How Clad Works

Clad reviews news broadcast segments — network newscasts, morning shows, roundtables, interviews, and opinion programs published on YouTube — along with occasional notable clips from independent creators and specialist channels submitted by the editor, all graded by the same criteria. Every report links directly to its source video, so you can watch what we watched and judge for yourself.

Each broadcast report carries three measurements:

• Letter grade (A+ through F). An overall quality grade that weighs severity, not just count: A means high accuracy, sound sourcing, and balance; B means mostly accurate with minor concerns; C means a mix of accurate and problematic claims with noticeable patterns such as one-sided sourcing or missing context; D means significant factual issues, heavy partisan framing, or major unsourced claims; F means pervasive misinformation or propaganda-level distortion. Every grade is published with a one-line rationale naming the specific strengths or failings that earned it. Browse the archive by grade: A+, A, A−, B+, B, B−, C+, C, C−, D+, D, D−, or F.
• Factuality score (0–100). Places the segment on a continuum from entirely false (0) through mixed (50) to entirely accurate (100). Severity matters: three minor missing-context issues are not the same as one outright false claim on a load-bearing point.
• Political lean. A direction and magnitude from -100 (strongly left) through 0 (centered) to +100 (strongly right), measuring the slant of the segment as presented — word choice, which facts are emphasized or omitted, guest and source selection, and framing. The same standard applies to every network and figure, and the lean is published with its own rationale naming the specific cues observed. Non-political content scores 0.

Each report isolates three to five of the segment's most load-bearing claims and labels each one verified, disputed, missing context, or unsupported, with a one-sentence note giving the source, context, or reasoning behind the verdict. Where evidence is genuinely thin, we prefer "missing context" or "unsupported" over guessing.

Each labeled claim is also published as machine-readable ClaimReview markup so fact-check aggregators can ingest it. The four labels map onto ClaimReview's 1–5 rating scale as: verified = 5, missing context = 3, disputed = 2, unsupported = 1 (no label maps to 4; "unverified" single-claim posts carry the label without a numeric rating).

Reports are AI-drafted: xAI's Grok analyzes the broadcast and verifies its factual claims against current authoritative sources via live web search before any grade is assigned. The AI's role is research and drafting — it produces the structured report card, the rationales, and the sourced claim-by-claim verdicts.

No report publishes itself. Every draft goes through a human approval queue, where the editor reviews it — and can send it back for revision — before it appears on the site. Every verdict is hand-reviewed. We disclose this because you deserve to know the role AI plays, why we use it (it can check every claim in a broadcast against primary sources at a speed no newsroom of one could match), and that a human remains accountable for everything published.

Reports cite their sources — typically four to six per report, and often more. Primary sources are preferred: official documents, statistics, statutes and filings, named experts, and reputable outlets across the spectrum. Wikipedia and other tertiary encyclopedias are listed under "Background," never as Sources Consulted. Dead links are pruned before publication.

Headlines describe; they don't judge. The verdict lives in the grade and the report, not the headline — no clickbait, no exclamation, no political adjectives applied to people. Adjectives in the assessment describe the evidence itself ("documented," "unsubstantiated"), never individuals.

When a report needs correcting, the correction is issued as a new post that links the original — never a silent edit. Both posts carry a visible correction banner, and every correction is logged publicly on the corrections page.

Every report carries a flag mechanism. If you believe a grade or political-lean rating is wrong, say so — your note goes to the editor and the report is re-run through the exact same grading process that produced it. If the fresh result confirms the original was off, the report is removed; if it holds up, it stays.

The full report — summary, editorial assessment, key moments, and sources — is free to read by anyone. The scoreboard — letter grade, factuality score, political-lean rating, and social sentiment — plus Reader Reactions, charts, and midterms tools unlock with a free account (no card required).