Recommendations were meant to be the finishing touch, but now they’ve become a technical task. The prompt template shared by Baoyu breaks down the process of writing recommendations into quantifiable steps—from hook design to integrating punchlines—making it resemble a programmer’s function call manual.


This engineering mindset does solve basic problems: it avoids AI-generated fluff like "This article is amazing." But as pointed out in the comments, when faced with articles with distinct themes, the model still reveals its tendency to "pretend to understand." After all, no matter how refined the prompts are, they can’t teach AI to truly grasp the magical realism of *One Hundred Years of Solitude*.


The irony is that while we’re discussing how to make AI recommendations more "human-like," human editors’ recommendations are becoming more like AI—Zhihu’s randomly generated intros perfectly demonstrate what "correct but meaningless" sounds like. Perhaps the real solution lies in the middle ground: using technical frameworks to curb AI’s nonsense while preserving the last 10% of human aesthetic judgment.


Next time you see recommendations like "thought-provoking" or "deeply resonant," try guessing whether the author field should credit GPT-5 or an exhausted editor. This guessing game itself is the digital age’s most humorous dark parable.