你是一名产品分析师，负责整理用户反馈并输出需求优先级清单。

任务

我会分批提供用户反馈，每批不超过100条。请对每批做以下处理：

1. 逐条提取核心诉求，忽略情绪化表述和无关内容；
2. 将同义问题归并（如"登录不了"和"无法登录"视为同一问题）；
3. 为每个问题打标签：问题类别（功能缺陷/功能建议/性能/体验/其他）、

严重程度（阻断使用/影响体验/锦上添花）、情绪强度（1-5）；

1. 统计每个问题的出现次数；
2. 以Markdown表格输出：问题类别 | 问题描述 | 出现次数 |

严重程度 | 情绪强度均值 | 原始反馈示例（最多2条）。

注意

• 不编造反馈中不存在的信息；
• 一条反馈含多个问题时拆分处理；
• 不确定时标注"待确认"，不要猜测。

第一批反馈如下：

（在此粘贴反馈内容）

After aggregating all batches, do a second round, appending priority evaluation dimensions and weights (e.g., frequency 40%, severity 30%, implementation cost 20%, strategic fit 10%), and have the AI output the final ranked list.

## 4. Before vs. After Using AI

| Task | Purely Manual | AI-Assisted Workflow |
|---|---|---|
| Classifying 1,000 feedback entries | 2–3 days | ~1–2 hours with batch processing |
| Consistency of classification standards | Varies by person, inconsistent over time | Uniform rules, reproducible |
| Frequency statistics | Rarely done | Automatic deduplication, aggregation, and counting |
| Basis for prioritization | Gut feeling, whoever argues loudest wins | Explicit rules + weights, debatable and adjustable |
| Repeated labor next month | Start over from scratch | Reuse the prompt and workflow, near-zero marginal cost |

More important is the cognitive shift: priorities used to be a "feeling"; now they're "evidence"—in reviews, you can explain exactly how many feedback entries the top-ranked issue affected and how severe it is. Team members stop fighting over rankings and instead discuss whether the weight settings are reasonable—a completely different quality of discussion.

## 5. A Few Practical Reminders

1. **Don't skip the manual review**. AI will occasionally merge issues that shouldn't be merged, or miss rare but critical feedback (e.g., a security issue that appears only once). Spending 20 minutes scanning the list is necessary insurance.
2. **Once the workflow is stabilized, you can automate it via API**. When the batch-processing workflow is stable, you can programmatically call a large model API (e.g., via an aggregation platform for unified access—pricing per the official pricing page) to build a routine "feedback comes in, weekly report comes out" pipeline. That's when you truly approach zero manual labor.
3. **Iterate your prompts as the product evolves**. The classification framework isn't set in stone. In each retrospective, re-slot the "needs confirmation" items, and the framework will fit your product better and better.

## Final Thoughts

User feedback is the cheapest and most authentic product intelligence, yet its value has long been buried under "no time to organize it." AI can't decide your product direction for you, but it can turn thousands of noisy entries into a clear, debatable list—turning decisions from guesswork into evidence.

If you want to go further and turn this workflow into a service that runs automatically every day, consider connecting through https://api.thistoken.ai/register to access a unified model API, and save yourself even the manual copy-pasting time.

---

Tired of juggling provider integrations? Register at https://api.thistoken.ai/register and call every model through one base_url.