Built by Metorial, the integration platform for agentic AI.
Scans text content against internet sources to identify duplicate or similar content. Returns an overall plagiarism score, matched sources with URLs and titles, word-level match details (identical vs. similar), citation detection, and obfuscation attack detection. Accepts plain text, a publicly accessible file URL (.pdf, .doc, .docx), or a website URL. Supports 45+ languages with auto-detection. You can exclude specific domains or URLs from the scan and tailor results by country.
Analyzes an image to determine whether it was generated by AI or created by a human. Returns a human score (0-100), AI and human probability percentages, and extracts provenance metadata including C2PA content credentials, EXIF/IPTC data, and AI watermark detection. Identifies images from major AI generators including ChatGPT Images, Midjourney, Stable Diffusion, Adobe Firefly, Ideogram, Leonardo, Runway, and others. Also detects manipulated faces, identity swaps, synthetic personas, and deepfakes.
Analyzes text content to determine if it was written by a human or generated by AI. Returns a human score (0-100) where higher values indicate more likely human-written content. Can also provide per-sentence scores for granular analysis, a readability score, and detection of obfuscation attacks such as zero-width spaces or homoglyph substitutions. Accepts plain text, a publicly accessible file URL (.pdf, .doc, .docx), or a website URL as input. Supports 12+ languages with auto-detection.
Compares two pieces of text to identify similarities and differences. Returns an overall similarity score (0-100), per-text matching word counts, similarity percentages, and detailed item-level match positions. Useful for checking if two texts share common content, detecting paraphrasing, or verifying text originality between documents.
Verifies the factual accuracy of content by extracting claims and checking them against trusted sources. Returns an overall accuracy score (0-100), and for each claim provides a verdict (SUPPORTED, PARTIALLY_SUPPORTED, NOT_ENOUGH_EVIDENCE, or REFUTED), a confidence score, an explanation, and reference links. Accepts plain text, a publicly accessible file URL (.pdf, .doc, .docx), or a website URL. Supports 38+ languages with auto-detection. Extracts up to 12 claims per request.