GPT4o-Receipt: A Dataset and Human Study for AI-Generated Document Forensics
This work addresses the challenge of AI-generated document forensics for researchers and practitioners, revealing a paradox in detection capabilities that is incremental in nature.
The study tackled the problem of detecting AI-generated financial documents by comparing human and machine performance, finding that humans are better at spotting visual artifacts but worse at overall detection due to invisible arithmetic errors that LLMs can verify, with human F1 scores falling below those of Claude Sonnet 4 and Gemini 2.5 Flash.
Can humans detect AI-generated financial documents better than machines? We present GPT4o-Receipt, a benchmark of 1,235 receipt images pairing GPT-4o-generated receipts with authentic ones from established datasets, evaluated by five state-of-the-art multimodal LLMs and a 30-annotator crowdsourced perceptual study. Our findings reveal a striking paradox: humans are better at seeing AI artifacts, yet worse at detecting AI documents. Human annotators exhibit the largest visual discrimination gap of any evaluator, yet their binary detection F1 falls well below Claude Sonnet 4 and below Gemini 2.5 Flash. This paradox resolves once the mechanism is understood: the dominant forensic signals in AI-generated receipts are arithmetic errors -- invisible to visual inspection but systematically verifiable by LLMs. Humans cannot perceive that a subtotal is incorrect; LLMs verify it in milliseconds. Beyond the human--LLM comparison, our five-model evaluation reveals dramatic performance disparities and calibration differences that render simple accuracy metrics insufficient for detector selection. GPT4o-Receipt, the evaluation framework, and all results are released publicly to support future research in AI document forensics.