Summary
In the last 90 days, Trump posted 2,362 times on Truth Social, counting reposts: 15% more than in the 90 days before.
Each post got 36% fewer comments, 29% fewer reposts and 33% fewer likes on average.
Adding up all his posts: comments down 25%, reposts down 17% and likes down 22%.
52% of his posts sounded positive, about the same as before; 20% sounded negative, down from 23%.
The last 90 days (Jul 26 – Oct 23, 2026) compared with the 90 days before (Apr 27 – Jul 25, 2026). “Holds up” means the number passed every check we run before you quote it, like other time frames and the typical post. Tap a card to see them. Data as of Oct 24, 2026. How it works
Follow it over time
Is it up or down? The chart puts the latest period next to the one before it, day by day. Tap any day to read its posts.
Read the posts
The posts behind each period on the chart.
Get the data
Free to use and quote, no sign-up. The files open in Excel, Google Sheets and any analysis tool.
Every post, every field
All posts with full text, dates, tone and engagement. Refreshed every morning.
Just the dates you picked
One row per post for the dates on the chart.
Link to this exact view
The page address always holds the chart settings and dates, so anyone you share it with sees the same thing.
How it works
What each number means, where it comes from and what it can’t tell you.
- Up or down, compared with what
- Each number at the top compares the last 10, 30 or 90 days with the same number of days just before, and the chart below shows the same period. Periods are whole days in Chicago time and end with the last full day before the morning update, so a half-finished day never counts as a whole one.
- “Holds up” or “Check before quoting”
- Before a number goes on a card, the site asks the questions a careful reporter would. Does the change show up over other time frames (10, 30, 60, 90 and 180 days and a year)? Is it the typical post, not just a few viral ones? Do all his posts added together agree? Is it more than one busy day? For tone, do the negative share and the average score agree? If every answer is yes, the card says “Holds up”. If not, it says “Check before quoting”, and the card’s quote panel says what to watch for, in plain words. Anything that changes the meaning of a quote is written into the quote itself.
- Posts
- Everything he posted in the period: his own posts, quote posts and reposts of other accounts (ReTruths).
- Comments, reposts and likes
- How many times other users commented on, reposted or liked each of his posts, on average, for the posts he made in the period; the quote panels also give the totals for all his posts combined. Every count comes from the same source, the archive’s live feed (see Sources), so both periods are measured the same way. In copies of this feed downloaded hours apart, the counts of recent posts kept rising, so the newest posts may still be gathering reactions. Reposts of other accounts carry no counts of their own and are left out.
- Positive, neutral, negative
- Each post’s own words are read by an open-source language model trained on social media posts. It labels the text positive (praise, thanks, optimism), negative (criticism, hostility, alarm) or neutral. “Positive tone” is the share of posts labeled positive. It reads wording only: it can miss sarcasm, and it says nothing about whether a post is accurate or about politics.
- The tone score
- Behind each label is a score from −1 (clearly negative) to +1 (clearly positive): the model’s confidence the post is positive minus its confidence it is negative. The chart’s Tone view and the downloads use this score.
- Posts without text
- 15,038 posts (40%) are only photos, video or links, so there are no words to read. They are left out of tone figures rather than counted as neutral.
- Where the posts come from
- Every morning this site downloads a public feed of the account’s posts: a file served from a cnn.io web address, which a public GitHub archive of the account links to as “an archive, updated every five minutes.” Neither says who runs the feed or how it is collected; the posts and counts in it are Truth Social’s. The record here starts on February 14, 2022. A post deleted before the feed saw it would not be included, so treat this as a near-complete record rather than an official one. Every post on this site links to its original on Truth Social, and every source, including every website and account in his posts, is listed under Sources, with links.
- Quoting this site
- Use any figure freely, and please credit “Truth Social Tracker” and the date of the data. To keep a figure hard to dispute: name the period and its dates, say “per post” or “in total” for reactions, describe tone as rated by an automated model, and check the card’s verdict first. Each card has a ready-made quote that already does this, and the full data is free to download so anyone can check it.
Sources
Every outside source this site relies on, including the sources behind its sources. Each line says only what its link shows.
What each number comes from
- Posts, their words, dates and times
- Truth Social, via the live JSON feed this site downloads every morning (see below); for 240 posts from October 2026, the text saved directly from Truth Social (see Direct captures). Times are converted to US Central (Chicago) time.
- Comments, reposts and likes
- The reply, ReTruth and like counts recorded for each post in the same feed, which can lag behind what Truth Social shows. Reposts (ReTruths) have no counts here, because the feed’s numbers for them are not the original post’s. This site calls them comments, reposts and likes.
- Post types (his own post, quote post, repost)
- Worked out by this site from how each post’s text in the same feed begins: “RT @” marks a repost, and “RT:” followed by a Truth Social post link marks a quote post. Many repost or quote his own earlier posts.
- Websites and accounts in his posts
- Read straight from each post’s own text: every web address he linked and every Truth Social account he reposted, quoted or linked. All are listed below with their links.
- Positive, neutral and negative, and the tone score
- An open AI model published by Cardiff NLP, which this site runs on the words of each post that has enough text. Photo-, video- and link-only posts are marked neutral by this site’s rule and left out of tone figures unless a reader includes them. The tone score is this site’s calculation from the model’s output: the chance of positive minus the chance of negative.
- Every percentage, average, total and check
- Calculated by this site from the items above, as explained in How it works. The full data is free to download so anyone can redo the math.
- The photograph
- The Library of Congress (copy via Wikimedia Commons).
Where every post comes from
The chain from the original post to this page, one step at a time.
- Truth Social, @realDonaldTrump
The original posts: their words, times and the reply, ReTruth and like counts Truth Social shows on each one. Every post on this site links back to its original here. Photo and video links in the full download point to Truth Social’s own media server.
truthsocial.com
- Trump Media & Technology Group
The company that calls Truth Social “our social media platform” on its website.
tmtgcorp.com
- Live JSON feed (ix.cnn.io)
The file this site downloads every morning, served from a cnn.io web address. The GitHub archive below links to it as “an archive, updated every five minutes,” but does not say who runs it or how it is collected. In copies of this feed downloaded hours apart, the reaction counts of recent posts kept rising, so the site uses the feed’s latest counts for every post.
ix.cnn.io
- trump-truth-social-archive (GitHub)
A public GitHub repository, licensed CC0 1.0 Universal, whose script “fetches posts directly from the Truth Social API.” Its README says its workflow “was disabled on Oct. 26, 2025,” links to the live feed above, and says the project “is not affiliated with my employer.”
github.com
- ScrapeOps
The proxy service the archive’s script used to request posts from the Truth Social API, according to the archive’s README. The README does not say how the live feed is collected.
scrapeops.io
- Direct captures from Truth Social pages (240 posts, October 2026)
For 240 posts from October 19–24, 2026, the dataset’s first collection, made before this site’s daily update began, saved text, post types and the counts shown on Truth Social directly. The full download marks these rows “Direct browser snapshot” (Engagement count basis), “Direct API text” (Text source) and “API fields” (Classification basis). Their text is used; their saved counts are not: every figure uses the live feed’s counts.
truthsocialtracker.com
- GitHub Actions
Runs this site’s daily update: it downloads the feed, rates the tone of new posts and saves the results.
docs.github.com
Websites and accounts in his posts
His posts link to 657 websites (11,569 links in 11,459 posts) and repost, quote or link 317 other Truth Social accounts. Every one is a source and is listed here, read straight from the posts’ text. In 864 reposts, the original account’s name can’t be read reliably from the archive text; the download lists those too, each with its link on Truth Social, which shows the original.
Loading the list…
How the data is prepared
Tools the daily update runs on the downloaded posts before anything is counted.
- IANA Time Zone Database
The database of “the history of local time” worldwide, including daylight-saving rules, used to show every post’s time in US Central (Chicago) time.
iana.org
- ftfy (“fixes text for you”) by Robyn Speer
Repairs garbled characters (“mojibake”) in post text before the text is saved and its tone rated.
github.com
- langdetect
Checks posts of 30 or more words for text that may not be in English. A flagged post would not be rated and would be marked neutral for review. Its result is the “Language review code” column in the full download, which is empty for every post so far.
github.com
- Full data download
Everything this site counts, one row per post, with the original link, full text, counts, tone and how each value was assigned.
truthsocialtracker.com
How tone is rated
Tone is rated by software, not by people, the same way for every post that has enough words. Posts that are only a photo, video or link are not rated.
- cardiffnlp/twitter-roberta-base-sentiment-latest
The open AI model that labels each post’s words negative, neutral or positive. Its model card says it was “trained on ~124M tweets from January 2018 to December 2021, and finetuned for sentiment analysis with the TweetEval benchmark.” The card lists its license as CC BY 4.0.
huggingface.co
- cardiffnlp/twitter-roberta-base-2021-124m
The original Twitter-based RoBERTa model, which the sentiment model’s card links to (“The original Twitter-based RoBERTa model can be found here”). Its own card describes it as “a RoBERTa-base model trained on 123.86M tweets until the end of 2021.”
huggingface.co
- TimeLMs: Diachronic Language Models from Twitter (Loureiro et al., 2022)
The paper the model card lists as its reference paper.
arxiv.org
- TweetNLP (EMNLP 2022)
A second paper the model card cites.
aclanthology.org
- TweetEval (Barbieri et al., 2020)
The benchmark of “seven heterogeneous Twitter-specific classification tasks” that the model card says the model was fine-tuned with for sentiment.
arxiv.org
- Xenova/twitter-roberta-base-sentiment-latest
The same model with ONNX weights. The daily update runs this repository’s quantized ONNX file, offline.
huggingface.co
- VADER sentiment (Hutto and Gilbert, 2014)
A second, lexicon- and rule-based method. Its scores appear in the full download for comparison only; nothing on this page uses them.
github.com
Photograph and typeface
- The White House, Washington, D.C., by Carol M. Highsmith (Library of Congress)
The photograph at the top of the page, part of the Carol M. Highsmith Archive. The library’s record says “No known restrictions on publication.”
loc.gov
- Wikimedia Commons file page
Where the full-size copy of that photograph was downloaded. Commons lists its source as the Library of Congress.
commons.wikimedia.org
- Public Sans
The typeface, “developed by the United States Web Design System” according to Google Fonts, which serves it to this site under the SIL Open Font License.
fonts.google.com
Software the site is built with
These shape how the page works, not what the numbers say.
- Next.js
The web framework.
nextjs.org
- Apache ECharts
Draws the chart.
echarts.apache.org
- shadcn/ui
Buttons, menus and pop-up panels, using its Radix UI versions.
ui.shadcn.com
- Radix UI
The open-source building blocks underneath those components.
radix-ui.com
- Tailwind CSS
Page styling.
tailwindcss.com
- Motion
The counting and rising animations.
motion.dev
- Lucide
The icons.
lucide.dev
- ONNX Runtime
Runs the tone model during the daily update.
onnxruntime.ai
- Vercel
Hosts the site.
vercel.com