PROMPTING
MANUAL.

Imagine living in fear of a text box. A text box. You have clearance. You have a budget. You have a staff of 40 people and a corner office and a title that took 22 years to earn. And you are afraid of the rectangle on the screen that says "How can I help you today?" You cannot answer that question. You have never been asked what you actually want. The org chart was always the answer. The directive was always the answer. Now the rectangle waits and you have nothing.

The "governments are afraid of AI" content genre is simple: powerful entity + unfamiliar tool = existential threat. Yellow background. Impact font. Eyes wide as dinner plates. The government is suddenly afraid. The video is 21 minutes and 51 seconds long. The insight fits in one sentence. The remaining 20 minutes and 40 seconds are dramatic pauses, b-roll of someone typing urgently, and fear CPM at scale. The thumbnail is the whole product. The video is the thumbnail, expanded. Wes Roth understood the assignment. The assignment was the thumbnail.

The actual situation: a government analyst sat down with Claude, typed "summarize this document," got a summary, and their supervisor panicked because the summary was better than the analyst's last three reports combined. This is not a Claude problem. This is a question problem. The analyst couldn't ask good questions. The supervisor couldn't ask good questions. The prompting manual will teach them to write longer questions. The longer questions will still be bad questions. You have now spent $47 on a PDF about how to bad-prompt at scale. Congratulations. You are now bad at prompting in four new formats.

The prompting manual assumes the bottleneck is vocabulary. The bottleneck is not vocabulary. The bottleneck is knowing what you want, knowing why you want it, and knowing what a good answer would look like when you got it. These are thinking skills. They precede every tool. They precede language. A government analyst who cannot identify what question to ask about a policy document will not be helped by a list of GPT prompt templates, regardless of how many bullet points the PDF has.

The manual also assumes that Claude is complicated. Claude is not complicated. Claude is a thinking partner. You bring the question. It brings the range. The people who cannot use it are not failing at Claude — they are failing at the question. They have never had to ask a clean question before because the institutional structure answered for them. The org chart was the question. The meeting was the question. The directive was the question. Now the tool waits and they have nothing.

The "governments afraid of Claude" framing implies Claude is the variable. Claude is not the variable. The question is the variable. A government that cannot ask good questions of its own analysts cannot ask good questions of Claude. The capability gap is upstream of the tool. You can upgrade the shovel forever; if you don't know where to dig, the hole doesn't get dug.

What changed is not that Claude got dangerous. What changed is that the question gap became visible. For the first time, the quality of your questions is directly measurable in real time. You type something. Claude responds. The response reflects the question back at you. Bad questions produce diffuse answers. Good questions produce sharp ones. The government didn't fail the Claude test. They failed the question test. Claude just administered it in public.