大多数”我该研究什么”的 Prompt 输出含糊的话题云。可辩护的研究空白要更锐利：点出”没做过什么”、“为什么重要”、“为什么你能做”。下面 15 个 Prompt 把你从一堆文献带到一段能在 committee 面前站住的空白陈述。
适合哪些场景
在写 proposal 的研究生、博士候选人、postdoc、做毕业研究的高年级本科生。
什么时候不建议这样写 Prompt
不要用来写大众科普——它需要 20+ 篇真实文献作为输入。也不要在你完全没读过候选文献时用：AI 替你读不了。
Prompt 结构公式
研究空白 Prompt 一定要带这六个要素：
- 角色：AI 扮演谁——研究导师、同行评议人、考试教练、辩论对手、图书馆员。
- 上下文：水平、学科、deadline、论文数量、引用风格、课程或项目。
- 目标：一个具体交付物——12 道题、1 页文献矩阵、5 条反论、4 周复习计划。
- 限制：字数、深度、允许的来源类型、跳过什么、绝不主张什么。
- 输出格式：编号清单、表格、JSON 或分级块（E / M / H），能粘到 Notion / Anki / Word。
- 示例 / 信号：1-2 段参考或反例（“不要像维基那样讲”）。
这套 Prompt 适合用在哪
- 博士 proposal 的”立项”段
- 硕士论文选题收窄
- 基金”未知部分”段落
- 注释性书目 → 空白综合
- discussion 段的”空白定位”
15 个可直接复制的 Prompt 模板
1. 文献集 → 未回答问题抽取
当你手上已有 15-30 篇 abstract 时用。
You are a research librarian in {field}. Below are {N} abstracts on {topic}. Cluster the recurring research questions, then list 5 questions that appear in the framing of these abstracts but are NOT answered in any of them. For each gap: 1-sentence statement, the 2-3 papers that almost addressed it, why they fell short.
{paste abstracts}可替换变量： field, topic, N, abstracts
优化建议： 如果模型凭空造空白，追加：“Cite the paper id (first author + year) supporting every claim. If you cannot cite, omit the gap.”
2. 方法学空白审计
For the topic {topic}, list the dominant 3 methodologies used in published work over the last {N} years. Then name 2 underused methodologies and the kind of question each could newly answer. Output as a 3-row table: dominant method | what it cannot see | better-suited method.3. 人群 / 样本空白
For research on {topic}, list the populations / samples most commonly studied. Then name 3 underrepresented populations (geography, age, profession, condition) and a 1-paragraph rationale for studying each.4. 理论框架空白
In {field}, list the dominant 2-3 theoretical frames applied to {topic}. For each, name 1 phenomenon it explains well and 1 it cannot reach. Then propose a 4th frame that would extend coverage; cite one paper using that frame in a neighboring field.5. 时间 / 情境空白
Most studies on {topic} use data from {era / region}. Name 3 underrepresented time periods or contexts and why the existing findings might not transfer. End with 1 testable hypothesis per context.6. 结论冲突的空白
Below are abstracts that disagree on {claim}. Group them into "supports", "rejects", "mixed". Identify the moderator variables that might explain the disagreement and propose a study design that would resolve it.
{paste}7. 复制研究空白
For the topic {topic}, list the 3 most-cited findings that have never been replicated independently. For each: 2-line description, why replication matters, what a tight replication would look like ({N} subjects, design, primary measure).8. 操作化空白
For {construct}, list the 3 most common operational definitions used in {field}. Name what each captures and misses. Propose a 4th operationalization that would close a measurement gap; describe instrument and validation steps.9. 实务 — 学术 断层
For {topic}, list 3 questions practitioners ({clinicians / teachers / engineers / founders}) keep asking that the academic literature has not seriously studied. For each, 1 study design that would generate practitioner-ready evidence.10. “弱证据传言”空白
In {field}, name 3 claims about {topic} that "everyone repeats" but have only weak supporting evidence. For each: where it likely came from, the failure mode of the original evidence, what a stronger test would look like.11. 跨学科引入
A method / construct from {adjacent field} that has not been imported into research on {topic}: name 3 candidates. For each: 1-paragraph case for why it would advance the field, 1 paper that piloted it elsewhere.12. 空白 → proposal 段重写
Below is my draft gap paragraph. Rewrite it in the voice of a NIH / NSF proposal: name the unknown, name why it matters, name what would change if we knew. Stay under 200 words. Cite at least 3 sources by first author + year.
{paste draft}13. 评议人压力测试
My proposed gap: "{gap statement}". Pretend you are a skeptical reviewer in {field}. List the 5 strongest objections (already-done, not-important, not-feasible, wrong-method, wrong-population) and how I should preempt each in the proposal.14. 引用链空白
Below are the 10 most-cited papers on {topic}. List the questions each "calls for future work on" in its discussion section. Group the recurring calls; rank by how many papers raise them.
{paste discussion excerpts}15. 一段空白陈述（最终版）
Write a 150-180 word research-gap paragraph following this structure: (1) what we know — 2 sentences citing 3 sources; (2) what remains unknown — 1 sentence; (3) why this matters — 1 sentence linking to a practical or theoretical stake; (4) what the proposed study uniquely contributes — 1 sentence. Topic: {topic}. Field: {field}.容易踩的坑
- 把”话题”当”空白”（“应该多研究 X”）——评议人秒拒。
- 让 AI 凭空编空白——一定喂真实 abstract，不许它幻觉。
- 把”没人研究”等同于”值得研究”——空白还要重要。
- 忽略负结果——有时空白就是”那个效应其实不存在”。
- 选了你做不了的方法学空白（如 6 周内做不完的纵向研究）。
- 把空白写成”抱怨”而不是”问题”——评议人要可操作问题。
- 让 AI 引用不存在的论文——发表前必须核对标题、作者、DOI。
优化技巧
- 每次喂 10-15 篇 abstract，30 篇以上质量明显下降。
- 强制每句都带”FirstAuthor Year”引用，没有就删。
- 同一 Prompt 跑两次不同文献子集，若同一空白两次都出现就稳了。
- 空白识别后立刻做可行性筛查：把 timeline / budget / IRB 限制也喂给模型。
- 5 个候选空白后让模型按”评议人可辩护度”而非”新颖度”排序。
- 在 Notion / Obsidian 里维护一个 candidate gap 文件，每周回看。
- 看 top 10 高引论文的 discussion 段——一半空白就明写在”future work”里。
FAQ
- 研究空白和研究话题有什么区别？: 话题是领域（“AI 辅导”）。空白是该领域内一个具体未解问题（“AI 辅导是否在初中数学里缩小还是扩大成就差距”）。
- AI 自己能找到真实空白吗？: 不能。它能聚类、对比、改写你喂的内容。没有真实文献它会编。
- 跑空白识别 Prompt 之前需要多少文献？: 至少 15-20 篇 abstract，最好 30 篇。再少聚类就太薄。
- 方法学空白和话题空白哪个更好？: 方法学空白：同样问题、更好方法。话题空白：没人问过的问题。评议人通常更接受前者。
- 命名空白后还要继续用 AI 吗？: 要——做文献矩阵和段落草稿。但最终推理和引用清单必须人工核对。