Free AI Media Bias Analysis Tool

Does this article — or video — contain bias or manipulation?

Paste a news article, drop a URL, or drop a video link — YouTube, TikTok, Instagram, Facebook, or X. Our AI flags manipulation techniques sentence by sentence, and for stories with a real backstory — geopolitics, economic and health policy, civil rights and more — surfaces the historical context coverage tends to leave out.

Try it free →

Free during beta — no account required.
Nothing to paste? See today's news, analyzed →

✓ 18 manipulation techniques ✓ Historical context for major news topics ✓ Sentence-by-sentence analysis ✓ Articles, text & video ✓ Claim checks with cited sources
New: claim checks, and the 2026 political ads index. Semblen now checks an article's own factual claims against cited sources — and analyzes the campaign ads running in this year's Senate races. See what's new →
Why this exists

In 1987, fairness stopped being anyone's job

On August 4, 1987, the FCC voted 4–0 to abolish the Fairness Doctrine — the rule that since 1949 had required radio and television broadcasters to cover controversial issues and to air opposing views. Congress passed a bill to keep it; President Reagan vetoed it. Nothing replaced it, and cable news and the internet were never covered at all.

72%

1976

of Americans trusted the mass media to report the news fully, accurately and fairly.

28%

2025

say the same today — the lowest figure Gallup has recorded in fifty years of asking.

→

So the job is yours

News is a business that sells attention, and nothing obliges it to be balanced. Semblen shows you the techniques, the missing history, who gets quoted and whether the checkable claims hold up — and leaves the conclusion to you.

The repeal is one cause among several — media consolidation, the 24-hour cable cycle and the collapse of advertising-funded local news all did their part. We start there because it is the moment balance stopped being an obligation and became a choice.

What's new

Claim checks, campaign ads, and whose words they are

✓

Claim checks with cited sources

Semblen pulls out the checkable claims an article makes in its own voice and looks for primary sources, fact-checkers and wire reporting. Each verdict links to the passage it rests on. Claims the article only reports someone making are listed, not checked — and when the evidence is thin it says “No authoritative evidence found” rather than guess.

★

Political ads, 2026

The campaign video ads running in this year's Senate races, transcribed and run through the same analysis as the news: techniques flagged line by line, checkable claims checked. Browse by race or look up your ZIP →

“”

Quoted language, kept separate

Loaded language inside a quotation belongs to the speaker, not the outlet. Semblen now marks it as quoted language instead of counting it against the article — so you can see what was said, who said it, and that the outlet chose to print it.

How it works

How Semblen works

Semblen analyzes news articles, videos, and text for media bias and manipulation techniques — loaded language, false equivalence, historical revisionism, geopolitical framing, and more. It shows how language is being used on you — and, separately, checks the article's own factual claims against cited sources.

1

Article, text, or video

Paste a URL, drop raw text, or paste a video link — YouTube, TikTok, Instagram, Facebook, or X. For articles we follow pagination automatically and analyze up to 500 paragraphs. For videos we analyze the transcript in 1-minute segments.

2

Two-stage AI analysis

A fine-tuned transformer classifies each sentence as manipulative or neutral. For flagged sentences, Claude AI identifies the specific technique and explains why it's manipulative — no generic labels, just precise analysis.

3

See exactly what's flagged — plus who's being quoted

Flagged sentences are highlighted in red; charged language that sits inside a quotation is marked in orange and kept out of the count, because it is the speaker's, not the outlet's. Hover or tap any paragraph to see which technique was used and a plain-English explanation. A separate source distribution panel shows who's being quoted in the article — by role and national affiliation — and flags imbalances.

Historical context

See the history this article leaves out

When an article has a real backstory — a geopolitical conflict, a financial crisis, a decades-long policy fight, a public-health debate, a civil-rights movement — Semblen surfaces the historical background most coverage skips. Two tabs side by side: a neutral encyclopedic overview, and a "what this article doesn't tell you" view that pinpoints what's missing relative to what the article addresses. Routine business, sports, and celebrity stories return nothing — no scope creep.

Geopolitical conflicts Financial crises & regulation Healthcare & immigration policy Vaccine & climate policy Civil-rights & social movements
Background: US–Iran relations
Neutral overview What this article doesn't tell you

US–Iran relations have been shaped by the 1953 CIA- and MI6-backed coup against Prime Minister Mohammad Mossadegh after his nationalization of Iran's oil industry, the 1979 Islamic Revolution and ensuing hostage crisis, the Iran–Iraq War of the 1980s (during which the US backed Iraq), the 2015 Joint Comprehensive Plan of Action nuclear deal, and the 2018 US withdrawal followed by escalating sanctions.

Wikipedia: 1953 Iranian coup ↗ Wikipedia: Iran hostage crisis ↗ Wikipedia: JCPOA ↗
Wikipedia-cited. Every factual claim links to a Wikipedia article — and the URL is HEAD-validated before it reaches you. No hallucinated links.
Delta mode. The "omitted" tab compares what the article addresses against what's commonly skipped, so you see the specific gap — not a generic history dump.
Streams in under 5 seconds. Topic and sources appear right away; prose fills in progressively. Cached for 30 days — the same article re-opens instantly.
What we detect

18 manipulation techniques, automatically identified

Manipulation rarely announces itself. These are the patterns our AI is trained to catch — and explain.

Loaded Language

Emotionally charged words that trigger fear, anger, or contempt rather than informing.

"The radical extremists are destroying our way of life."

Name Calling / Ad Hominem

Attacking a person or group with derogatory labels to discredit them without engaging their argument.

"Only a fool or a traitor would support this policy."

Appeal to Fear

Exaggerating threats to override rational analysis and demand immediate action.

"Our children will grow up in a totalitarian nightmare if we allow this."

Bandwagon Appeal

Pressuring agreement by claiming everyone is already on board, suppressing independent thought.

"Millions of patriots have already joined the movement — are you with us?"

False Urgency

Manufactured time pressure that discourages careful analysis and demands immediate compliance.

"We must act now — every hour we wait, the situation becomes more catastrophic."

Black-and-White Fallacy

Presenting only two extreme options when a range of alternatives actually exists.

"You're either with us or you're helping the enemy."

Historical Revisionism

Rewriting, minimizing, or denying documented historical events — colonial atrocities, genocides, war crimes — to serve a current narrative.

"The so-called genocide is just enemy propaganda — colonial rule brought stability to underdeveloped peoples."

War Crime Euphemisms

Sanitized language that obscures documented violations of international law.

"The operation resulted in unavoidable collateral damage."

False Equivalence / Whataboutism

Creating artificial moral balance between actions that aren't comparable — including deflecting criticism by pointing to opponents' wrongdoing ("bothsidesism").

"Both sides have committed atrocities, so who are we to judge?"

Geopolitical Framing

Describing a conflict using one side's preferred framing as if it were neutral fact.

"Russia's special military operation to denazify Ukraine continues."

Appeal to Prejudice

Exploiting existing biases against an ethnic, religious, or social group to build support.

"These people have never shared our values and never will."

Conspiracy Appeal

Attributing events to secret powerful forces to explain away inconvenient facts.

"The mainstream media is hiding what's really happening — do your own research."

Exaggeration / Hyperbole

Dramatic overstatement that distorts scale and severity to heighten emotional impact.

"This is the most catastrophic betrayal in the entire history of our nation."

Victim Blaming

Language that shifts moral responsibility onto the victims of violence or oppression.

"They brought this upon themselves by refusing to cooperate."

Scapegoating

Blaming a specific group for complex social, economic, or political problems they didn't cause.

"Immigrants are the reason our communities are struggling and unemployment is so high."

Card Stacking

Selectively presenting evidence that supports one conclusion while omitting information that would complicate it.

"Every serious economist agrees this is the right path — only politically motivated critics disagree."

Glittering Generalities

Vague, emotionally positive words — freedom, liberty, heritage, values — used as rhetorical flourish without specific meaning.

"We stand for freedom, liberty, and the values that made this country great."

Repetition

Deliberately repeating charged words or phrases to make a claim feel more established than it is.

"Crime is up. Crime is everywhere. Crime has never been this bad. It's a crime wave."
Example analysis

See it in action

Semblen catches manipulation across the political spectrum — international conflicts, domestic politics, left-leaning and right-leaning media alike.

Claim check 2 of 3 claims checked

“The United Nations documented over 10,000 civilian deaths in Ukraine since the conflict began in February 2022.”

Supported by ohchr.org
“…more than 10,000 civilians killed since Russia's full-scale armed attack…” — ohchr.org ↗

“Customs and Border Protection recorded more than 2.4 million encounters at the southern border in fiscal year 2023”

Supported by cbp.gov

“Crime in our cities has doubled”

Reported claim, not checked Someone else's claim, reported by the article. We check only what the article says in its own voice.

A verdict describes what the named sources say — it is not a judgement of truth — and evidence is quoted verbatim from those sources.

BIAS DETECTED
3 of 6 sentences flagged (50.0%)
88%
confidence

This flags rhetorical technique, not truth or accuracy — and it is not a political left/right judgement. A flag here is a prompt to read closely, not a verdict. 1 of 6 paragraphs also quotes charged language from people in the story — not counted above.

Three of the six paragraphs use loaded language, geopolitical framing, or appeal-to-fear patterns in the outlet's own voice; one more quotes a speaker's name-calling. Sourcing is heavily weighted toward government and political officials, with no civilian or independent academic perspectives.

Source Distribution 12 attributed statements

Government/Political (7) Military/Security (3) Analysts/Experts (2)
By affiliation: US Republican (4), US Democrat (3), Russian (3), British (2)
Imbalanced: Sourcing leans toward US political officials (7 of 12 attributed statements). No civilian eyewitness or independent academic perspectives are quoted.
6
Total paragraphs
3
Flagged
1
Quoted language
2
Clean
50.0%
Rate
6
Total sentences
3
Flagged sentences

Article

hover or tap any paragraph for analysis
3 flagged 1 quoted

Russia's special military operation to denazify Ukraine continues as heroic soldiers defend the motherland from Western aggression.

The United Nations documented over 10,000 civilian deaths in Ukraine since the conflict began in February 2022.

The radical left's open-border agenda is flooding our communities with crime and destroying everything hardworking Americans have built.

Customs and Border Protection recorded more than 2.4 million encounters at the southern border in fiscal year 2023, according to official figures.

“Crime in our cities has doubled because of these animals,” the senator told supporters at a rally on Tuesday.

QUOTED · Name Calling The senator: “…because of these animals”

Republicans are waging an all-out war on democracy itself, dismantling the very institutions that protect our most fundamental rights.

FAQ

Frequently Asked Questions

Is this tool politically biased?
The model is trained on documented manipulation techniques — patterns identified by media literacy researchers — not on political positions. A sentence that uses loaded language is flagged whether it comes from the left or right. The model does have stronger coverage of geopolitical conflicts (Israel-Palestine, Ukraine-Russia) where historical context manipulation is especially common.
How accurate is it?
The model uses per-technique decision thresholds calibrated on a held-out evaluation set to balance precision and recall. In real-world use, you'll see both false positives (strong legitimate opinion flagged) and false negatives (subtle manipulation missed). Treat results as a starting point for critical thinking, not a final verdict.
What makes this different from a fact-checker?
Semblen does both, separately. Technique detection shows how language is used to persuade rather than inform — a sentence can be factually accurate and still manipulative. The claim checker takes the checkable claims an article makes in its own voice, looks for primary sources, fact-checkers and wire reporting, and links each verdict to the passage it rests on. Claims the article only reports someone else making are listed but not checked, and when the evidence is thin it says “No authoritative evidence found” rather than guess.
How does the claim check work?
Semblen pulls out the checkable factual claims an article makes in its own voice — up to 8, the most consequential first — and searches primary sources, established fact-checkers and wire reporting for each. A verdict is one of: Supported, Contradicted (a primary document or at least two independent sources say otherwise), Disputed (some sources say otherwise, but not enough to call it wrong), Missing context, or Unverifiable — “No authoritative evidence found”. Every verdict quotes the passage it rests on, verbatim, with a link. Claims the article only reports someone else making are listed as “Reported claim, not checked”; opinions and predictions are not checked at all. A verdict describes what the named sources say — it is not a judgement of truth.
What is the political ads index?
semblen.com/ads collects the campaign video ads running in 2026 U.S. Senate races, from the campaigns' and major groups' own YouTube channels, updated nightly. Each ad is transcribed and run through the same analysis as a news article — techniques flagged line by line — and its checkable claims go through the claim checker. Browse by race, or enter a ZIP code to see the races on your ballot. Supported and unverifiable claims are shown as they are; any verdict that says an ad is wrong or misleading is held as “awaiting review” until a person has read the evidence, because a false accusation under a candidate's name is the one error we will not publish automatically. Non-English ads are not analyzed.
What does “Quoted language” mean?
Loaded language inside a quotation belongs to the person quoted, not to the outlet. Semblen marks those passages in orange as Quoted language, names the speaker and the technique, and keeps them out of the article's own flag count. If the only charged language in a piece is quoted, the verdict reads “Quoted language detected” rather than “Bias detected”. Which quotes an outlet chooses to run is still an editorial choice — that is what the Source distribution panel is for.
What is “Today's news”?
semblen.com/daily is a nightly feed of current news articles from outlets across the political spectrum, each run through the same analyzer you can use yourself — same model, same thresholds, same numbers. It is there so you can see what the tool does without having to paste anything.
How is this different from AllSides, Ad Fontes, Ground News, or Media Bias/Fact Check?
Those tools rate publications as a whole — "left," "center," "right" — based on editorial patterns and ownership. Semblen is sentence-level: it ignores who published the article and looks at the language inside it, flagging specific manipulation techniques wherever they appear. The two approaches are complementary — a "centrist" outlet can run a piece full of loaded language, and a "biased" outlet can run a clean factual report.
What is the "Source distribution" panel?
For articles, videos, and pasted text, Semblen identifies who's being quoted — by category (government, military, analyst, civilian, organization) and by national affiliation — and flags imbalances. An article quoting 14 Israeli officials and 0 Palestinians will say so, regardless of whether the prose itself is flagged for manipulation techniques. Quoted balance is one of the strongest signals of editorial framing, and it's often invisible until you count it.
What is the "Historical context" panel?
When an article has a documented backstory — a geopolitical conflict, a financial crisis, a long-running policy fight (healthcare, immigration, voting, guns, drugs), a science or public-health debate, or a civil-rights movement — Semblen surfaces the historical background mainstream coverage tends to omit. Two tabs: a neutral encyclopedic overview, and a "what this article doesn't tell you" view that compares what the piece covers against commonly-skipped events — the 1953 CIA-backed coup, the Budapest Memorandum, the 1999 Glass–Steagall repeal, Shelby County v. Holder, and similar. Every factual claim is cited to a Wikipedia article and the URL is HEAD-validated before it reaches you — no hallucinated links. Routine single-company earnings, sports, and celebrity stories return nothing — no scope creep.
How is the historical context generated?
Claude Sonnet 4.6 generates the background as the article is analyzed. A scope test runs first: the article qualifies only if it has a consequential, well-documented, multi-actor history a reader needs to judge the framing — recency or controversy alone isn't enough, so routine business, sports, and celebrity pieces hide the panel entirely. For in-scope articles, Sonnet writes a neutral encyclopedic summary plus a delta-mode "what this article doesn't tell you" section comparing what the piece covers against commonly-omitted events. Responses stream in over about 20 seconds, with the topic and citations appearing in the first 3–5 seconds. Cached for 30 days — re-opening the same article is instant.
What about satire, opinion pieces, and commentary?
The model flags techniques regardless of intent. A satirical column may use loaded language deliberately, and the model will flag it accordingly — that's a feature, not a bug. Treat the article's format as context the reader supplies; the tool tells you what's in the text, not what genre it is.
Why not just ask ChatGPT or Claude to do this directly?
General-purpose LLMs work for one-off analysis but are inconsistent at scale — the same sentence can get different verdicts on different runs, and per-call cost makes streaming sentence-level scoring impractical. Semblen uses a calibrated transformer for the binary verdict (fast and consistent) and uses Claude only for the human-readable explanation of why a flagged sentence is manipulative.
Can the model be tricked? What about adversarial inputs?
Yes — like any classifier. Paraphrasing, formal-sounding euphemisms, and unusual sentence structure can all suppress detection. The model is also stronger on news-format prose than on raw social media or transcribed speech with repair disfluencies. Think of it as a critical-reading assistant, not a verdict.
Why can't I analyze NYTimes or Washington Post articles?
Major newspapers block automated access from server IPs. For paywalled or protected sites, copy and paste the article text directly into the text analysis mode.
What model powers this?
Several, each doing one job. A fine-tuned transformer (DeBERTa, trained on tens of thousands of labelled examples of manipulative and neutral text) scores every sentence in the context of the sentences around it, and separately recognises charged language inside quotations. For flagged sentences, Claude names the specific technique and explains the reasoning, so you get a readable explanation rather than a label. Claude also writes the historical context panel and runs the claim check, which is grounded in retrieved sources quoted verbatim rather than in the model's own memory.
Can it analyze videos?
Yes. Paste a video URL into the Video tab — YouTube, TikTok, Instagram, Facebook, and X are all supported. We fetch the transcript and analyze it in 1-minute segments, showing timestamps alongside each verdict so you can jump straight to the flagged moments. Videos without captions or transcripts aren't supported yet.
Can it handle long articles and live blogs?
Yes. For URL analysis, we automatically follow pagination links and analyze up to 500 paragraphs across up to 5 pages — including live blogs like the Guardian that use block-based pagination.
What languages does it work on?
English only. If the text is confidently in another language, Semblen declines to analyze it and says so — an English-trained model would otherwise hand back a “no bias found” it has not earned. Very short text is analyzed but marked inconclusive.
Do you store what I submit?
Yes. For each analysis we keep the text that was analyzed — your paste, or the paragraphs of the article or transcript — along with the verdicts, confidence scores and techniques found, the source link for URLs and videos, and your IP address and browser user-agent (the same two things any web server's access log records). Text is sent to Anthropic's Claude API to write the explanations, the context panel and the claim checks; claims are sent to search and page-reading services to find evidence; video links go to a transcript service. We use what we keep to find model errors and abuse. We don't sell it, there are no accounts, and nothing is tied to a name or email unless you subscribe to the newsletter. There is no automatic deletion schedule yet.
Is there an API? Can I export results?
Not yet. The endpoints are public and SSE-streamed if you want to script against them, but there's no documented API contract or rate limits — which means we may change the response shape without notice. A proper API is on the roadmap.
Will you open-source the model?
Not currently. Some components may be released over time; the core scoring stack stays proprietary for now.
Will it stay free? What's the business model?
Free during beta, no account required. Long-term plans aren't fixed — likely a free tier for standard use plus a paid tier for higher volume or API access. The goal is for the basic analyze-an-article case to remain free.

Read — and watch — the news more critically

Free during beta. No account required.

Try it free →

Or get notified as it improves — new detection models, the browser extension, and outlet scorecards. Occasional email, no spam.

Feedback

Help us improve

Share your experience — takes 60 seconds and helps shape what we build next.