Know what’s on your ballot before you vote

Not in the usual sense. Media bias checkers rate an outlet or an article on a left-to-right scale. Semblen does not score political slant: it flags the rhetorical techniques in the language itself — loaded language, appeals to fear, false urgency and 15 more — marked word by word, wherever they appear and whichever side uses them, and it checks the factual claims separately against cited sources. A flag says how the wording tries to persuade, not which way it leans or whether it is true.

The model is trained on documented rhetorical 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 the background a reader needs is often missing from the coverage.

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 techniques missed). Treat results as a starting point for critical thinking, not a final verdict.

Semblen does both, separately. Technique detection shows how language is used to persuade rather than inform — a sentence can be factually accurate and still persuade by technique rather than evidence. The claim checker takes the checkable claims an article makes, and up to 5 it reports other people making, looks for primary sources, fact-checkers and wire reporting, and links each verdict to the passage it rests on. When the evidence doesn’t settle a claim it says so (“The sources we found don’t settle it”) rather than guess.

Semblen pulls out the checkable factual claims an article makes itself — about three-quarters of them, at least 8 and at most 25, the most consequential first — plus up to 5 claims it reports someone else making, each restated as that speaker’s own assertion and shown with who said it; then it 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 — “The sources we found don’t settle it”. A claim about what caused what is never ruled on, and says so. Every verdict quotes the passage it rests on, verbatim, with a link. A verdict on a reported claim describes what the sources say about the speaker’s assertion, and one saying the speaker is wrong or misleading is shown only when it cites a fact-checker or a primary source that is not the speaker’s own; otherwise it reads “No verdict shown”. Reported claims past the 5, or too vague to state as a checkable fact, 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.

semblen.com/races collects the political ads running in 2026 U.S. Senate races — Facebook and Instagram ads from Meta's Ad Library, and video ads from the campaigns' and major groups' own YouTube channels — updated nightly. Each race page also shows the money behind every candidate from FEC filings: fundraising totals, outside spending for and against them (linked to the spender's own ads on the page), money bundled through PACs — every PAC that forwarded $100,000 or more to a candidate is called out, whatever its cause — and each super PAC's largest donors. Each ad is run through the same analysis as a news article — techniques flagged line by line — and on any ad you can press “Check this ad's claims” to run its checkable claims through the claim checker (the result is then shown to everyone). Browse by race, or enter a ZIP code to see the races on your ballot. Supported and unverifiable claims are shown as they are. A verdict that says an ad is wrong or misleading is shown only when it cites an established fact-checker or an institutional record — Congress, a federal agency, a court — and its confidence is not low; a weaker one is not shown at all. Every verdict shows what it rests on and how confident it is, and has Agree / Disagree buttons; a verdict readers dispute and no one defends comes down. Non-English ads are not analyzed.

semblen.com/government analyzes what the government says, nightly: releases and statements from the White House, House Speaker Mike Johnson and House Democratic Leader Hakeem Jeffries, the floor speeches of Senate Majority Leader John Thune and Democratic Leader Chuck Schumer from the Congressional Record, and releases from the Departments of War, Justice and Homeland Security. One rule for every source: each night, the newest three items from each that we have not analyzed before. Every item is listed whether or not anything was flagged, the two parties sit side by side in each chamber, and we do not publish rates by party.

Loaded language inside a quotation belongs to the person quoted, not to the outlet. Semblen marks those passages as Quoted language (a dashed edge in the article), names the speaker and the technique, and keeps them out of the article's own count. Which quotes an outlet chooses to run is still an editorial choice — that is what “Who gets to speak?” is for.

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.

Those tools rate publications as a whole — "left," "center," "right" — based on editorial patterns and ownership. Semblen reads the article itself: it ignores who published it and looks at the language inside it, paragraph by paragraph, flagging specific rhetorical 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.

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 rhetorical techniques. Quoted balance is one of the strongest signals of editorial framing, and it's often invisible until you count it.

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: a short list of what the article leaves out, compared against commonly-skipped events — the 1953 CIA-backed coup, the Budapest Memorandum, the 1999 Glass–Steagall repeal, Shelby County v. Holder, and similar — and a short list of background, with the full text one tap away. 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.

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 show no “What’s missing?” card. 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.

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.

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 trained model to decide which paragraphs use a technique (fast and consistent) and uses Claude for what needs language: the TL;DR, the explanation of how the flagged words work on a reader, the background and the claim checks.

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.

Many big newspapers block requests from servers like ours, and a paywalled article can only be read with your subscription. On a computer, use the Semblen Chrome extension: open the article and click the icon. It reads the page you already have open, so any article you can read works, paywalled ones included. On a phone, select the article's text, copy it and paste it into the analyzer's text box.

Several, each doing one job. A fine-tuned transformer (DeBERTa, trained on tens of thousands of labelled examples of rhetorical techniques and neutral text) scores each whole paragraph, then the sentences in it to find the words to mark, and separately recognises charged language inside quotations. For flagged paragraphs, Claude names the specific technique and explains the reasoning, so you get a readable explanation rather than a label. Claude also writes the TL;DR and the “What’s missing?” background and runs the claim check, which is grounded in retrieved sources quoted verbatim rather than in the model's own memory.

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 finding so you can jump straight to the flagged moments. Videos don't get a TL;DR yet, and videos without captions or transcripts aren't supported.

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.

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 clean result it has not earned. Very short text is analyzed but marked inconclusive.

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 TL;DR, the explanations, the background 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. You don't need an account; if you sign in, we keep your email address and your plan. Analysis records are deleted automatically after 180 days. See the privacy policy.

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.

Not currently. Some components may be released over time; the core scoring stack stays proprietary for now.

Yes: 5 analyses a day are free, with no account required, and the daily news, government and political-ads pages are free to read. Semblen Pro, a monthly subscription, adds more analyses, the full findings for each article, how other outlets covered it, and a daily digest of the news you follow. Every article's TL;DR is free.