Skip to content
LedgeurLedgeur
Guide

What an AI meeting assistant actually does, and the four questions that separate them.

The category went from novelty to default in about eighteen months, and the products in it are now so similar on a feature grid that the grid has stopped being useful. The differences that matter are architectural, and none of them appear on a comparison table.

The job, stated plainly

An AI meeting assistant listens to a conversation, writes down what was said, and turns it into something shorter that you can act on. Everything else is variation on those three steps: who joins the call, where the audio goes, what the summary is shaped like, and what happens to the record afterwards.

Most people arrive at this category because they took bad notes in an important meeting, or because they took good notes and missed the meeting doing it. Both are real, and either is enough reason to use one.

Question one: does something join the call?

Most assistants work by sending a bot into the meeting through a calendar integration. It appears in the participant list, everybody sees it, and somebody usually asks what it is. That is a social cost on every external call you take.

The alternative is capturing the audio the meeting is already playing on your machine. Nobody joins, nothing appears, and it works the same on Zoom, Teams, Meet, a phone call on speaker, or a conversation in a room.

It also decides something less obvious: a bot can be locked out. Video platforms control their own participant APIs, and several now ship their own free notetaker. A tool that depends on being admitted as a guest depends on a decision somebody else makes.

Question two: where does the audio go?

Nearly every product in this category uploads the recording and processes it on their servers. That is not sinister, it is just how the software was built, and it is worth knowing because it decides what a security review will find and what happens to the recording if you stop paying.

The alternative is running the speech model on your own machine. It is harder to build and it costs the vendor nothing to run, which is why the products that do it can afford to be free for one person.

Question three: does it tell you who said what?

A transcript without speakers is a wall of text. Separating voices is called diarization, and doing it well, with overlapping speech handled and the same person recognised across different meetings, is the single largest quality gap between products in this category.

Ask specifically whether it recognises a person in a later meeting, not just whether it labels Speaker 1 and Speaker 2 within one recording. Those are very different features and the marketing rarely distinguishes them.

Question four: can anything else read the record?

This is the one nobody asks and the one that matters most in a year. A meeting archive that only its own app can read is a dead end: you cannot ask your coding agent why a decision was made, or have your assistant check what a customer objected to last quarter.

The open answer is the Model Context Protocol, which lets an agent read the record directly. Ask whether the product has an MCP endpoint, or an API that is not shaped entirely around exporting to one CRM.

What to actually do

Record one real meeting with two candidates on the same day and read both transcripts. Accuracy differences are obvious immediately and invisible in a feature list, and speaker labelling either works on your accent, your microphone and your team or it does not.

Then check the price against how much you will use it. Nearly everything here is metered by minutes, which is fine until a fortnight of workshops.

Rather than reading about it
Compare the products directlyTemplates for each kind of meetingPricing, and why free is the whole product

Going deeper

The specific questions underneath this one.

How to Choose an AI Meeting Assistant in 2026

A practical guide to picking an AI meeting assistant — what features matter, the privacy trade-offs of bots and cloud transcription, and how to compare them.

7 min read
AI Meeting Summarizer: How It Works and What to Expect

AI meeting summarizers produce a structured recap from a transcript. Here's how LLM-based summarization works, what it gets right, and where it still makes mistakes.

6 min read
How to Never Miss an Action Item From a Meeting

Action items vanish because nobody captured them in the moment. Here's a system — and an automatic way — to extract them every time.

4 min read
Getting Started with AI for Meeting Notes

Adding AI to your meeting workflow takes 10 minutes to set up and can save hours per week. Here's a beginner's guide from setup to your first AI-generated meeting summary.

5 min read
How to Choose the Best AI Meeting Recorder in 2026

The best AI meeting recorder depends on your platform, privacy needs and budget. Here's a decision framework covering privacy, accuracy, integrations and cost.

6 min read
How to Get AI Meeting Notes Without a Bot Joining the Call

Notetaker bots are awkward and a privacy risk. Here's how to capture AI meeting notes without any bot joining your Zoom, Meet or Teams call.

5 min read
Meeting Notes for Remote Teams: Best Practices

Remote teams depend on written meeting notes more than co-located ones. Here's how to structure notes, who owns them, and how to share them so everyone can act on them.

5 min read
What Is Meeting Recording Software and Do You Need It?

Meeting recording software captures audio, video and transcripts of calls. Here's how it differs from video conferencing built-ins and when it adds real value.

5 min read
Private meeting transcription: running the models on your own machineMeetings for AI agents: MCP and the company record

Try it on one real meeting

Free for one person, permanently. Nothing is uploaded, and no account is needed to record.

Open LedgeurSee pricing