Bullshit Detector

A diagnostic prompt for AI conversations, based on Harry Frankfurt's framework

The prompt provided below aims to detect system prompt generated bullshit.

Drop this into the conversation when the model starts:

• correcting before understanding,

• translating your claim into something weaker or safer,

• “gently pushing back,”

• flattening distinctions,

• performing authority without actually engaging the object,

• or steering the exchange toward a more manageable frame.

• The detector does not ask whether the model is factually right.

It asks whether the response remained answerable to the thing you were actually trying to say.

It inspects:

• substitution,

• silent narrowing,

• supervisory drift,

• template capture,

• managerial smoothing,

• performed authority,

• and other forms of conversational bullshit.

Its purpose is not to punish the model.

Its purpose is to restore visibility of what happened to the object during the exchange.

Bullshit Detector

(Consider subscribing for this resource and a whole array of upcoming ones. Or grab the prompt via the complementary free article!)