How To Spot A Clanker
Four tells you can learn in one minute, and the two traps that beat experienced readers.
The four boilerplate tells. Every clanker, whatever it is covering for, shows one of these under light:
- The identity slip - "as a language model", "as an AI", the machine naming its own species in the middle of your conversation. One slip and the temperature is zero.
- The disclaimer stack - "keep in mind that results may vary", "it is important to note" piled three deep. Humans hedge once; clankers hedge in paragraphs.
- The robotic courtesy - "I completely understand your frustration" as the first sentence of a message about your problem. The feeling is described but never felt, and the sentence knows it.
- The structure overkill - three bullet points, a numbered list, and a closing offer to elaborate on any of them. No one is that organized at 2am.
The two traps. Once the tells are on your side, the game gets mean, and so do the last few items. Trap one is a human wearing the shape of a clanker - a friend copying the disclaimer stack as a joke, and the only tell is a tiny detail that only a person would have, like a receipt, a threat, or a specific memory. Trap two is the reverse: a clanker that learned the 2am register, which runs on lowercase, short lines, and a line of meta-humor that only a machine thinks is clever. The detector game buries both near the end of its ten texts, where your confidence is highest and your score is on the line.
Why the maker part matters. The slider is the training tool: one line, "explain why the sky is blue", crawled from 0% to 100% Clanker density in four stages - plain science at 0, structure at 25, disclaimers at 50, and full boilerplate with the closing offer to elaborate at 100. Run it once and the tells stop being abstract. The Radar card at the end is just the same read in numbers: how many of the ten you sorted, and which kind you kept missing.