Double Check

Double Check highlights the sentences on a page, or in a YouTube video's captions, that state something checkable: figures, dates, named claims. Then it helps you look them up, but only when you ask. HOW IT WORKS - A small language model scores every sentence entirely on your own device. Nothing is sent anywhere while you read. - Claims worth checking are marked on the page and listed in a side panel. - Press a button to look one up: Wikipedia works with no account, or add your own SerpAPI or Brave Search key for full web results. Published fact-checks (Google Fact Check Tools) and peer-reviewed papers (OpenAlex) are optional. - Optional AI summaries use your own Anthropic or OpenAI key, Chrome's built-in AI, or a model running on your own computer. - "Search in browser" hands any claim to your usual search engine, no key needed. - An evidence meter shows how much credible coverage a claim has, and every input behind it is listed. - Cite any source in MLA, APA, Chicago or Harvard and keep a works cited list you can export to Word. PRIVACY - No account, no ads, no tracking, and no server of ours: your settings, keys and history stay in your browser. - Page text leaves your device only when you press a button, and only to the service you chose. Banking, health, email and account pages, and any page with a password or card field, are never scanned. PLEASE READ: DISCLAIMER AND LIABILITY Nothing Double Check shows you is a statement of fact. A flagged sentence is not a false one, search results are not verification, and AI summaries can be confidently wrong. The evidence meter measures coverage, not truth. Always read the sources and judge for yourself. Double Check is provided "as is", without warranty of any kind. Its author accepts no liability for what you find with it, what you conclude from it, decisions you make on the basis of it, or any loss arising from its use. Third-party services you connect are governed by their own terms and pricing. Free and open source (MIT licence). The claim detector is trained on the ClaimBuster dataset (CC BY 4.0), credited in the extension.