About verifAI

Short-form video has become the dominant way many people find out about the world, and it is structurally hostile to verification: loops are short, framing is emotional, claims are often hidden in caption overlays rather than spoken aloud. verifAI exists to put a fast, source-cited, transparent second opinion one paste away.

Paste any Instagram reel, TikTok video, or news article URL. The pipeline downloads the media (videos only, kept just long enough to process and then deleted), transcribes the audio, reads the on-screen text, splits the content into atomic factual claims, and checks each claim against Google Fact Check Tools and a live web search.

Each claim ends up on a verdict card with one of five ratings (true, false, partially true, misleading, unverified), a short explanation, and the URLs that were actually used to reach the verdict. The verdict is yours to read, share, or disagree with — all source links are visible so you can verify the work.

1. Canonicalise the URL and queue the job
2. Download the media via Apify → yt-dlp → RapidAPI fallback chain (videos only)
3. Transcribe audio with Whisper-large-v3-turbo and read on-screen text with Llama-4 Scout
4. Extract atomic claims and discard opinions, jokes, and rhetorical framing with GPT-OSS-120B
5. Cross-check each claim with Google Fact Check Tools and a live Tavily web search; stream verdicts to the browser as they resolve

The product runs on Next.js, deployed on Fly.io, with Supabase for auth and history storage and Stripe for the optional paid plans. Inference runs through Groq (Whisper-large-v3-turbo for speech, Llama-4 Scout for vision, GPT-OSS-120B for claim extraction and verdict synthesis). Search providers are Google Fact Check Tools and Tavily, with optional SearXNG support for self-hosters. Product analytics use PostHog.

We do not retain submitted media. The pipeline uses a per-job temporary directory that is wiped when the job finishes. Only the resulting text verdicts and their source URLs are persisted in your account so you can return to your history later. Read the full privacy policy for the complete subprocessor list, retention periods, and your data-subject rights.

Read the privacy policy →

verifAInow.es is an independent project operated by Fernando Rueda Oliva, based in Spain. It is not affiliated with Instagram, TikTok, Google, OpenAI, or any of the news organisations whose verdicts we surface. The codebase is small enough that one person can read it end-to-end, which we consider a feature, not a limitation.

Bug reports, factual corrections to a verdict, partnership ideas, or just an interesting reel you'd like us to look at — all welcome at fernandoruedaoliva@gmail.com.