Live feed
Every tracked leader's posts, newest first. Updates by itself every minute.
Daily games
Four quick games built from this week's real posts, starting with Guess Who. New ones every day at midnight UTC.
Which leader posted this?
This week's superlatives
Last 7 days. Personal accounts only; office accounts are left out of rankings.
Elections
Upcoming votes in tracked countries, with links to the latest news. A general-interest guide, not research data.
Diplomacy Desk
Posts by type, last 7 days
Busiest posting hours
Who tags whom
The Condemnation Web
Every condemnation by a tracked leader, drawn from the leader to whoever they condemned. Thicker lines mean more posts.
AI-made posts
Posts by tracked leaders that used AI-generated images, video or audio. A post appears here only when there is outside evidence, and every entry shows its source (Rule 9).
- Loading…
Research data
Everything behind Leader Watch, free to download and cite, with the rules we follow to build it.
Leader Watch grew out of an MA thesis on how diplomats shame each other when the world is watching. That study covered UN missions on Twitter from 2017 to 2022; this site extends the question to heads of state and government, in real time and across platforms.
How to cite
Cite the version (date) you downloaded, so others can reproduce your numbers.
Archived on Zenodo with a permanent DOI: 10.5281/zenodo.23299870. Each archived version is a dated snapshot with its own DOI, listed on the Zenodo page; the live downloads here are always the newest data.
Datasets
| File | What's in it |
|---|---|
| leaders.csv | Every country, its leader and title, region, the account tracked and its type (personal, office or none). |
| posts.csv | One row per post: post ID, platform, country, handle, account type, time (UTC and local), tag and how it was made, language, whether an AI translation exists, who a condemnation targets, engagement, virality, Community Notes counts and link. Engagement is a snapshot taken about 48 hours after posting, not the post's lifetime totals: see engagement_window and engagement_age_hours. |
| notes.csv | Community Notes written on tracked leaders' posts, from X's public Community Notes data: note ID, post ID, when it was written, its status and whether it is shown on X. |
| handcheck.csv | The hand check behind Rule 4: a fixed sample of 200 AI-tagged posts (40 of each type), coded by hand without seeing the AI tag, with the AI tag, the hand code and whether they agree. |
| labels.csv | How players of the former Condemnation or Not? game labelled each post, next to the AI tag. A check on the tagging; no longer updated. |
| ai_flags.csv | Posts flagged as using AI-generated media, one row per piece of evidence: the type (Community Note, the leader's own statement, or a fact-check), its source and link (Rule 9). |
X's developer terms limit sharing post text, so posts.csv lists post IDs and links. Researchers can retrieve full posts from X by ID. Truth Social posts come from the Trump's Truth archive. Telegram posts come from leaders' public channels. Data licensed CC BY 4.0.
Codebook
How each post is tagged
The rules
Numbered so you can cite them, e.g. "Leader Watch Rule 3".
Found an error? Email
Isabella Topp studies diplomacy in the age of the timeline: how social media and AI change what states say and who they say it to.
She holds an MA from the Committee on International Relations at the University of Chicago and a BA in International Relations from American University. Her research spans international relations theory, comparative politics and technological innovation, with a focus on how digital diplomatic audiences shape what states are willing to say in public. extends that work from UN diplomats to heads of state, in real time.
The Audience Structure of Shame: Twitter, the UPR, and the Logic of Diplomatic Condemnation
Naming and shaming is a three-way relationship among a shamer, a target and an audience, yet the audience is rarely varied. This thesis holds the speaker constant and varies the audience. It compares the 33,484 recommendations 102 states made in the third cycle of the UN Universal Periodic Review (2017–2022) with 50,328 country-directed tweets from 125 UN mission accounts over the same period, placing every communication on a common five-point severity scale. Shaming is both more severe and more frequent on Twitter: with each mission compared against its own conduct in the review, moving to the timeline raises the probability that a communication is a condemnation by 6.3 percentage points. The audience helps structure diplomatic condemnation.
How to cite the thesis
Topp, Isabella. 2026. “The Audience Structure of Shame: Twitter, the UPR, and the Logic of Diplomatic Condemnation.” MA thesis, University of Chicago, August 2026. https://doi.org/10.6082/cvt28-4an97
Admin
Private. Unlock with your admin key to hand-check tags and review AI-made posts.
Hand check of AI tags
Code each post yourself. The AI's tag is never shown here, so the check stays blind. Results are published as handcheck.csv (Rule 4).
AI-made posts
Community Notes saying a post is AI-generated appear on the site by themselves. Posts where a leader seems to say they used AI wait here for you to confirm. You can hide any flag.
Add from a fact-check
Only from a fact-checking organisation or a news report that says this post's media is AI-made. Older posts that Leader Watch never collected can be added too; you'll be asked for a few details.
Zenodo archive
On the 1st of every month the site publishes a new version of the dataset on Zenodo by itself. "Archive now" does the same immediately.
Loading…
Elections and polls
Add elections, pull polls from Wikipedia, and approve each poll before it appears on the site.
Open elections admin