Zhihu Answer Prompt Builder

Conclusion-first structure is what makes a Zhihu answer land

Zhihu (知乎) is China's leading question-and-answer community, and the answers that rise to the top there share a recognisable shape. They lead with a clear conclusion, back it with arguments, data and real cases, and lay everything out point by point so a reader skimming on a phone can follow it in seconds. Writing that by hand every time is tedious; getting an LLM to write it well means describing that shape precisely in the prompt. This builder does exactly that. You fill in the question you are answering, the stance or conclusion you want to make, your professional background, the evidence you want cited, the structure, the tone and a target length — and it joins them into one clean, well-headed prompt with a leading "you are…" identity line, ready to paste into DeepSeek, Qwen, Doubao, Kimi, ERNIE or Zhipu. The result is the kind of brief a thoughtful Zhihu writer would give themselves before they start typing.

The two highest-leverage fields are the question and your stance. Together they fix the spine of the whole answer: the model knows exactly what it is responding to and which side it is arguing, so it stops hedging and writes with conviction. The default structure — conclusion first, then point-by-point argument, then examples, then a short summary — is not arbitrary; it mirrors how the strongest answers on the platform are built, and it is also how people actually read on a small screen. By stating the structure explicitly in the prompt, you keep the model from collapsing your answer into one undifferentiated wall of text, which is the single most common reason a good point gets scrolled past.

Identity, evidence and length separate a draft from an upvoted answer

The fields people skip and regret are identity, evidence and length. Identity is what builds trust on Zhihu, where readers visibly care about who is answering: one concrete line — "a doctor with fifteen years in emergency medicine", "a founder who has shipped two SaaS products" — does more for credibility than any amount of confident phrasing. Evidence is the skeleton: arguments, figures and real cases are what let an answer survive the follow-up questions and counter-arguments that fill the comments. And length matters more than it seems — set it too short and the answer feels thin, too long and readers abandon it; a target keeps the model honest, and a common sweet spot for a substantive answer is somewhere between six hundred and fifteen hundred characters.

Because the output is structured plain text, the same prompt is portable across every major Chinese model and works just as well on ChatGPT, Claude or Gemini, and the same framework — conclusion first, point-by-point, examples, summary — adapts cleanly to column articles and WeChat posts when you change the tone and length. Treat whatever the model returns as a first draft, never a finished post: verify the figures, swap in your own real experience, and tighten the wording to the platform's voice before you publish. And because the whole tool runs locally in your browser, you can iterate as much as you like — adjust one field, copy again, regenerate, compare — without anything you type ever leaving your device, being sent to a model, or being stored. Two or three rounds of that usually turn a generic reply into an answer that reads like you wrote it, with a clean, reusable prompt to keep at the end.