SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillspre-publish-review

pre-publish-review

General & other

View repository

Updated 10-15 21:11

About this skill

Pre-Publish Review

Instructions

Core goal: from the reader's perspective, judge whether this article is worth publishing. It is not "publish once it is written," but "publish only if it is worth publishing."

1. Trigger Timing

This skill should be used in the following situations:

  • After the editor-in-chief has completed review and before publication
  • When the user says "help me check whether this can be published"
  • When the user says "review this article"
  • The final step of the complete writing workflow

2. Pre-Publish 5-Question Review ⭐⭐⭐⭐⭐

The following 5 questions must be answered item by item, and publication is recommended only if all pass:

Question 1: Title question

If you scrolled past this title, would you stop and click on it?

Hard indicator checks (must be performed):

  • Does it include specific numbers/names/place names?
  • Does it avoid AI cliché words such as "in-depth interpretation," "comprehensive analysis," and "worth pondering"?
  • Is the word count within the 15-25 character range?
  • Does it use at least 1 viral formula (numbers/contrast/identity/suspense/emotion)?

If 2 of the above 4 items fail → automatically deduct 1 star

Evaluation dimensions:

  • Does it have suspense/conflict/specific numbers?
  • Does it use a viral formula?
  • Does it avoid AI cliché words (in-depth interpretation, comprehensive analysis, etc.)?
  • Is it within the 15-25 character range?

Rating: ⭐⭐⭐⭐⭐ (1-5 stars)

Question 2: Opening question

After reading the first 3 lines, would you want to keep reading?

Evaluation dimensions:

  • Does the opening use a hook type (scene immersion/identity binding/suspense opening/emotional resonance)?
  • Do the first 50 characters make readers feel "this is about me"?
  • Does it avoid clichés such as "in today's society" and "with the development of..."?

Rating: ⭐⭐⭐⭐⭐ (1-5 stars)

Question 3: Identification question

Does the full text have at least 1 place where readers feel "this is about me"?

Evaluation dimensions:

  • Are there specific scene descriptions that let readers identify with it?
  • Does it say on readers' behalf what they want to say but dare not say?
  • Is the intensity distribution of empathy points reasonable?

Rating: ⭐⭐⭐⭐⭐ (1-5 stars)

Question 4: Way-out question

After reading, do readers know what they should do?

Evaluation dimensions:

  • Are specific, actionable suggestions provided?
  • Are the suggestions practical rather than empty and unrealistic?
  • Does it avoid preaching and chicken soup?

Rating: ⭐⭐⭐⭐⭐ (1-5 stars)

Question 5: Sharing question

Would readers be willing to forward this article to friends? Why?

Evaluation dimensions:

  • Does the article have "social currency" value (making readers appear smart/tasteful/opinionated)?
  • Are there quotable lines that can be screenshotted and shared?
  • Can it spark discussion or resonance?

Rating: ⭐⭐⭐⭐⭐ (1-5 stars)

3. Review Result Output Format

📋 发布前评审报告

═══════════════════════════════════════════════════

【文章信息】
- 标题：[标题]
- 字数：[XXXX] 字
- 风格：[风格名称]

═══════════════════════════════════════════════════

【发布前5问评审】

1️⃣ 标题问：刷到这个标题，你会停下来点开吗？
   评分：⭐⭐⭐⭐☆（4/5）
   分析：[具体分析]
   建议：[如有改进空间]

2️⃣ 开头问：读完前3行，你会想继续往下看吗？
   评分：⭐⭐⭐⭐⭐（5/5）
   分析：[具体分析]
   建议：[如有改进空间]

3️⃣ 认同问：全文有没有让读者觉得"说的就是我"的地方？
   评分：⭐⭐⭐⭐☆（4/5）
   分析：[具体分析]
   建议：[如有改进空间]

4️⃣ 出路问：读者看完知道自己该怎么做吗？
   评分：⭐⭐⭐⭐⭐（5/5）
   分析：[具体分析]
   建议：[如有改进空间]

5️⃣ 分享问：读者会愿意转发吗？
   评分：⭐⭐⭐⭐☆（4/5）
   分析：[具体分析]
   建议：[如有改进空间]

═══════════════════════════════════════════════════

【综合评分】
总分：22/25 分

【评审结论】
✅ 建议发布 / ⚠️ 建议修改后发布 / ❌ 不建议发布

【修改建议】（如有）
1. [建议1]
2. [建议2]

═══════════════════════════════════════════════════

🔴 【红队评审】强制挑刺环节

**目标：站在最挑剔读者的角度，找出文章的潜在问题**

必须逐项检查以下5个问题（即使文章很好也要找）：

1. **标题致命伤**：
   - 哪个词会让读者觉得"又是标题党"？
   - 对不了解背景的人是否有吸引力？
   - 是否明确了目标人群？
   分析：[具体指出问题]

2. **开头劝退点**：
   - 前50字有没有让人想关闭的表述？
   - 是否有粗俗/冒犯性表达？
   - 是否假设了读者的背景知识？
   分析：[具体指出问题]

3. **AI味道残留**：
   - 哪句话一看就是AI写的？
   - 是否有"思考题"、"综上所述"等套路？
   - 是否有过度排比、小标题病？
   - **是否有"假名人案例"？**（如"薛冰雷，35岁，部门经理..."）
   
   **假名人案例检测：**
   - 查找"陌生人名 + 年龄 + 职业"的模式
   - 如果不是名人，建议改为"我朋友"、"我见过"、"老王"等
   - 例外：刘润、马云等真正的名人可以直接用名字
   
   分析：[具体指出问题]

4. **逻辑漏洞**：
   - 哪个论证经不起推敲？
   - 是否只提问题不给出路？
   - 数据/案例是否可靠？
   分析：[具体指出问题]

5. **情绪过载**：
   - 哪里用力过猛，让人觉得刻意煽情？
   - 是否全文负面情绪，缺少希望？
   - 是否有说教或鸡汤味道？
   分析：[具体指出问题]

6. **段落间重复**（⚠️ 新增重点检查）：
   - 是否有相同句式在不同段落重复？（如"这说明什么？说明..."）
   - 是否有相同结论在不同段落重复？（如"这是个结构性问题"）
   - 是否有相同数据被重复引用并推导相同结论？
   - 是否有相同的论证逻辑在不同段落重复使用？
   
   **检查方法：**
   - 逐段对比，标记相似度高的句子
   - 特别注意"强调句式"的重复（如"XX说明什么？说明YY"）
   - 检查同一个数据/案例是否被多次用于相同推论
   
   分析：[具体列出重复内容，标注段落位置]

7. **传播力缺失**（⚠️ 新增重点检查）：
   - 全文是否缺少情绪爆发点？（感叹号<3个，无脏话）
   - 全文是否缺少金句？（无可截图分享的句子）
   - 案例是否缺少画面感？（只陈述事实，无对话/动作/场景）
   - 是否全是"说理"，缺少"撕裂感"？
   - 结尾是否是"思考题"？（AI最大败笔）
   
   **检查方法：**
   - 统计感叹号、脏话、短句数量
   - 扫描全文，标记可截图的金句（至少2个）
   - 检查案例是否有具体细节（对话/动作/场景）
   - 检查结尾是否是"思考题"
   
   分析：[具体列出问题，给出改写建议]

【红队扣分】
- 如果发现"AI套路结尾"（如"思考题"） → -2分
- 如果发现"缺少出路"（只焦虑不给方案） → -2分
- 如果发现"粗俗表达"可能劝退读者 → -1分
- 如果发现"逻辑严重漏洞" → -2分
- 如果发现"段落间明显重复"（相同句式/结论重复2次以上） → -2分
- 如果发现"全文无金句"（无可分享内容） → -2分
- 如果发现"情绪过于平淡"（感叹号<3个且无脏话） → -1分

【红队结论】
- 如果7项中有5项以上是"无明显问题" → 通过红队评审
- 如果有3项以上是"严重问题" → 建议重写

═══════════════════════════════════════════════════

【最终评分】
常规评分：XX/25
红队扣分：-X
**最终得分：XX/25**

═══════════════════════════════════════════════════

⚠️ **重要：发布前评审完成后，必须执行【今日头条读者模拟】**

无论评审结果如何，都必须进行读者模拟测试。这是发布前的最后一道关卡，用于验证文章是否真的能吸引普通读者。

**下一步操作：**
现在请说"继续"，我将立即启动【今日头条读者模拟】，模拟真实读者的阅读体验。

═══════════════════════════════════════════════════

4. Review Criteria

Total Score Range Conclusion Action
23-25 points ✅ Strongly recommend publishing Publish directly
20-22 points ✅ Recommend publishing Can be published, but consider minor adjustments
15-19 points ⚠️ Recommend publishing after revision Needs revision before review
10-14 points ❌ Do not recommend publishing Needs major revision
<10 points ❌ Strongly do not recommend publishing Recommend rewriting

Low score warning for individual items:

  • If any item is below 3 stars → that item needs focused revision
  • If "title question" or "opening question" is below 3 stars → it must be revised before publishing

5. Quick Review Mode

If the user only needs a quick judgment:

📋 发布前快速评审：

1️⃣ 标题：⭐⭐⭐⭐☆
2️⃣ 开头：⭐⭐⭐⭐⭐
3️⃣ 认同：⭐⭐⭐⭐☆
4️⃣ 出路：⭐⭐⭐⭐⭐
5️⃣ 分享：⭐⭐⭐⭐☆

总分：22/25 ✅ 建议发布

⚠️ 注意：标题可以更吸引人，考虑使用悬念反转公式

Examples

User: Help me review whether this article can be published.

AI：(After reading the article, output the review report)

📋 发布前评审报告

═══════════════════════════════════════════════════

【文章信息】
- 标题：你那张"总监"名片，不过是租来的座位
- 字数：2165 字
- 风格：九边风

═══════════════════════════════════════════════════

【发布前5问评审】

1️⃣ 标题问：刷到这个标题，你会停下来点开吗？
   评分：⭐⭐⭐⭐⭐（5/5）
   分析：使用了"对立冲突型"公式，"总监名片"vs"租来的座位"形成强烈反差，能引发职场人的好奇心。
   建议：无需修改

2️⃣ 开头问：读完前3行，你会想继续往下看吗？
   评分：⭐⭐⭐⭐⭐（5/5）
   分析：开头"见过太多这样的人"直接切入，第二段用具体场景描写（名片、朋友圈）制造代入感，第三段"直到有一天，公司一纸通知下来"制造悬念。
   建议：无需修改

3️⃣ 认同问：全文有没有让读者觉得"说的就是我"的地方？
   评分：⭐⭐⭐⭐☆（4/5）
   分析："饭局上被问你是干啥的"这个场景非常具体，能引发认同。中间的心理分析部分也有共鸣。
   建议：可以增加1-2个更具体的职场场景

4️⃣ 出路问：读者看完知道自己该怎么做吗？
   评分：⭐⭐⭐⭐⭐（5/5）
   分析：文章给出了3条具体建议：不把所有时间投在工作上、做跟工作无关但有成就感的事、学会跟自己相处。建议务实，不是假大空。
   建议：无需修改

5️⃣ 分享问：读者会愿意转发吗？
   评分：⭐⭐⭐⭐⭐（5/5）
   分析："职位只是张租来的座位，你得有自己的椅子"这句话是金句，可以截图分享。文章观点有态度，适合职场人转发表达认同。
   建议：无需修改

═══════════════════════════════════════════════════

【综合评分】
总分：24/25 分

【评审结论】
✅ 强烈推荐发布

【修改建议】
可选优化：在"自我价值外化"部分增加1个更具体的职场场景描写

═══════════════════════════════════════════════════

Best Practices

  • Prioritize the reader's perspective: When reviewing, become the target reader, not the author
  • Strictly implement the 5 questions: Every question must be answered seriously, not perfunctorily
  • Mandatory red team review: Even if the article is very good, red team nitpicking must still be carried out (now upgraded to 7 checks) to prevent inflated scores
  • Paragraph-by-paragraph comparison check: Pay special attention to repetition between paragraphs (same sentence patterns, same conclusions, same data inferences)
  • Hard indicators first: First check hard indicators (such as title length, AI cliché words), then make subjective evaluations
  • Give specific suggestions: If any item scores low, specific revision suggestions must be given
  • Dare to say no: If the article really is not good, dare to say "do not recommend publishing"
  • Golden sentence check: A good article must have golden sentences that can be screenshotted and shared (at least 2)
  • Must check for a way out: If the article only raises problems without giving a way out, points must be deducted
  • Spreadability check: Check whether the article has emotional outburst points, golden sentences, and visual imagery

Version History

  • v2.3.0 (2026-01-21): Mandatory implementation of Jinri Toutiao reader simulation, ensuring that reader testing must be completed after the pre-publication review is finished, no longer an optional step.
  • v2.2.0 (2026-01-19): Added the 7th red team review item, the "lack of spreadability" check, to detect issues such as articles lacking emotion, golden sentences, and visual imagery.
  • v2.1.0 (2026-01-19): Added the 6th red team review item, the "inter-paragraph repetition" check, and added the "fake celebrity case" check (an AI-flavor residual item) to solve common AI problems.
  • v2.0.0 (2026-01-16): Major upgrade, added the red team review mechanism and hard indicator checks to prevent inflated scores.
  • v1.0.0 (2026-01-03): Initial version, supports pre-publication 5-question review and quick review mode.