Fact Brief · Summer 2026
Social Media and Information Quality
Like more than half of American adults, you may get news from social media — and from others who use social media. You may also think the platforms are flooded with lies. But what are the facts about how much, and by whom?
No verified figure exists for how much of social media is false. Research is thin and slow: the best measurements of ordinary users still come from the 2020 election, while the tactics changed in 2025 and 2026. Every fact here carries the year it was measured. Where nobody has measured yet, this brief says so.
Know the Terms
| Disinformation — false information deliberately designed to mislead, deceive or manipulate. Made-up news, manipulated video, deepfakes.[6] | Misinformation — partial or inaccurate information spread without the intent to deceive.[6] |
| Malinformation — true information used out of context to harm. A real photo captioned with the wrong war; a real quote with the next sentence cut off.[6] | Propaganda — information structured to achieve an agenda or promote an ideology. |
| Gaslighting — psychological manipulation meant to replace what we know, experience or see with our own eyes with another version of events. | Information warfare — a state or group attacking another country's information space to shift opinion or break trust. |
| Algorithmic micro-targeting — using big data and computing power to aim tailored, manipulative messages at specific groups. Common in politics and advertising. | AI-powered disinformation swarms — thousands of AI-generated accounts, websites and bots spreading coordinated lies about individuals, groups and more.[17] |
| Deepfake vs. cheapfake — a deepfake is made by AI. A cheapfake is real footage made misleading by editing, slowing or a false caption. | Supersharer — the very small number of human accounts behind most false material.[1] |
Where Americans Get News
Where U.S. adults regularly get news
| Facebook 38% | TikTok 20% | Nextdoor 6% | Rumble 2% |
| YouTube 35% | X 12% | WhatsApp 5% | Truth Social 2% |
| Instagram 20% | Reddit 9% | Threads 3% | Bluesky 2% |
Television is a different question on a different scale, so it does not belong in that ranking: 64% of adults get news from TV at least sometimes, 32% often, and 65% get news from a local TV station at least sometimes. [21]
Your age predicts where — and what you trust
76% of adults under 30 get news from social media at least sometimes, compared to 28% of those 65 and older. On trust: under-30s trust social media (50%) and national news organizations (51%) about equally. For adults 65 and older it is 24% for social media against 62% for national news organizations. [22]
Human Superspreaders
Three in every thousand — Twitter, 2020
2,107 people — 0.3% of 664,391 registered voters studied — shared 80% of the fake news on Twitter in 2020. Researchers matched accounts to voter registration records and counted what each one posted. [1]
About one percent — Facebook, 2020
About 1% of Facebook users produced most misinformation re-shares. It moved friend to friend, one step at a time, reaching about as many people as Pages that broadcast to every follower at once. From over 1 billion posts by 110 million users. [2]
Not bots — and who they were
They produced that volume by hand: persistent manual retweeting, all day. Twitter supersharers were disproportionately women, older adults and registered Republicans. [1] Facebook's top re-sharers also skewed older and more conservative. [2] And 97% of 2020 political news pages rated false by fact-checkers were seen by more conservative than liberal users. [3]
Machine Superspreaders
After 2024 the operators changed. This is where the field is moving — and where the numbers you would want do not yet exist.
An untraceable fake press, built in a year
One Russian network, CopyCop, tied to Russia's military intelligence service, launched more than 300 fake news sites during 2025 — aimed at the United States, France, Canada, Germany, Moldova and Armenia. It writes the articles with Meta's Llama 3 software running on its own servers, so nothing is logged with any AI company. An earlier version ran more than 1,000 invented journalist personas. [18]
Swarms are the next step
In January 2026 researchers writing in Science warned of malicious AI swarms: AI-run accounts with lasting identities and memory. They coordinate while varying their tone, adapting in real time to your replies, and operating across platforms with little human supervision. The danger is less false content than manufactured consensus — the feeling that everyone agrees. [17]
Old tricks, faster
OpenAI has shut down more than 40 networks misusing its tools since February 2024, including covert influence operations. What it found: these operators are not inventing new tactics. They are running the same old propaganda playbooks, and AI simply lets them run them faster and cheaper. [19]
Nobody is counting
No published figure exists for what share of accounts or posts on any major platform is AI-run. Nor can we find reliable data on what share is a deceptive AI depiction of a real person. The 2026 International AI Safety Report says such content is already used for scams, fraud, blackmail and non-consensual imagery, but that data on how much remains limited. [17] [5]
Think You Can't Be Played by AI? Think Again!
About as well as a coin toss
1,276 people sorted real from AI-generated media in a mock social feed. They got an average of 51.2% right — not much better than a coin toss (50%). On images, the average was 49.4%. On AI-generated faces, 46.6% — worse than guessing. Across every kind of media in the study: shown something genuine, they called it correctly 64.6% of the time; shown something AI-generated, only 38.8%. [4]
Experience did not help
People who called themselves highly familiar with deepfakes scored 51.9%. Those who called themselves unfamiliar scored 51.1%. Not a significant difference. [4]
Platforms Unfettered
The fact-checking and research tools are switched off
Meta ended U.S. third-party fact-checking on January 7, 2025, moving to Community Notes — where a note appears only if contributors who normally disagree both rate it helpful. [7] YouTube dropped its 2020 election-fraud policy in 2023, and in September 2025 told Congress that creators banned under policies no longer in effect could return. [15] Meta shut down CrowdTangle on August 14, 2024, for a decade the main tool for tracking how content spreads. [8] X charges for research access: its API is billed per request, with no free tier. TikTok's research tool is limited to academic and non-profit institutions. Reddit requires a university affiliation and ethics-board approval. [14]
On X, the algorithm moved opinion
U.S. users were randomly assigned an algorithmic or a newest-first feed for seven weeks. Switching to the algorithmic feed raised engagement and shifted political opinion toward more conservative positions. Switching the other way had no comparable effect. [16]
Your worst posts get the most attention
Over 10 million posts carrying news links, seven platforms, January 2024 — X, Bluesky, Truth Social, Gab, GETTR, Mastodon and LinkedIn. On every one, the same user's lower-quality news posts drew more engagement than their higher-quality ones: left-leaning platforms and right-leaning ones, and platforms with no ranking algorithm at all. The study also found platforms with more conservative user bases carried lower-quality news on average, and that each platform's own politics won the most engagement there. [23]
It's a bi-partisan issue
Sens. Bill Cassidy (R-LA) and Chris Coons (D-DE) condemned the CrowdTangle shutdown the day it took effect. On January 7, 2026 they reintroduced the Platform Accountability and Transparency Act (S. 3292), which would force large platforms to open data to vetted researchers. [9]
How Much Is AI-Made
On the platforms
AI-generated posts on Medium rose from 10.2% in December 2022 to 37.0% by October 2024. On Reddit over the same period: 1.9% to 3.0%. [11] One example: between April and June 2026, about one in four posts over 250 words across LinkedIn, Medium, Substack, X and Reddit were classified AI-generated — 41% on LinkedIn, 2% of Reddit replies. From 1,002,627 posts collected through a browser tool users chose to install, so not a random sample. [12]
Synthetic is not the same as false
Much AI-made political content is openly a meme, a joke, an illustration, or a person working with an AI tool. A count of AI-made posts is not a count of posts meant to deceive.
In the false stuff
European fact-checkers count all the disinformation they catch, monthly. The share created or altered with AI: 16% in December 2025, 20% in March and April 2026, 14% in May 2026. [13]
What Actually Helps People Detect More Falsities
Ninety seconds of training works — for about a month
Short videos teaching five manipulation techniques made viewers 1.5 to 1.67 times better than a control group at spotting manipulative content. Seven pre-registered studies, nearly 30,000 participants. [10] A separate set of five pre-registered experiments with 11,759 participants tested how long it lasts: video and text training stayed effective for about a month, game-based training faded faster, and a short refresher extended it. [20] Watch free at inoculation.science.
When you see a post, ask…
- Who is the original source, and can I reach it in one click?
- Real footage with a false caption? Far more common than a deepfake.
- Is it trying to make me feel something before it tells me anything?
- Do a lot of strangers suddenly agree? Manufactured consensus is the point of a swarm.
- Is this a measurement or a forecast? A raw count or a share of a population?
- Would I share it if it made the other side look good?
What Do the Facts Say?
Research that tracks misinformation is sparse.
The sophistication of threats from AI and large-scale manipulation is growing.
Low-quality posts draw more engagement than high-quality ones — even on platforms with no ranking algorithm.
Adults under 30 are far more likely to get news from social media than adults over 65.
Among ordinary users, false content comes from very few accounts.
The platforms are not protecting you from manipulation. It is up to YOU to be a fact sleuth.
Shown AI-made content, people took it for real about six times in ten.
European fact-checkers publish a monthly brief counting the disinformation they detected — which topics it targeted, and how much was made with AI. A new one appears each month; the latest covers May 2026. No comparable monthly count is published in the United States. See the EDMO Fact-Checking Briefs.
All links verified August 6, 2026. Source 11 is a preprint, not peer-reviewed. Source 12 is vendor-published from an opt-in sample. Source 17 is a policy forum — expert warning, not measurement. Source 18 is commercial threat intelligence with proprietary methods.
Every Fact Brief follows the same discipline — so you can trust what's on these pages, and check it yourself.
Prepared for educational purposes from publicly available sources believed reliable at the time of publication. Readers are encouraged to consult the original sources. A www.ClaimYourVoterPower.org Fact Brief. Free to print, copy, host, and distribute unaltered, with attribution (CC BY-ND 4.0). Quotation with citation welcome. Published August 2026.