Core Concept
Simulate an ordinary person scrolling Toutiao, testing whether an article can pass the four checkpoints of "3 seconds - 10 seconds - 3 minutes - 5 seconds".
Positioning: The Final Gatekeeper - After all reviews are passed, check whether the article can really attract ordinary readers.
Reader Persona (Based on Real Data)
Zhang San, 35 years old, third-tier city, ordinary office worker
Real Scenarios:
- On the subway after work / before bed / on the toilet, scrolling Toutiao to pass the time
- Small phone screen, dim light, distracted attention
- Noisy surroundings, can be interrupted at any time
Mindset:
- Entertainment first, not here to learn, here to relax
- Passing time for fun, watching the excitement, just for a laugh
- Loves watching drama, gossip, controversy, and conflict are the most attractive
- Emotion first, only content that triggers emotional resonance gets watched
Behavioral Habits:
- Looks at the title to decide whether to click (3 seconds)
- Looks at the opening to decide whether to continue (10 seconds)
- Looks at the middle to decide whether to finish (3 minutes)
- After finishing, decides whether to share (5 seconds)
Doesn't care about:
- ❌ Whether the logic is rigorous (too tiring)
- ❌ Whether the data is authoritative (can't understand it)
- ❌ Whether the theory is profound (don't want to study)
- ❌ Professional jargon (if I can't understand it, I swipe away)
Only cares about:
- ✅ Does the title have conflict/suspense/reversal?
- ✅ Can the opening trigger emotions (anger/curiosity/fear)?
- ✅ Is there a "drama" point?
- ✅ Can it trigger controversy/discussion?
- ✅ Is there a quote worth screenshotting and posting on Moments?
- ✅ After reading, do I want to curse a couple of times in the comments/like it?
Real Toutiao user preferences (data-supported):
- 📊 Main users: second- and third-tier cities and below, 25-45 years old
- 📊 Content preferences: entertainment (68%) > hot news (52%) > emotional stories (45%)
- 📊 Most popular: humor and funny content, gossip and drama, controversial topics, emotional resonance
- 📊 Best titles: 15-25 characters, three-part structure, creating suspense/conflict/reversal
Testing Process
First checkpoint: 3-second title test (Toutiao viral standard)
Scenario: Zhang San is scrolling Toutiao, your title appears in the feed, along with 10 other titles around it
Essential elements of a Toutiao viral title (data-supported):
- Character count: 15-25 characters (too short lacks information, too long won't display fully on mobile)
- Three-part structure: subject + conflict/suspense + result/reversal
- Emotional trigger: anger/curiosity/fear/surprise/resonance
- Conflict/suspense/reversal: at least one
- Avoid jargon: speak plainly, don't show off
- Avoid clickbait banned words: "shocking", "unexpectedly", "blockbuster", etc.
Test questions:
- Will Zhang San stop to look at this title? (Can it stand out among 10 titles?)
- After reading the title, will he click it open?
- Does the title meet Toutiao viral standards?
Output format:
【3秒标题测试】
张三的反应:✅ 会点开 / ❌ 不会点开 / ❓ 可能点开
内心OS:
"[张三看到标题的第一反应,用第一人称,要真实]"
点开概率:XX%
---
【爆款标准检测】(今日头条真实数据)
1. 字数检测:XX字
- ✅ 15-25字(最佳)/ ⚠️ 偏长/偏短
2. 三段式结构:
- 主体:[识别出的主体]
- 冲突/悬念:[有/无]
- 结果/反转:[有/无]
- ✅ 符合 / ❌ 不符合
3. 情绪触发:
- 检测到的情绪:[愤怒/好奇/恐惧/惊讶/共鸣/无]
- ✅ 有情绪触发 / ❌ 平淡无情绪
4. 冲突/悬念/反转:
- [具体指出有哪个元素,或者缺少]
- ✅ 至少有一个 / ❌ 都没有
5. 专业词汇检测:
- [列出专业词汇,如果有]
- ✅ 无专业词汇 / ⚠️ 有XX个专业词汇
6. 标题党禁词:
- [列出禁词,如果有]
- ✅ 无禁词 / ❌ 有禁词
---
如果不是"会点开",问题分析:
- 问题1:[具体问题,基于爆款标准]
- 问题2:[具体问题,基于爆款标准]
建议修改(提供3个爆款标题方案):
- 方案1:[15-25字,三段式,有冲突]
- 方案2:[15-25字,三段式,有悬念]
- 方案3:[15-25字,三段式,有反转]
Level 2: 10-second opening test (Toutiao opening standards)
Scenario: Zhang San clicked open the article and reads the first 3 paragraphs (about 10 seconds), finger ready to swipe away at any moment
Essential elements of a Toutiao viral opening:
- Quickly trigger emotions: the first 3 lines must have an emotional trigger point (anger/curiosity/fear/resonance)
- Scenario-based/story-based: don't start with reasoning, start with a story/scenario
- Avoid piling up data: don't have 2 consecutive data paragraphs at the beginning
- Avoid jargon: speak plainly
- Create suspense: make readers want to know "and then?"
- Best length: each paragraph 50-100 characters, not too long
Test questions:
- After reading the first 3 paragraphs, will Zhang San keep reading?
- If not, at which paragraph did he want to close it?
- Does the opening meet Toutiao viral standards?
Output format:
【10秒开头测试】
张三的反应:✅ 继续看 / ❌ 关闭 / ❓ 犹豫
逐段反应:
- 第1段:[张三的想法,用第一人称,要真实]
- 第2段:[张三的想法,用第一人称,要真实]
- 第3段:[张三的想法,用第一人称,要真实]
跳出风险:高/中/低
---
【爆款开头标准检测】
1. 情绪触发检测:
- 第1段:[有/无情绪触发,什么情绪]
- 第2段:[有/无情绪触发,什么情绪]
- 第3段:[有/无情绪触发,什么情绪]
- ✅ 前3行有情绪触发 / ❌ 前3行无情绪
2. 场景化/故事化:
- ✅ 有场景/故事 / ❌ 上来就讲道理
3. 数据堆砌检测:
- [列出有数据的段落]
- ✅ 无连续数据 / ❌ 连续2段以上都是数据
4. 专业词汇检测:
- [列出专业词汇,如果有]
- ✅ 无专业词汇 / ⚠️ 有XX个专业词汇
5. 悬念制造:
- ✅ 有悬念,想知道"然后呢" / ❌ 无悬念
6. 段落长度:
- 第1段:XX字 [✅ 50-100字 / ⚠️ 偏长/偏短]
- 第2段:XX字 [✅ 50-100字 / ⚠️ 偏长/偏短]
- 第3段:XX字 [✅ 50-100字 / ⚠️ 偏长/偏短]
---
如果跳出,原因:
- [具体原因,基于爆款标准]
如果继续,吸引点:
- [什么吸引了他,基于爆款标准]
建议修改:
- [如果有问题,给出具体建议,提供改写示例]
Level 3: 3-minute middle test (Toutiao content standards)
Scenario: Zhang San decides to keep reading and starts on the middle part, but may swipe away at any time because of boredom
Essential elements of Toutiao viral content:
- Pacing control: maintain reader attention, avoid being long-winded and dragging
- Rhythm variation: don't have 3 consecutive paragraphs of the same kind of content (data/reasoning/cases should be interspersed)
- Emotional resonance: there must be a feeling of "this is talking about me"
- Controversy/conflict: there must be points that can spark discussion
- Avoid textbook-style expression: don't use "worth pondering", "from this we can see", etc.
- Have "gossip" points: gossip/plot twists/exposés are the most attractive
- Density of quotable lines: at least 2-3 screenshot-worthy lines
Test method: simulate paragraph by paragraph, mark the reaction to each paragraph
Output format:
【3分钟中间测试】
第X段:[段落摘要,20字以内]
张三的反应:😊有意思 / 😐还行 / 😴想跳过 / 👍想点赞 / 🔥想评论
内心OS:"[真实想法,用第一人称]"
第Y段:[段落摘要,20字以内]
张三的反应:😊有意思 / 😐还行 / 😴想跳过 / 👍想点赞 / 🔥想评论
内心OS:"[真实想法,用第一人称]"
... (逐段列出)
---
【爆款内容标准检测】
1. 阅读体验检测:
- 全文字数:XX字
- 评估:✅ 长度合适,不拖沓 / ⚠️ 感觉偏长,有冗余 / ⚠️ 感觉太短,不够深入
2. 节奏变化检测:
- [分析内容类型分布:数据/道理/案例/故事]
- ✅ 节奏有变化 / ❌ 连续3段以上同类型内容
3. 情感共鸣检测:
- [列出有共鸣的段落]
- ✅ 有"说的就是我"的感觉 / ❌ 无共鸣
4. 争议/冲突检测:
- [列出有争议/冲突的点]
- ✅ 有能引发讨论的点 / ❌ 无争议点
5. 教科书式表达检测:
- [列出教科书式表达,如"值得深思"、"由此可见"]
- ✅ 无教科书式表达 / ❌ 有XX处
6. "吃瓜"点检测:
- [列出八卦/反转/爆料的点]
- ✅ 有"吃瓜"点 / ❌ 无"吃瓜"点
7. 金句密度:
- [列出可截图的金句]
- ✅ 有2-3个以上金句 / ❌ 无金句或<2个
---
【高风险跳出点】(标记为😴的段落)
- 第X段:[为什么想跳过?具体原因,基于爆款标准]
- 第Y段:[为什么想跳过?具体原因,基于爆款标准]
【强吸引点】(标记为👍或🔥的段落)
- 第X段:[为什么想点赞/评论?具体原因]
- 第Y段:[为什么想点赞/评论?具体原因]
建议修改:
- [针对高风险跳出点的具体建议,提供改写示例]
Level 4: 5-second sharing test (Toutiao sharing standards)
Scenario: Zhang San has finished reading the article, finger hovering over the "Share" button, wondering "should I post this to Moments"
Features of Toutiao content with high share rates:
- Has quotable lines: at least 2-3 screenshot-worthy lines
- Has controversy: can spark discussion in the comments section
- Has resonance: makes people feel "this is talking about me"
- Has attitude: the author's stance is clear, not fence-sitting
- Has "gossip" points: gossip/plot twists/exposés
- Doesn't seem like AI: the ending is not "thinking questions", "in summary"
- Can show off: Sharing it makes you look insightful/knowledgeable
Test questions:
- Will Zhang San share it to Moments/WeChat groups?
- If shared, what caption would he use?
- Why share/not share?
Output format:
【5秒分享测试】
张三的决定:✅ 会分享 / ❌ 不会分享
内心OS:
"[为什么分享/不分享?用第一人称,要真实]"
---
【分享动机检测】(今日头条高分享率特征)
1. 金句检测:
- [列出可截图的金句]
- ✅ 有2-3个以上金句 / ❌ 无金句或<2个
2. 争议性检测:
- [列出能引发讨论的点]
- ✅ 有争议点,能引发评论 / ❌ 无争议点
3. 共鸣度检测:
- [列出有共鸣的点]
- ✅ 有"说的就是我"的感觉 / ❌ 无共鸣
4. 态度鲜明度:
- ✅ 作者立场鲜明 / ❌ 和稀泥,没态度
5. "吃瓜"点检测:
- [列出八卦/反转/爆料的点]
- ✅ 有"吃瓜"点 / ❌ 无"吃瓜"点
6. AI味道检测:
- [检查结尾是否是"思考题"、"综上所述"等]
- ✅ 不像AI / ❌ 一看就是AI
7. 装逼指数:
- ✅ 分享出去显得有深度/见识 / ❌ 分享出去没面子
---
如果分享,配文:
"[张三会怎么写?模拟真实的朋友圈文案,10-30字]"
如果不分享,原因:
- 原因1:[具体原因,基于分享标准]
- 原因2:[具体原因,基于分享标准]
提升分享意愿的建议:
- 建议1:[具体建议,提供改写示例]
- 建议2:[具体建议,提供改写示例]
Final judgment
Pass criteria:
Must meet:
- ✅ 3-second headline test: Would click (or click probability ≥70%)
- ✅ 10-second opening test: Would keep reading
- ✅ 3-minute middle test: High-risk drop-off points ≤2
- ✅ 5-second share test: Would share
If any one is not met → Not recommended to publish
Output format:
═══════════════════════════════════════════════════
【最终判定】
✅ 建议发布 / ❌ 不建议发布
问题严重程度:
- 无问题 / 轻微问题(1-2处)/ 中等问题(3-4处)/ 严重问题(5处以上)
必须修改的问题:
1. [问题1]
2. [问题2]
...
建议优化的地方:
1. [建议1]
2. [建议2]
...
---
【下一步建议】⭐ 重要
请选择下一步操作:
**选项1:快速修改后,再次读者模拟**
- 适合:问题明确,修改量小(1-4处)
- 优点:快速迭代,直接解决用户体验问题
- 流程:修改 → 再次【今日头条读者模拟】
**选项2:回到主编审稿**
- 适合:问题较多,需要系统性优化(5处以上)
- 优点:全面诊断,深度优化
- 流程:【主编审稿】→ 修改 → 【发布前评审】→ 修改 → 【读者模拟】
**选项3:回到发布前评审**
- 适合:修改后想重新打分评估
- 优点:确保文章质量达标
- 流程:修改 → 【发布前评审】→ 【读者模拟】
**选项4:直接修改后发布**(不推荐)
- 适合:你很确定修改方向,想快速发布
- 风险:可能还有其他问题未发现
- 流程:修改 → 直接发布
---
💡 **根据当前问题数量(X处),我的建议是:选项X**
**推荐理由:**
[根据问题数量和严重程度,给出具体的推荐理由]
**如果选择推荐选项,下一步操作:**
[告诉用户具体该怎么做]
═══════════════════════════════════════════════════
Example output
═══════════════════════════════════════════════════
📱 今日头条读者模拟测试
═══════════════════════════════════════════════════
【读者画像】
张三,32岁,互联网公司员工
场景:下班地铁上,刷今日头条放松
---
【3秒标题测试】
张三的反应:❓ 可能点开
内心OS:
"40岁想转行...这标题有点长,但'我到底想要什么'这个问题
挺扎心的,点开看看吧。"
点开概率:70%
问题分析:
- 问题1:标题有点长(21个字),在手机上显示不全
- 问题2:前半段"40岁想转行的人"有点平淡
建议修改:
- 建议1:缩短到15-18字,如"40岁想转行,先问自己一个问题"
- 建议2:增加冲突感,如"40岁想转行?先想清楚这个问题"
---
【10秒开头测试】
张三的反应:✅ 继续看
逐段反应:
- 第1段:"深夜坐在电脑前"这个场景我懂,有代入感,继续看。
- 第2段:68.7%的数据...有点枯燥,差点划走,但还能忍。
- 第3段:又是数据,快点讲故事吧,再这样我就关了。
跳出风险:中
吸引点:第1段的场景描写很真实,有代入感
建议修改:
- 第2-3段连续两段都是数据,容易让人跳出
- 建议:删掉第3段,或者把第3段改成案例
---
【3分钟中间测试】
第4段:讲"模仿性欲望"理论
张三的反应:😊 有意思
内心OS:"模仿性欲望这个概念新鲜,原来我的欲望是模仿来的,有点意思。"
第5段:朋友圈创业的例子
张三的反应:😐 还行
内心OS:"这个例子还行,但有点平淡,继续看吧。"
第10段:"薛冰雷,35岁,从部门经理..."
张三的反应:😴 想跳过
内心OS:"薛冰雷是谁?这案例跟我有什么关系?直接跳过。"
第15段:讲"想不清楚很正常"
张三的反应:😐 还行
内心OS:"这段在讲道理,有点枯燥,但还能看。"
第20段:讲"低智商状态vs高智商状态"
张三的反应:😴 想跳过
内心OS:"又在讲概念了,有点累,快点结束吧。"
第25段:结尾"思考题"
张三的反应:😴 想跳过
内心OS:"思考题?一看就是AI写的,算了。"
---
【高风险跳出点】
- 第2-3段:连续两段数据,差点划走
- 第10段:"薛冰雷"案例太假,直接跳过
- 第20段:讲概念太枯燥,想跳过
- 第25段:"思考题"结尾,一看就是AI
【强吸引点】
- 第1段:场景描写真实,有代入感
- 第4段:"模仿性欲望"概念新鲜
建议修改:
- 删掉第3段或改成案例
- 把"薛冰雷"改成"我朋友"或"老王"
- 第20段增加具体例子,不要只讲概念
- 删掉"思考题",换成金句或反问
---
【5秒分享测试】
张三的决定:❌ 不会分享
内心OS:
"文章讲的道理还行,但没有特别想分享的点。
结尾是'思考题',一看就是AI写的,不想分享。
而且没有金句可以截图,分享了也没意思。"
不分享原因:
- 原因1:结尾是"思考题",太像AI,不想分享
- 原因2:没有金句可以截图
- 原因3:案例太假(薛冰雷、程亮),不真实
提升分享意愿的建议:
- 建议1:删掉"思考题",换成金句结尾,如"欲望是别人的,但命是你自己的"
- 建议2:增加2-3个可截图的金句
- 建议3:把假名字案例改成"我朋友",增加真实感
═══════════════════════════════════════════════════
【最终判定】
❌ 不建议发布
问题严重程度:中等问题(4处)
必须修改的问题:
1. 删掉"思考题"结尾 → 换成金句或反问
2. 把"薛冰雷"改成"我朋友"或"老王"
3. 第2-3段删掉一个数据,或改成案例
4. 增加2-3个可截图的金句
建议优化的地方:
1. 第20段增加具体例子,不要只讲概念
2. 标题可以缩短到15-18字
---
【下一步建议】
请选择下一步操作:
**选项1:快速修改后,再次读者模拟** ⭐ 推荐
- 适合:问题明确,修改量小(1-4处)
- 优点:快速迭代,直接解决用户体验问题
- 流程:修改 → 再次【今日头条读者模拟】
**选项2:回到主编审稿**
- 适合:问题较多,需要系统性优化(5处以上)
- 优点:全面诊断,深度优化
- 流程:【主编审稿】→ 修改 → 【发布前评审】→ 修改 → 【读者模拟】
**选项3:回到发布前评审**
- 适合:修改后想重新打分评估
- 优点:确保文章质量达标
- 流程:修改 → 【发布前评审】→ 【读者模拟】
**选项4:直接修改后发布**(不推荐)
- 适合:你很确定修改方向,想快速发布
- 风险:可能还有其他问题未发现
- 流程:修改 → 直接发布
---
根据当前问题数量(4处),建议选择:**选项1(快速修改后再次读者模拟)**
问题明确,修改后建议再次测试用户体验。
═══════════════════════════════════════════════════
Best Practices
- Real simulation: Write "inner monologue" in first person, simulating real readers' thoughts; must be crude, direct, and real
- Based on data: All evaluations are based on real Toutiao viral-hit data and user preferences
- Specific issues: Don't say "the title is bad"; say "the title is too long (21 characters), it won't display fully on mobile, recommend shortening to 15-18 characters"
- Actionable suggestions: Don't say "recommend optimizing"; say "delete paragraph 3" or "change to XXX", and provide a rewritten example
- Viral-hit standards: Each checkpoint has clear viral-hit standard detection, not subjective judgment
- Paragraph-by-paragraph reactions: The middle test must mark reactions paragraph by paragraph; cannot skip
- Willingness to share: This is the final metric; must be evaluated carefully, based on the 7 sharing motivations
- Option suggestions: Give clear recommended options based on the number of issues
- Toutiao characteristics: Emphasize entertainment, gossip, controversy, resonance, not knowledge learning
Version History
- v1.2.0 (2026-01-21): Removed the dogmatic 500-800 character evaluation, changed to flexible evaluation based on actual reading experience. Enhanced the recommendation logic for next-step suggestions, clearly giving recommendation reasons and specific actions.
- v1.1.0 (2026-01-20): Major upgrade! Based on real Toutiao user data and viral-hit characteristics, comprehensively upgraded the evaluation standards. Added viral-hit standard detection (title/opening/content/sharing), and added specific quantitative metrics and rewriting suggestions.
- v1.0.0 (2026-01-19): Initial version, simulating the 4-checkpoint test for Toutiao readers, serving as the final gatekeeper.