An Old Problem for Managers

After leading a team for several years, I've found that the most energy-draining part of meetings isn't the meeting itself—it's what happens afterward.

A one-hour meeting produces thousands of words of transcription. Who's responsible for what, when it's due, what the acceptance criteria are—all of this information is scattered throughout the conversation. My old approach was to have a team member manually compile the minutes, which usually led to three problems:

First, delays. After the meeting, everyone returns to their desks, and the minutes often don't come out until the next day, by which point the momentum has faded. Second, vagueness. The minutes say "we discussed the launch plan," but not who would revise it or when it would be done. Third, buck-passing. Two weeks later, someone asks "who's following up on this?" and you can't find a clear owner anywhere in the chat history.

For a small team, this means every meeting incurs extra management overhead—and that cost ultimately falls on the manager.

What AI Can Do Here

Later, I handed this step over to AI. The core idea: AI doesn't make decisions, but it's responsible for locking decisions down in a structured way.

Specifically, AI can do four things in the meeting minutes scenario:

1. Extract action items from lengthy conversations. In a meeting transcript, genuinely useful information may only account for twenty percent. AI can take colloquial expressions like "maybe we'll take a look at this next week" or "this part might need further confirmation," recognize the action signals within them, and distill them into list items.

2. Fill in the three essential elements of action items. What managers care about most is: what to do, who does it, and when it's due. AI enforces a check of these three items for every action item—those missing an owner get explicitly flagged with "⚠️ No owner assigned during the meeting," and those missing a deadline get flagged with "⚠️ No deadline agreed." This flagging itself has management value—it forces you to fill the gaps before sending out the minutes.

3. Identify risks and points of contention. When someone says in a meeting "I'm not sure this plan can be delivered on time," the human ear tends to miss this kind of statement, but AI will categorize it under "items pending confirmation" and remind you to follow up separately.

4. Generate a format ready for distribution. Markdown tables and task lists grouped by person—just paste them into your collaboration tool and they're ready to use.

My Actual Workflow

The process is simple, four steps:

1. Transcribe the recording. Use a transcription tool (many conferencing apps have one built in) to get the text.
2. Feed it to AI. Process the full text with the prompt template below.
3. Manually proofread. Focus on verifying owners and deadlines—AI might misattribute "Xiao Li said Xiao Wang could do this," so this step can't be skipped.
4. Send it out within half an hour. While everyone still remembers, sync it directly in the group chat; anyone wrongly assigned as an owner will immediately correct it.

Here's the prompt template I use—feel free to copy it directly:

你是一名会议纪要助手，请从以下会议转写文本中提取信息，输出三部分：

## 一、会议概要
用3-5句话概括本次会议的核心议题与结论，只写已达成的内容，不推测。

## 二、行动项清单
用表格输出，列：序号 | 任务描述 | 负责人 | 截止时间 | 备注
要求：
- 任务描述必须具体可执行，含验收标准（如文本中未提及则写"未提及"）
- 负责人：只根据文本中的明确指派填写；未指派的填"⚠️ 会上未指定"
- 截止时间：只填写文本中明确出现的时间；未出现的填"⚠️ 未约定"
- 如有多人负责同一任务，注明第一责任人

## 三、待确认事项
列出会议中的争议、疑虑或未闭环的问题，格式：问题 | 提出人 | 建议跟进方式

注意：不要虚构任何文本中不存在的信息；口语表达请转为书面语但保持原意。

会议转写文本：
{在此粘贴转写内容}

The key design in this template is those "⚠️" markers—it's not about having AI fill in the blanks, but about having AI expose information gaps. That's where the risk control lives.

Before and After

The biggest change isn't saving time—it's that the minutes go from being a "memo" to a "contract." Once action items with all three elements are sent out, team members have no ambiguity about what they should do, and future follow-ups have a documented basis.

Three Risks Managers Must Watch

When using AI to process minutes, three things must remain under human control:

Misattribution. When references in the transcript are unclear (e.g., "let him handle this"), AI might guess the wrong person. So the first thing to do during proofreading is verify owners item by item.

AI's over-inference. Models sometimes "helpfully" fill in details that aren't in the context. The "do not fabricate" instruction in the prompt is necessary but not sufficient—the person who signs off is you.

Sensitive information. If a meeting involves personnel matters, compensation, or trade secrets, think carefully about which model you're using and where the data goes. This is the point I care about most as a manager: tool selection is itself a risk-control decision. When choosing an API service, look beyond capabilities to data usage policies, whether data can be excluded from training, and similar terms (policies vary by platform—refer to official pricing pages and terms of service).

Final Thoughts

Automating meeting minutes essentially frees managers from the role of "information courier," letting them focus their energy on real judgment—which tasks take priority, which people are at risk, which commitments need to be tracked to fulfillment.

If you're also looking for a stable, controllable AI API for running this kind of text processing task, give https://api.thistoken.ai/register a try—register and connect to start building your own meeting minutes pipeline.

---

Every example in this post runs with a single API key — get yours at https://api.thistoken.ai/register and start in minutes.