YOUPOL

An observatory of political YouTube and TikTok.

YouPol tracks more than 100 political channels and accounts in France, Quebec and the United States. Using machine-learning and artificial-intelligence methods, the observatory collects, archives and analyses the videos, what they say and the reactions they draw from the public.

Enter the observatory The method The methods paper (PDF)
Seven days of content on YouTube and TikTok · 29 September to 5 October 2026

Political YouTube and TikTok, the last week in brief

Every video published by the tracked channels is transcribed, summarised and counted by our local models. This panel gathers the synthesis of the last seven days, the records to read, the themes, the public figures named, the arguments and the removals; the three countries together or one of them, as you choose.

255
videos published
114
transcribed
4
records
42
hours of speech
66.5M
views
The synthesis of the week
United States · 29 September – 5 October 2026
The synthesis proposed by our local model · last 7 days

US Secret Programs, Donald Trump's Travels, and Claude AI Morality

On the PowerfulJRE channel, a former intelligence officer claims that the US government is hiding non-human craft and bodies to avoid financial fraud prosecutions. He argues that the "deep state" is blocking the disclosure of this evidence and establishes a correlation between nuclear weapons tests and the increase in unidentified anomalous phenomena activity.

Breaking Points reports that Donald Trump visited Oklahoma, a state described as his comfort zone. Additionally, the channel claims that US aerial superiority over Iran no longer exists following the loss of several aircraft. In another video, Breaking Points asserts that Benjamin Netanyahu's camp is using an attempted hijacking of a Dubai-Tel Aviv flight to present the Prime Minister as Israel's sole protector.

Regarding technology, Breaking Points reports that Anthropic consulted religious experts to implement morality and explore the consciousness of its AI, Claude. One speaker on the channel argues that this reliance on religion shows that the creators no longer understand how their technology works.

Generated by our model on 5 October 2026 at 04:06 from the 4 following videos:

The synthesis in full and the daily syntheses

The records to readAll the videos

The most viewed videos of the week, in each country, whose content synthesis is produced by our local model; each opens in a side panel.

United States255 videos
The themes, month by month Since 2016

Coming soon: the theme classifiers are trained on French, and their English counterparts will soon be trained and deployed. Meanwhile, this country’s channels are transcribed and summarised.

The public figures most named The ranking
September – October 2026 · transcribed sentences naming them
List
  1. Donald Trump 1,224
  2. Jeffrey Epstein 96
  3. Joe Biden 72
  4. Barack Obama 42
  5. Ben Shapiro 41
  6. Zohran Mamdani 40

Number of transcribed sentences naming them; bar: channel orientation; line: twelve months. Closed list of public figures. The full ranking

The arguments of the moment By theme
the first arguments of each theme, drawn from the sentences of the last seven days and read again every day

Content in preparation: it will appear here as soon as it is ready.

Gone from the platforms… All

No recent removal observed.

Latest transcripts1 h ago
Updated every hour; the last seven days, recomputed every day. The collection figures are in the counter, bottom left. Explore the observatory
The corpus

Three countries, two platforms, 11 channels

Each channel is chosen and coded by the team: an orientation, a country, a language. The far right is deliberately over-represented, to study it closely; the American channels, in English, are transcribed and summarised without going through the classifiers, which are trained on French.

Our method

Detect, extract, transcribe, classify, name, summarise

Six methods, run on the project’s computers, each documented with its models, thresholds and reliability. Pick a step.

01 · Detect

The channels’ RSS feeds are read every six hours; every week their full list of videos is reread. A detected video enters the queue with its date; those of the last seven days go ahead of everything else.

The detail in the method

Cadence
every 6 h, full listing every week
Priority
the last seven days first

The method in full · The methods paper (PDF) · Cite the project

The team

Three researchers

Tristan Boursier

Tristan Boursier

Postdoctoral researcher · co-lead

Sciences Po Paris · Université du Québec en Outaouais

Research design, orientation coding and political analysis.

WebsiteWrite
Antoine Lemor

Antoine Lemor

Postdoctoral researcher · co-lead

Université de Sherbrooke · CIRST · RFICS

Methods, machine learning, infrastructure and website.

WebsiteGitHubWrite
Arnaud Miranda

Arnaud Miranda

Research collaborator

CEVIPOF · Sciences Po Paris

Political theory and analysis of reactionary discourse.

WebsiteWrite

The team and the affiliations · Write to us · Cite the project

Project paper