismiss it, so a high count there is not a forecast. The same may hold, to a lesser degree, for other sources. The first extraction mixed in spending "blowouts" and foreign landslides, and those were removed in the rerun.

Hope versus forecast. Left-leaning sources may have wanted a wave. Only predictive sentences are counted, but a source can phrase a hope as a prediction.

Coverage. The Dispatch Podcast transcripts stop on Nov 6 and TYT was not collected. Commentary Magazine could not be fetched in full, so it has no text. The Morning Dispatch has 44 emails for about 54 weekdays. We were unable to retrieve articles from Just the News from Oct 18 to Oct 31. A few columnists (Bouie, French, Bruni, Wallace-Wells) have small samples because only some columns were recovered.

Rule-based reading. The sentence rules are regular expressions, so some sentences are misread in both directions. Sample sentences are shown so any row can be checked.