The 2026 U.S. midterm elections are producing a new kind of newsroom experiment.
The Wall Street Journal has built an internal artificial-intelligence system designed to help its reporters understand what Americans are talking about online—by processing millions of TikTok videos and thousands of hours of podcasts and looking for emerging themes in political conversation.
The significance goes beyond the Journal.
For decades, journalists have tried to understand the electorate through polling, interviews, focus groups, campaign events and traditional reporting. Now a major newspaper is effectively building a machine capable of listening to a vast portion of America's digital political conversation at a scale no human newsroom could manage manually.
The question is no longer simply:
What are voters saying?
It is becoming:
Can AI listen to millions of voters simultaneously—and tell journalists what matters?
Meet ORCA: the newsroom's digital listening machine
The Wall Street Journal calls the system ORCA.
The system was initially developed to make the enormous podcast ecosystem searchable. According to the Online Journalism Awards' description of the project, ORCA automatically monitors podcast feeds, downloads new episodes and uses AI to transcribe them while identifying individual speakers. Google's Gemini then analyses transcripts, generates summaries and identifies topics.
Reporters can subsequently interact with the system through a chat interface.
Instead of listening to thousands of hours of podcasts themselves, journalists can ask questions and retrieve relevant material, including links back to the precise moments in episodes where particular statements were made.
That last feature is particularly important.
The AI isn't supposed to become the final authority.
It is supposed to help reporters find the evidence faster.
The scale changes everything
Consider the problem facing a political journalist.
There are thousands of podcasts.
Each produces hours of conversation.
There are millions of social-media posts.
New content appears continuously.
No newsroom, regardless of its resources, can have journalists manually listen to everything.
AI changes the economics of that process.
The Journal says its election tool has ingested millions of TikToks and thousands of hours of podcasts to help reporters track conversations influencing the electorate.
That means the newsroom can potentially observe patterns that would be almost impossible to identify through traditional reporting alone.
A phrase starts appearing.
A political issue suddenly begins dominating conversations.
A previously obscure personality becomes a frequent subject.
A narrative spreads between podcasts.
A particular community starts discussing an issue long before it appears in mainstream polling.
Those signals can then become leads for journalists.
This isn't a poll—and the distinction is crucial
The Wall Street Journal has explicitly cautioned readers not to interpret the system as a replacement for polling.
Editor in Chief Emma Tucker described the technology as “not a poll and not a replacement for reporting.” Journalists will choose the questions, verify the material and make editorial judgments.
That disclaimer is important.
A million TikTok videos do not equal a representative sample of American voters.
People who post on TikTok aren't necessarily representative of people who vote.
People who appear on podcasts aren't necessarily representative of the population listening to them.
Online conversations can also be distorted by:
- algorithms,
- coordinated campaigns,
- bots,
- influencers,
- partisan communities,
- viral controversies,
- engagement incentives.
A topic becoming extremely popular online doesn't automatically mean it is the country's most important political issue.
This is why the technology should be understood as a social listening system, not a magic electorate simulator.
But it can see things traditional polling may miss
That limitation doesn't make the technology useless.
It makes its purpose different.
Traditional polling asks a structured question to a sample of people.
AI-powered social listening can ask a different question:
What are people spontaneously talking about?
That distinction can matter.
Imagine that a polling organisation asks:
"Which of these five issues is most important to you?"
A voter chooses inflation.
But that same voter may spend two hours on TikTok watching videos about housing, healthcare or AI.
A poll captures the answer to the survey question.
A social-listening system captures parts of the conversation surrounding people's lives.
Neither is automatically superior.
They measure different things.
The podcast revolution is particularly important
American political discourse increasingly takes place outside traditional political media.
Podcasts have become enormous political ecosystems.
Politicians appear on them.
Activists appear on them.
Influencers appear on them.
Commentators debate on them.
Communities develop their own political vocabulary.
And unlike a television interview that might last 10 minutes, podcasts can run for two, three or even four hours.
That creates a huge amount of political material.
ORCA was partly created because journalists could no longer reasonably treat this enormous ecosystem as something they could monitor manually. The system has already been used to analyse more than 22,000 episodes from 150 shows while tracking political narratives.
That is a fundamentally different approach to journalism.
Instead of starting with:
“What did the major newspapers say today?”
the journalist can increasingly ask:
“What are hundreds of political communities talking about today?”
AI can also track how narratives move
This may be one of the most interesting applications.
A political narrative rarely stays in one place.
It can begin with:
a podcast
then move to:
X
then:
TikTok
then:
YouTube
then:
television
then:
campaign speeches
then:
mainstream news coverage.
ORCA's clustering and search capabilities are designed to help reporters identify these movements. The system can group similar episodes and help journalists investigate where particular ideas emerged and how they spread.
That gives journalists something closer to a map of information flow.
And in modern politics, information flow can be almost as important as polling numbers.
The 2026 election is happening in an unusually digital environment
The experiment comes as AI itself becomes a political issue.
The Washington Post reported in August that AI had entered hundreds of 2026 congressional and gubernatorial races, with candidates addressing issues including automation, jobs, data centres and national security. Its analysis found AI-related policy positions appearing in nearly 40% of races nationally.
Reuters has also reported that AI concerns are increasingly entering political debate, although polling indicates that voters continue to place issues such as the economy, immigration and international conflicts ahead of AI itself.
At the same time, political campaigns are using AI themselves.
That creates an unusual situation.
The technology isn't simply reporting on the election.
It is increasingly becoming part of the election.
The campaigns are using similar technology
Axios reported earlier this year that Republican campaign strategists were using AI to simulate voter attitudes and scan large amounts of social-media content for changes in public opinion.
That means campaigns and news organisations are beginning to use similar technological capabilities for different purposes.
A campaign might ask:
"How can we persuade this group?"
A newsroom might ask:
"What is this group talking about?"
Those questions are fundamentally different.
But they may rely on some of the same underlying technology.
That creates an increasingly important distinction between understanding public opinion and engineering public opinion.
Journalism is becoming computational
The traditional investigative journalist might spend weeks:
- calling sources,
- requesting documents,
- reading transcripts,
- attending meetings,
- interviewing people,
- travelling to communities.
AI doesn't eliminate those activities.
But it can change what happens before them.
A reporter could ask the system:
“Find emerging discussions about healthcare costs among politically conservative podcasts.”
The system could identify several hundred relevant conversations.
The journalist can then investigate.
Another reporter could ask:
“When did this particular political narrative begin appearing across these communities?”
The system could identify dates and sources.
A third could ask:
“Which politicians are repeatedly mentioned alongside this issue?”
Again, the AI can narrow the search.
The journalist then performs the most important part:
verification.
The danger is confusing digital volume with public opinion
This may be the biggest weakness of AI political listening.
The internet isn't a democracy in miniature.
It is an ecosystem governed by algorithms.
The loudest people aren't necessarily the most numerous.
The most viral issue isn't necessarily the most important.
The most repeated statement isn't necessarily true.
A coordinated group can create the appearance of a spontaneous movement.
An influencer can generate millions of views without having millions of politically engaged followers.
A single controversy can dominate digital conversation for 48 hours and then disappear.
AI can process these signals incredibly efficiently.
But processing more data does not automatically produce a more accurate understanding of society.
That is why human reporting remains essential.
There is also a representation problem
Suppose ORCA finds that 80% of the podcasts discussing an issue are strongly conservative.
Does that mean 80% of Americans support the position?
Absolutely not.
It could mean:
- conservative podcasts talk about the issue more,
- the issue is especially interesting to that audience,
- the algorithm surfaced those podcasts more effectively,
- liberal audiences discuss it elsewhere,
- many voters aren't discussing it online at all.
This is why digital conversation should be treated as evidence about discourse, not automatically as evidence about the entire electorate.
The distinction is subtle but fundamental.
AI could actually make journalism more empirical
There is another side to the story.
Used properly, systems like ORCA could make journalism more evidence-driven.
Instead of a reporter saying:
"People online seem angry about this."
The journalist can investigate:
Which people?
Where?
When did the conversation begin?
How quickly did it spread?
Which communities are participating?
Who introduced the claim?
What evidence supports it?
Did the conversation move into mainstream politics?
Those are much better questions.
AI doesn't necessarily replace journalism.
It can make better questions possible.
The WSJ experiment also exposes the changing role of journalists
The journalist of the future may spend less time searching for information and more time evaluating information discovered by machines.
That changes the skill set.
Reporters increasingly need to understand:
- datasets,
- algorithms,
- model limitations,
- sampling bias,
- transcription errors,
- statistical inference,
- social-media dynamics,
- verification.
The journalist isn't simply writing anymore.
The journalist is increasingly becoming an investigator working alongside computational systems.
That could be one of the most important changes in journalism this decade.
But who watches the AI?
There is an unavoidable paradox.
The Journal is using AI to understand the political conversation.
But the AI itself is making decisions about what to surface.
That means journalists need to know:
What did the system miss?
What did it misunderstand?
Which sources did it overweight?
Which communities aren't represented?
Could its clustering create artificial relationships between unrelated conversations?
Could transcription errors change meaning?
Could the model interpret sarcasm as sincerity?
Could a viral campaign overwhelm genuine grassroots discussion?
These aren't theoretical questions.
Large-scale automated analysis inevitably involves filtering.
And whenever something filters information, it shapes what humans see.
The election newsroom may therefore become a new kind of information infrastructure
This is the deeper significance of the WSJ project.
A major newspaper is no longer simply receiving information from reporters and sources.
It is building a system capable of continuously monitoring a huge portion of the digital information environment.
That resembles something closer to a political intelligence platform.
Not because the newspaper is secretly manipulating voters—the evidence does not support that conclusion—but because the underlying capability is powerful.
A system capable of detecting political narratives at scale could potentially be useful to:
- journalists,
- researchers,
- campaigns,
- governments,
- political parties,
- advertisers,
- advocacy groups,
- corporations.
The same technology can answer very different questions depending on who controls it.
The coming election may therefore produce two simultaneous experiments
One experiment is happening among voters:
How much will digital platforms influence political behaviour?
The other is happening inside institutions:
How effectively can AI understand what voters are doing online?
The Wall Street Journal is now part of the second experiment.
Its stated goal is not to predict the election with an algorithm.
It is to help journalists understand the conversations surrounding it.
That distinction is important.
The system is being positioned as another instrument in the newsroom—alongside reporters, polling, interviews, documents and traditional reporting.
From polling the public to listening to the internet
For most of modern political history, understanding voters required sampling.
Now there is another source of information:
the enormous stream of conversations people voluntarily publish online.
It is messier than polling.
It is noisier.
It is less representative.
But it is also continuous.
And AI finally makes it possible to process that stream at a scale humans simply cannot.
That is the real breakthrough.
Not that an AI can tell journalists who will win.
It can't reliably do that.
Not that millions of TikToks represent millions of voters.
They don't.
But AI can help journalists see patterns in a digital conversation that would otherwise be almost impossible for humans to observe at scale.
And that could fundamentally change political journalism.
The new political sensor
The Wall Street Journal's experiment points toward a future in which newsrooms operate something resembling a digital political sensor.
It listens continuously.
It searches broadly.
It identifies emerging themes.
It detects changes.
It gives reporters leads.
Then humans investigate.
That last step is crucial.
Because the most important question isn't:
“What does the AI think voters believe?”
It is:
“What evidence did the AI surface—and does that evidence actually tell us something meaningful about the country?”
That is the line between AI-assisted journalism and AI-generated political mythology.
The 2026 midterms will provide a real-world test.
Not only of the candidates and political parties, but of something quieter happening inside the media itself:
Can machines help journalists understand a country that is increasingly expressing its political identity through millions of fragmented digital conversations?
The answer may determine what political journalism looks like long after the 2026 election is over.
Comments
Post a Comment