What is an AI meeting assistant?

The name covers a range of products, from a simple recorder to a tool that follows a meeting from the calendar invite to the last action item. Most of them do some part of three jobs.

• Before the meeting: pull together what was said last time, what is still open, and who is attending.
• During the meeting: record the audio, transcribe it as people talk, and label who is speaking. Some also show prompts or reminders while the call is live.
• After the meeting: write a summary, pull out decisions and action items, and send or file the result somewhere you will see it again.

Nearly every product does the middle job. Fewer do the first, and the third is where they differ most. A summary that sits in its own app still leaves you to copy the tasks into your task list and write the follow-up email yourself.

Four steps happen in order, and each one limits the next.

• Capture. The tool has to hear the meeting. It either joins the call as a participant, records the audio on your own device, or is part of the meeting app itself.
• Transcription. Speech recognition turns the audio into text. Quality depends on the microphone, background noise, accents, and how often people talk over each other.
• Speaker identification. The tool splits the transcript by voice and, where it can, attaches a name to each one. A bot can read names from the participant list. A tool that listens on your device has to recognize voices.
• Summarization. A language model reads the transcript and writes the summary, decisions and action items. It can only work with what the transcript says, so an error upstream becomes a confident sentence downstream.

How a tool captures the meeting is the first thing to check, because it decides who sees it, where it works, and what your IT team will ask.

None of these is the correct one. A sales team that wants every customer call logged may prefer a visible bot. Someone who moves between video calls, phone calls and a meeting room will get more from device audio.

These tools save real time, and they also make mistakes that read as fluent, tidy text. The errors below are common to the whole category.

• Names, numbers and jargon. Product names, acronyms, prices and dates are the words most likely to be transcribed wrongly, and the ones that matter most.
• Who said it. Speaker labels slip when voices are similar, when several people share one room microphone, or when someone joins late.
• What was actually decided. A summary can turn a suggestion into a decision, or drop the condition attached to one. “We will ship if the test passes” becomes “we will ship.”
• Owners and dates. An action item with no owner in the conversation may be given one in the summary. Check that every assignment was said out loud.
• Tone. Sarcasm, hesitation and disagreement rarely survive summarization.

The practical rule: treat the summary as a first view and the transcript as the record. If a tool does not let you jump from a summary line to the words behind it, you cannot check it quickly enough to trust it.

Recording a conversation has rules, and they differ by country and, in the United States, by state. Some places allow recording when one participant agrees. Others require everyone to agree. This guide is not legal advice; ask your own counsel if you record customers, candidates or patients.

Whatever the law says where you are, tell people. A tool that joins as a bot announces itself. A tool that records on your device does not, so the notice has to come from you. Beyond that, ask each vendor where recordings and transcripts are stored, how long they are kept, who at the vendor can read them, and whether your meetings are used to train models.

• Capture method. Bot, device audio, or built in. Does it cover the meetings you actually have, including the ones in a room?
• Speaker identification. Does it label people correctly in your real meetings, with your real microphones?
• A checkable summary. Can you get from any summary line to the transcript behind it?
• Search across meetings. Can you find a decision from three months ago without knowing which meeting it was in?
• What happens to action items. Do they become tracked tasks with the meeting attached, or a list to copy?
• Help before the meeting. Does it remind you what you promised last time?
• Where the output goes. Does it work with the email, calendar, chat and project tools you already use?
• Data handling. Storage location, retention, access, and training use, in writing.

Then try it on two or three real meetings, including a messy one. Our comparison of eight AI meeting assistants sets out how the main products differ on these points.

We make steer, so read this section with that in mind. steer is an AI work assistant, and meetings are one part of what it does. It captures a meeting from your computer’s microphone and speakers, so no bot joins the call, and it works for in-person conversations as well. It transcribes live, identifies speakers by voice, and can surface a relevant earlier commitment while the meeting is still going.

Afterward, the summary and transcript stay linked to the notes and to-dos that come out of the meeting, and the same assistant can draft the follow-up email or check the related ticket. The AI meeting assistant page describes that flow, and AI meeting transcription covers the transcript itself. steer is in beta and the desktop app currently runs on macOS.