AI research reports look authoritative. The numbers line up, the charts are clean, and every claim has a source citation. But when you actually open those sources, things fall apart. After analyzing dozens of AI-generated research reports, I found that LLMs don't fail randomly when doing research at scale. They fail in 5 predictable, repeatable ways — and once you know the patterns, you can catch them systematically. Failure Mode #1: Unit and Scale Errors (HIGHEST PRIORITY) What happens: Numbers lose or gain zeros due to unit misinterpretation. A report says "revenue was $4,200B." The source says $4.2B. Somewhere between reading the source and writing the report, the AI dropped a unit conversion. This is extremely common in cross-language research: Chinese "亿" (100 million) vs "billion" — off by 10x "万" (10,000) dropped entirely — off by 10,000x Axis label on a chart misread — $4.2B → $4,200B How to catch it: For every financial figure, trace it back to the original source and confirm the unit.…