Nightshift · Data Forensics

Truth Social · 32,856 posts · 50 months
DATASET RANGE:
PRIMARY SOURCE: stilesdata.com/trump-truth-social-archive
Deep-dive analysis · Updated April 19, 2026

After hours on Truth Social — what the data actually shows.

A forensic look at 32,856 posts from Donald Trump's Truth Social account spanning 50 months — from February 2022 all the way through April 19, 2026. The record has been cross-checked against an independent 18,476-post scrape and 1.5M unique commenters, with the final 3,447 posts backfilled from trumpstruth.org (Nov 2025–Apr 2026). We analyze nighttime posting cadence (23:00–06:00 ET), topic distribution, engagement patterns, audience scale, and how all of them evolved over four years of political cycles. This is what the timestamps say — not the commentators.

Posts analyzed
Feb 2022 – Apr 19, 2026
Nighttime posts
of all posts
Total likes received
Avg /post
Busiest hour (ET)
Quietest:
01 · Overview

The 30,000-foot view

Volume, cadence, and the big-three topic split across 50 months. Trump averages posts per day but the distribution is far from uniform — daily volume, hourly rhythm and topical mix all shift visibly across 2022, 2024 and 2025.

Avg posts / day
across 50 months
Foreign policy
of posts
Domestic politics
of posts
Personal / lifestyle
of posts
Category distribution
Foreign · Domestic · Personal
Full dataset · n=32,856
Monthly volume
Posts per month · stacked by category
Daily volume — last 18 months
Total posts vs. nighttime posts (23:00–06:00 ET)
Weekday pattern
Total vs. night-only

⏱️ Posting peak

Activity concentrates between 11:00–21:00 ET — a 10-hour window accounting for the bulk of daily volume. The quietest hours are 08:00–09:00 ET, where traffic drops near zero.

🌙 Nighttime share

Roughly of all posts land in the nighttime window (23:00–06:00 ET). That's individual posts over four years — enough to draw behavioral patterns from.

📈 Monthly volatility

The busiest month was . Monthly volume fluctuates from ~350 to 650+ posts, closely tracking political news cycles: impeachment trials, election sprints, foreign-policy flashpoints.

01b · Recent 6 months

What the last half-year actually looked like

From November 1, 2025 to April 19, 2026 — 3,447 posts collected directly from trumpstruth.org after the stiles pipeline stopped updating. December 2025 broke every prior nighttime record in the dataset.

Monthly volume — Nov 2025 through Apr 2026
Total posts vs. nighttime posts (23:00–06:00 ET) · 3,447 posts

December record

December 2025 saw 250 nighttime posts out of 599 — a 41.7% night share, more than triple the lifetime average of 13%. Christmas Day 2025 alone produced 112 nighttime posts, eclipsing the Jan 24, 2024 record of 60.

January surge

January 2026 posted 762 total posts, the single highest month in the full 50-month record. 28.9% landed after 23:00 — the second-highest night share ever measured.

Topic shift

Recent 6 months are domestic-heavy: 85% of posts are domestic policy vs. the lifetime 66%. Foreign policy holds steady around 5–8%. Personal/lifestyle collapses to under 5%.

02 · Nighttime behavior

What changes after 23:00?

Do topics drift at night? Do personal/lifestyle posts dominate? The data reveals a small but consistent night skew — certain topic families systematically over-index in the 23:00–06:00 ET window, while others under-index.

Total night posts
23:00–06:00 ET
Night · foreign
of nights
Night · domestic
of nights
Night · personal
of nights
Hourly distribution — all 32,856 posts
24-hour clock (ET) · stacked by category · nighttime window highlighted
Night-vs-day topic skew
Percentage-point difference (night % − day %)
Topic Night % Day % Skew (pp)
Activity heatmap
Hour × weekday · intensity = post count
Biggest nights
Most posts in a single 23:00–06:00 window
Recent 2026 nighttime posts
74 posts · Mar–Apr 2026 · via trumpstruth.org

🎯 Strongest night skew

The category over-indexes the most at night (). Post-midnight is where amplification of supporter content and self-referential posts becomes relatively more common.

🔇 Strongest night under-index

Topics requiring business-hour context — — under-index at night (). Market-sensitive messaging clusters in daylight hours.

🌋 Record-breaking nights

The single busiest night logged posts (). Reference points from news: TIME reported 160+ posts in a single December 2025 evening.

03 · Topic drift

What Trump talks about — and how it shifts

Eight topic families tracked across 24 months, plus the long-tail "other." Topical focus is a leading indicator of the political calendar — foreign policy rises with geopolitical events, elections dominate pre-vote windows, and media/culture-wars content is a constant baseline.

Topic evolution — trailing 24 months
Stacked areas · monthly post counts by topic family
Topic distribution (lifetime)
Full dataset · all 50 months
Top words
Normalized content · stopwords removed
04 · Engagement

What actually lands

Likes, reblogs, replies — broken down by category, topic and time. Engagement is highly skewed: a tiny fraction of posts drive the bulk of interactions, and category has a measurable effect on median response.

Total likes received
across posts
Avg likes / post
median ≈ 1/3 of avg
Avg reblogs / post
heavy long-tail
Top single post
likes on one truth
Avg engagement by category
Likes & reblogs per post
Reference study — sentiment
Google Cloud NLP · 26,669 text posts · via Hubert Taler
Top-10 posts by likes (all time)
Highest favourites_count · stripped content
04b · Audience scale

The audience on the other end

Engagement is only half the picture. A second, independent archive — notmooodoo9 on HuggingFace — scraped both Trump's posts and the people who replied. It surfaces the audience dimension that the stiles archive does not capture.

Unique commenters
people who replied at least once
Total comments
across all tracked posts
Avg comments / post
audience response rate
Posts covered
through Oct 8, 2025
Multi-source cross-check — monthly post volume
stiles archive + fresh backfill (32,856 posts) vs. HuggingFace scrape (18,476 posts) · three independent pipelines

Audience mass

Over 1.5 million unique accounts commented on Trump's truths between Feb 2022 and Oct 2025 — a standing population an order of magnitude larger than active posters.

Comment density

31.8M comments across 18,476 covered posts works out to ~1,720 comments per post on average — the long-tail is huge, but the floor is already conversational.

Cross-validation

Monthly cadence from the HF scrape tracks stiles within ±10% on most months, independently validating the post timing used throughout this dashboard.

05 · Post explorer

Browse the data

Sample posts per topic family, with engagement metrics and direct links to the source. Filter, inspect, and use the links to open individual truths on the source platform.

Topic
Sample posts
06 · Methodology & sources

How this was built

This dashboard pulls from multiple primary and secondary sources. The main dataset is an auto-updating S3 archive of all posts from @realDonaldTrump. Classification is heuristic (keyword-based topic taxonomy) — not model-based — and deliberately transparent so readers can interrogate the labels.

Data pipeline
  1. Primary dataset — 29,409 posts fetched from stilesdata.com/trump-truth-social-archive. The pipeline's last successful run was October 26, 2025 — it has stopped updating since.
  2. Fresh backfill — additional 3,447 posts for Nov 1, 2025 – Apr 19, 2026 collected directly from trumpstruth.org. This archive does not expose likes/reblogs, so engagement for this period is omitted rather than guessed.
  3. Recent supplement — 74 nighttime posts (Mar–Apr 2026) collected from trumpstruth.org archive.
  4. Timezone normalization — UTC timestamps converted to ET with DST-aware offset (–4 or –5 hours).
  5. Topic classification — 8 keyword families + "other"; each post assigned to its top-scoring family.
  6. Broad categorization — topics collapsed into foreign / domestic / personal.
  7. Aggregation — monthly, weekly, hourly, daily rollups with engagement metrics preserved.
Caveats & limits
  • Heuristic classification — keyword-based; a post mentioning both "Iran" and "election" gets the higher-scoring topic. No LLM post-processing.
  • Retruth detection — posts with empty/media-only content are flagged as retruths; content-bearing retruths may not be.
  • Archive gap — primary S3 archive ends October 25, 2025. The 74 recent posts supplement but do not continuously cover Nov 2025 – Apr 2026.
  • Engagement reflects a single snapshot — likes/reblogs are frozen at scrape time.
  • Nighttime definition — 23:00 ≤ hour < 06:00 ET. Choice of window affects night-vs-day comparisons.
  • HF ID mismatch — the stiles archive and HuggingFace scrape use different internal post-ID schemas; only ~1,100 IDs overlap. Both independently cover the same account and comparable time window, which is why we treat them as cross-validation rather than merging.
HuggingFace audience layer

Dataset notmooodoo9/TrumpsTruthSocialPosts (CC BY 4.0) is a parallel archive totalling 5.89 GB: 18,476 Trump posts, 31.8M comments, and 1,500,898 unique commenter accounts. Timestamps are decoded from Mastodon/Snowflake IDs (upper 48 bits = ms since epoch). We use it for audience scale and as an independent check on post cadence — the content-level analysis still runs on the stiles archive, which has engagement counts and richer metadata.

Factbase — off-platform speech archive

Truth Social is one channel. Roll Call Factbase documents the parallel channel: speeches, press briefings, gaggles, remarks and interviews with video, full transcripts, speaker timestamps, hierarchical topic tags and FiscalNote StressLens sentiment scores — from January 20, 2025 onward. Read together, the two surfaces approximate Trump's full public message volume.

Sources