AI-generated content, text, images, audio, and video, is now indistinguishable from human-authored material at a glance. Fact-checking it requires a different workflow than checking traditional sources: instead of asking 'is the source reliable', you first ask 'is this content authentic', then 'is the claim it makes accurate'.
For text, the two questions to answer are provenance (was this written by a human or an LLM) and factual accuracy (are the claims true regardless of author). VerumVerify's Press-tier AI detection scores text on the 0–100 scale, flags mixed / AI / human, and highlights suspect passages. The claim itself is then routed through the standard fact-check engine so you get both authorship signal and truth verdict.
For images, the check has three layers: statistical AI-generation detection (Midjourney, Stable Diffusion, DALL·E, Flux artifacts), forensic analysis (compression fingerprints, splice detection, ELA, lighting inconsistencies), and reverse image search via Google Vision to find the earliest known appearance and original context. VerumVerify's Verify-an-Image tool runs all three in one pass and OCRs any embedded text into the standard claim engine.
For audio and video, the immediate goal is deepfake and voice-clone detection combined with source-of-truth verification: where did this clip first appear, was it edited, is the transcript accurate, and does the claim inside it hold up. Wayback / dead-link recovery matters here too, since manipulated clips often reference sources that were later scrubbed.
Practical workflow: (1) run the content through an authenticity check appropriate to its medium, (2) extract the underlying factual claim in plain text, (3) fact-check that claim against multiple independent sources, (4) archive the original with a timestamp so the evidence chain survives even if the source is deleted. VerumVerify handles steps 1–3 in a single pipeline and stores the result in your journal for step 4.