Library.

“Substantive disagreements between AI models on fact-checking outcomes are common.”

Evidence from multiple studies shows that AI fact-checking models often reach materially different verdicts on the same claim, with reported substantive conflicts commonly in the roughly 15% to 30% range on challenging datasets. That is frequent enough to count as common in real-world use. Rates do vary by claim difficulty, ambiguity, prompting, and evidence quality.

“Publicly posted online content can be scraped and used to train artificial intelligence models.”

The claim is accurate as a statement of technical capability and widespread industry practice. Publicly posted online content is routinely scraped to train AI models—confirmed by academic research, corporate disclosures (e.g., Google's privacy policy), and the existence of major datasets like Common Crawl. However, the claim omits critical legal context: copyright law, privacy regulations, terms of service, and the EU AI Act (fully enforced in 2026) all impose significant restrictions. "Can be done" is true; "can be done freely and lawfully in all cases" is not.