You asked for a three-sentence summary. What arrived was three sentences, wrapped in an opening line explaining what it was about to do, a caveat about how this depends on your specific situation, and a closing offer to expand on any section. You now delete more than you keep.
This is not a defect. Models are trained to be helpful, and helpfulness, in the absence of any instruction to the contrary, looks like giving you more. More context, more hedging, more disclaimers, more balanced both-sides paragraphs, more of the passive voice that lets a sentence avoid committing to anything. The model has no way to know that the extra material is the reason you cannot use the output.
So tell it. Not in the form of another request, which just adds material, but as a fence.
Constraint prompting is saying what must not appear, in advance, in writing.
The failure modes it removes
- Disclaimers and hedging. "It is worth noting", "generally speaking", "you may want to consult a professional". Each one is a sentence that survives review because nobody can call it wrong.
- Padding around the point. Preambles announcing the answer, summaries repeating it, closing offers to help further.
- Boilerplate expertise. Generic best practices bolted onto a specific question, which read as filler to anyone who actually does the job.
- Structure you did not ask for. Bullet points in a letter, headings in a paragraph, an example where you wanted a definition.
This matters most in legal, medical and financial writing, where the padding is not merely annoying. One stray hedge turns a clear instruction into an ambiguous one, one generic disclaimer in a client-facing note reads as a failure to take a position, and the entire document loses its authority to a line that nobody chose to write.