Three Methods I’ve Tried for AI Meeting Notes: Here’s What Actually Works

Meeting notes are one of those things every workplace struggles with. Not the attending meetings part—that’s unavoidable. The struggling happens after: converting what was discussed into actionable notes, distributing them, and making sure everyone follows up on commitments.

I’ve been experimenting with AI-assisted meeting notes for about six months now. I’ve tried several approaches, paid for tools that didn’t deliver, and eventually found a workflow that actually works for my team. Here’s what I learned.

## Method 1: Otter.ai for Real-Time Transcription

**The Setup:**
Otter.ai is a dedicated transcription service that integrates with major video conferencing platforms. You connect it to your Zoom, Google Meet, or Microsoft Teams meeting, and it transcribes in real-time while identifying different speakers.

**My Experience:**

For English meetings, Otter is impressive. The transcription accuracy is genuinely good—probably 90%+ for clear audio with native speakers. The speaker identification works well for smaller meetings, and the ability to see live captions during the meeting is genuinely useful for accessibility.

After the meeting, Otter generates an automatic summary. It pulls out key points and action items, which sounds great in theory.

**The Problems:**

The Chinese transcription quality drops significantly. Any non-standard accent, industry terminology, or name pronunciation leads to garbled output. For our team, which operates in both English and Chinese, this is a serious limitation.

The AI-generated summaries are also mechanical. They capture the words but miss the context. “Discussed Q3 strategy” doesn’t capture that we decided to pivot away from our original plan because of budget constraints. You still need human interpretation to make the notes meaningful.

The free tier gives you 300 minutes per month, which isn’t much if you run several meetings weekly. The paid plans start at $20/month for individuals.

**Verdict:** Good for English-only teams on a budget. Limited beyond that.

—

## Method 2: Feishu (Lark) Miaoji (Meeting Highlights)

**The Setup:**
We use Feishu (Lark internationally) for our internal communication. Their built-in “Miaoji” feature provides automatic meeting transcription and AI summarization for recorded meetings.

**My Experience:**

The Chinese language processing is excellent—this is its core strength. If your meetings are primarily in Chinese, Feishu Miaoji handles transcription and understanding significantly better than Western tools.

The integration with Feishu Docs means meeting notes automatically appear in your team workspace. The action items can be converted to Feishu tasks with assignees and deadlines, which sounds like a seamless workflow.

**The Problems:**

The summarization quality is limited. It generates something, but the output is often generic. “Team discussed project status” tells you less than nothing—it wastes your time reading obvious content while missing the nuanced details that matter.

You can’t interact with the summary. If something is unclear, you can’t ask “what specifically did they decide about the timeline?” You get what the AI generated, whether it’s useful or not.

The quality depends heavily on the audio quality and speaking clarity. In larger meetings or with people speaking over each other, transcription accuracy drops.

**Verdict:** Excellent for Chinese-language teams using Feishu. Limited for anything requiring deeper analysis.

—

## Method 3: Recording + ChatGPT/Claude (My Current Approach)

**The Setup:**
This method requires a bit more manual work but gives me significantly better results:

1. Record the meeting using my phone or a dedicated recorder
2. After the meeting, I use my phone’s built-in transcription (or a dedicated app if needed)
3. Paste the transcript into ChatGPT or Claude
4. Use a carefully crafted prompt to generate structured notes

The prompt I use:
“`
This is a meeting transcript. Please organize it into a professional meeting notes document with:

1. Meeting basics (date, participants, topic)
2. Main discussion points
3. Decisions or conclusions reached
4. Action items with owners and deadlines (if mentioned)
5. Next meeting date (if mentioned)

Make the language professional but natural. Use clear, accurate terminology. Focus on capturing what was actually decided and what needs to happen next, not just what was discussed.
“`

**My Experience:**

The quality is significantly better than automated tools. Because I’m working with the full transcript and can ask follow-up questions, I can clarify ambiguities and ensure the notes capture what actually matters.

Example: After pasting in a client call transcript, I might follow up with:
– “The client seemed concerned about timeline. What specifically did they mention?”
– “List only the action items, organized by owner”
– “Summarize the key risk they flagged in one sentence”

The AI can do this because I can have a conversation with it. It’s not limited to what an automated summary algorithm decided to highlight.

**The Problems:**

This method requires manual recording, which is an extra step. Some meeting platforms record automatically; others require me to remember to start a recording.

Long meetings (over an hour) can produce transcripts that are too long for a single prompt. I either split them or summarize in sections.

The transcription step is still manual unless your phone has good built-in transcription. (On iPhone, the voice memos app transcribes automatically and quite accurately. Your experience may vary.)

**Verdict:** Best quality for teams that prioritize accurate, useful notes. Requires slightly more effort than automated solutions.

—

## My Honest Comparison

Factor	Otter.ai	Feishu Miaoji	Recording + AI
English transcription	Excellent	Good	Depends on recording
Chinese transcription	Poor	Excellent	Depends on recording
Summary quality	Average	Below average	Good
Ability to ask questions	No	No	Yes
Integration	Good (Zoom, Meet)	Excellent (Feishu)	Manual
Cost	$20+/month	Free/Included	ChatGPT/Claude subscription
Effort required	Low	Very Low	Moderate

—

## Which Should You Use?

**Use Otter.ai if:**
– Your meetings are primarily in English
– You need automatic integration with video conferencing
– Volume is high and you need a set-it-and-forget-it solution

**Use Feishu Miaoji if:**
– Your team uses Feishu/Lark
– Most meetings are in Chinese
– You want seamless integration with your existing workflow

**Use Recording + AI if:**
– Quality matters more than convenience
– You want to be able to ask questions about the meeting content
– You don’t mind the extra step of recording and processing
– You want accurate notes that capture nuance

—

## My Current Workflow

For important meetings: I record (or ensure they’re recorded in the platform), then process with ChatGPT using the structured prompt above. This takes about 30 minutes for a one-hour meeting—less if I’m thorough with my initial prompt and follow-up questions.

For routine meetings: I sometimes use Feishu Miaoji for quick notes that just need to capture the basics. Not great, but good enough for low-stakes follow-up.

For client calls: Always the recording + AI method. The extra effort is worth it for accuracy.

—

## Additional Tips

Regardless of which method you choose:

**Record meetings whenever possible.** Even if you don’t process them right away, having the recording means you can go back if something comes up later.

**Process notes within 24 hours.** The longer you wait, the more context you lose. I try to do it immediately after the meeting while everything is fresh.

**Share notes within 24 hours.** People remember meetings briefly, but start forgetting quickly. Get notes out while they’re still relevant.

**Include clear action items.** If there’s one thing that matters in meeting notes, it’s making sure someone knows they’re responsible for something and when it’s due. Don’t let good discussion lead to nothing being accomplished.

—

## Final Thoughts

AI-assisted meeting notes are better than manual notes, but the specific solution matters. The “right” tool depends on your language, your existing workflow, and how much quality matters for your specific use case.

I’ve landed on the recording + AI method for high-stakes meetings because the quality is worth the extra effort. For lower-stakes situations, automated tools do enough.

Your needs might be different. The important thing is to actually experiment and find what works for you, rather than giving up on the problem because no single solution is perfect.