Policy

Updated September 23, 2026.

ClipCast.News compares creator commentary with links to the underlying evidence. Creator opinions are not independent verification that a claim is true. Political profile labels do not establish a creator’s position on an individual event.

How articles are produced

Transcript-based articles are AI-written from the linked recording and transcript. New transcript articles and edits must carry a completed assessment from our desktop fact-checking software before publishing automatically. The language model runs on the administrator’s computer; this workflow does not send articles or transcripts to OpenAI or another hosted AI model and needs no AI API key. It uses internet search and reads public source pages, so search providers receive claim-based queries and publishers receive normal page requests. Completed articles, transcripts, claim assessments, short evidence excerpts and source links are uploaded to ClipCast.News. Existing articles may have no current check. Publication does not imply independent human verification. AI can mishear words, confuse speakers, omit qualifications and produce inaccurate summaries. The preserved transcript is machine-generated evidence, not a certified record.

Fact-check ratings

True, Mostly true, Maybe, Mostly untrue and Lie are assessments of the specific factual claim shown in a fact check, not an entire article or a creator’s political opinions. Each published assessment requires an explanation and evidence links. Lie requires evidence of deliberate deception. Not checked is separate from Maybe. A changed article requires a new fact check before its previous verdict is displayed.

Reader discussions

Reader comments are public contributions, not endorsed or verified reporting. Keep discussion relevant and respectful. Do not post harassment, threats, private information or spam. Authors may remove their own comments; administrators may remove abusive comments. Send concerns to our contact address with the article or comment link.

Original value and accuracy

Before considering an article for advertising, a human reviewer must check source context, material factual claims, attribution and independent supporting evidence; add meaningful original analysis, context or comparison where needed; and inspect the complete page and its linked media. Reviewers must not invent experience, expertise, quotations or reporting. Nine paragraphs alone do not demonstrate quality or Google policy compliance.

Rights and sensitive content

Publicly accessible videos do not automatically grant permission to republish their full transcripts. Rights for recordings, transcripts, images and quotations must be checked. Coverage involving health, elections, violence, sexual content or other sensitive topics may need corrections, removal or exclusion from advertising depending on the content and context.

Advertising review is separate

Every article requires a separate human review before consideration for advertising; changes to the article invalidate that review. Google advertising remains disabled until account approval and required configuration are complete. Google makes its own site and content decisions. We do not claim Google approval.

How our desktop fact-checking works

The local model extracts verifiable assertions from the headline, summary and each paragraph. It distinguishes factual reporting, attributed statements and opinion. A creator’s transcript can establish what was said; it does not prove that the statement itself is true. Political affiliation and popularity do not determine a verdict.

The software searches for original records, independent reporting and potential corrections or contradictions. It reads accessible text pages rather than treating search snippets as proof. For each factual claim, it requires evidence identified as an original source plus corroboration from at least two independently hosted publishers. The model assesses whether coverage is independent; different domain names alone do not establish independence. Copied, syndicated and mutually citing reports are not intended to count as independent confirmation.

The local model compares the claim, its article context, relevant transcript passages and retrieved evidence. Citations must include short passages that exactly match the text read by the software. It records source links and retrieval times. The visible main verdict concerns the identified central claim; the expanded evidence section lists the other examined claims. The website checks the report’s structure, required coverage, freshness, source records and match to the submitted article. It trusts submissions only from authorized administrator software. It does not independently fetch all those pages or guarantee that a local model’s reasoning, claim extraction or independence classification is correct.

True means the cited evidence supports the identified claim; Mostly true means it is supported with qualifications; Maybe means researched evidence remains unresolved or conflicting; Mostly untrue means the evidence substantially contradicts it. The desktop software never assigns Lie automatically: that label requires an administrator’s documented evidence of deliberate deception. Opinions receive no truth rating. A report may accurately describe a false claim without endorsing it.

A search outage, inaccessible source or incomplete check is not a Maybe verdict. If the program cannot retrieve enough evidence, cannot match its citations or finds unsupported reporting, it retains the article unpublished and explains the problem. Paywalls and access restrictions are not bypassed. PDF-only, interactive, private or removed sources may be unavailable to this text reader. Reddit, reposts and other secondary accounts are not direct creator statements; any displayed creator attribution relying on them remains “Uncertain · not a direct creator statement.”

Checks are cached locally for up to 24 hours to resume interrupted work. Changed article text requires a matching new assessment. The desktop program must be running and connected to process new submissions and queued edits. It also works through existing transcript articles without a check when their original transcripts are available; older news briefs without stored transcripts are not automatically backfilled by this version. A missing check must never be presented as a completed verification. AI assessments can miss context, overlook claims, misread sources or produce incorrect conclusions. Readers should inspect the linked evidence and report errors.

Reader challenges

Each signed-in account may cast one vote per fact-check version. At least 10 votes are required; 75% or more requesting review flags the check for administrator review and pauses advertising eligibility. This is a share of voters, not all readers or registered accounts. Votes do not establish truth. A flag stays open until an administrator publishes a new sourced assessment. Vote records are retained with your account identifier for integrity and review history.

Corrections

Send the article URL, the disputed passage and supporting evidence to contact@noderealm.com. Administrators can correct or unpublish articles while retaining source evidence and an audit record. Read how creator perspectives are assessed.