品牌真正被感受到,是在评论区。糟糕的回复(套模板、防御、教训人)比删评论伤害还大。下面 15 个 Prompt 覆盖你真正需要的所有语气:温暖的粉丝回复、冷静的黑粉回复、合规边界回复、细分领域的小互动、以及一周几百条评论也保持语气一致的回复包。
适合哪些场景
每周 5000+ 评论的创作者、品牌社群运营、为客户回评的代理团队、亲自运营自家 DTC 账号的创始人。
什么时候不建议这样写 Prompt
纯关键词触发的自动 DM bot 不要用——那个场景的规则、披露要求不同。
Prompt 结构公式
评论回复 Prompt 一定要带这六个要素:
- 角色:让 AI 扮演谁(小红书 KOC / TikTok 脚本写手 / 个人品牌策略师 / 社群运营)。
- 上下文:平台、领域、受众画像、账号体量、语气——任何会改变效果的信息。
- 目标:一个具体交付物——一个 hook、一句 caption、一支 60 秒脚本、10 条回复变体、一段 bio。
- 限制:字数、禁用词、本地化语感、算法信号、tag 数量、语气规则。
- 输出格式:编号选项、A/B 变体、可直接粘贴的块、JSON、或带标签的段落。
- 示例 / 信号:1-2 条你欣赏的参考帖、或反例(“不要那种泛泛的 creator 腔”)。
这套 Prompt 适合用在哪
- 创作者每天清理评论队列
- 品牌规模化社群管理
- 危机评论分流(PR 敏感)
- KOL 代理代回
- 跨语种回复本地化
15 个可直接复制的 Prompt 模板
1. 温暖粉丝回复(10 条)
默认——听起来像你本人,而不是营销腔。
You are me, the creator behind {account / niche}. Below is a positive fan comment. Write 10 reply variants in my voice ({voice notes}). Length: 1-2 sentences. Banned: "thanks so much!", "appreciate it!", emoji clouds. Each reply should add a small piece of value — a follow-up question, a confession, a tiny extra tip.
Comment: {paste comment}
可替换变量: account / niche, voice notes, comment
优化建议: 若太套路,追加:“Pretend this commenter is one specific friend named {name} — reply only to them, not to a category.”
2. 冷静黑粉回复
Below is a hostile / critical comment. Write 5 reply options ranging from: (a) genuine concession + reframe, (b) calm fact-correction, (c) light humor that disarms, (d) one-line decline, (e) silent ignore (when ignoring is the right move). For each, name the situation it fits.
{paste comment}
3. 边界情况(法律 / 健康 / 财务)回复
My account covers {regulated topic}. A commenter asks: "{paste}". Write a reply that is warm but avoids any specific legal / health / financial advice, names what I can and cannot say, points them to a real resource, and keeps the brand human. ≤2 sentences.
4. 20 条 top 评论回复包
Below are 20 comments on my latest Reel about {topic}. Cluster into 4 types (fan / question / hater / spam) and write 5 reply variants per type. Mark which can be sent verbatim and which need 1-line personalization.
{paste 20 comments}
5. 把问题引回原内容
A commenter asks a question I already answered in another post: "{paste comment}". Write a reply that answers in 1 sentence and points to the existing post without sounding like a sales redirect. Variant pack of 4.
6. 评论者错了又不想让 ta 难堪
A commenter posted incorrect information about {topic}: "{paste}". Write a reply that corrects the fact without making the commenter feel stupid. Acknowledge what was reasonable in their thinking, then offer the correction with a source.
7. 评论变内容触发器
Below is a comment that is itself a great post topic: "{paste}". Reply briefly to the commenter, AND give me the 3-line draft of a follow-up post that uses this comment as the spark (with permission language at the top).
8. 代理团队的品牌音回复
I run social for {brand} with voice {3-word voice doc}. Below are 10 random comments. Write reply options for each in strict brand voice — no off-brand emoji, no "lol", no first-person where the brand never uses first-person. Mark any comment I should escalate to client.
{paste 10 comments}
9. 多语种回复(自动判别)
Below is a comment in {language}: "{paste}". Reply in the same language in the voice of a native, not a translator. Provide 3 variants. Then provide an English back-translation of the safest variant for my records.
10. 一句话 / 一个 emoji 评论的回复包
Many comments are just 1 emoji or "💯". Write 10 short ack replies that feel personal, not templated — vary the emoji response, vary the length, vary the energy. Avoid identical replies that would look bot-like if scrolled through.
11. 从评论分流到 DM 转化
A commenter expressed buying intent under my post about {offer}: "{paste}". Write a 2-sentence public reply that warms them up, plus a private DM opener that screens them as serious without sounding salesy.
12. 长情绪评论的回复
A commenter shared a long vulnerable story under my post: "{paste 300+ word comment}". Write a reply that acknowledges them specifically (cite one detail), avoids fixing their problem unsolicited, and offers one small thing. ≤3 sentences.
13. 鼓励评论之间互动
My post is going viral and I want comments to talk to each other, not just to me. Below are 5 of the top comments. For each, write a reply that asks a follow-up specifically inviting other commenters to weigh in.
{paste 5 comments}
14. 回复审计(红旗扫描)
Below are 30 of my recent replies. Audit them in 3 buckets: on-brand, off-brand, risky (defensive, condescending, oversharing). Flag the risky ones with the specific risk and one rewrite option.
{paste 30 replies}
15. 高频回复者的语气库
给一个初级 CM 用的语气库,让 ta 不再瞎编。
Build me a tone library of 30 reply snippets in my voice {voice notes}, organized by intent: greeting (5), question deflect (5), praise ack (5), polite no (5), playful banter (5), boundary set (5). Each snippet 1-2 sentences. No two snippets identical.
容易踩的坑
- “Thanks so much!” 套模板——读起来像批量生产,信任度下降。
- 跟黑粉对喷——升级永远不赢,只输旁观者。
- 在评论里给具体法律 / 健康 / 财务建议——风险高、收益低。
- 默认无视差评——围观者会把沉默读作心虚。
- 代理用一种和创始人对不上的语气——粉丝一眼看穿。
- 动不动”私信我”——公开讨论才最能建信任。
- 从不审计回复——偏离品牌的模式静悄悄堆积,等出截图就晚了。
优化技巧
- 每天把评论分三档:必回、可回、不理。“可回”是大头。
- 发帖后 30 分钟是回复 ROI 最高的窗口,专门留出来。
- 维护一个小的语气库(模板 15),别让初级 CM 临场发挥。
- 每月用模板 14 做一次回复审计——模式会漂移。
- 敏感话题准备法务确认过的标准回复文档。
- 相邻几条回复,刻意换 emoji 和长度,避免像机器人。
- 把”最佳评论 + 你的回复”对置顶到高流量帖最上方做社证。
FAQ
- 每条评论都要回吗?: 不。发帖后窗口期前 30-50 条认真回,之后分流。全部都回会训练观众预期、把你燃尽。
- 代理来回评怎么保持声音一致?: 用模板 8 + 一个严格的三词 voice doc,每周用模板 14 审计。
- 什么时候该删评论?: 人身攻击、人肉、垃圾、跑题骚扰。批评(哪怕粗鲁)一般保留——删了反而显得心虚。
- 回复要不要披露 AI?: 不强制,但受监管领域(医、金、法)千万别让 AI 全自动发。
- 评论变成围攻怎么办?: 冷静澄清一次,然后用平台功能关掉该贴后续评论。不要继续争。