Set Up an AI Meeting Assistant in 30 Minutes (and Keep It From Embarrassing You)
Most people try an AI meeting assistant once, get a wall of text that nobody reads, and quietly turn it off. The tool usually isn't the problem. The setup is. A good assistant is a small system with three parts: what it listens to, what it's allowed to do with what it hears, and how you check its work.
Here's a setup that takes about half an hour and holds up in real use. (If your assistant seems to forget things between meetings, our last post on AI memory explains why.)
Step 1: Connect only what it needs
Start with two connections: your calendar and one place for notes, such as a shared doc or a folder. The calendar tells the assistant which meetings exist and who attends, so it can name speakers and file summaries under the right event. The notes location gives it somewhere to write.
Resist connecting your whole inbox and drive on day one. Every extra connection is more data the tool can see and more that can go wrong. You can widen access later, once you trust the output.
Most major platforms now have this built in. Zoom offers AI Companion meeting summaries and action items, and Google Workspace has its own AI note-taking for Meet. Dedicated tools like Fireflies, Otter, and others add features such as search across past meetings. Whichever you pick, the steps below are the same.
Step 2: Settle consent before the first meeting
This is the step people skip, and it's the one with legal teeth. An August 2025 class action against Otter.ai alleged that its assistant recorded not just its own customers but also other meeting participants who never agreed to it, raising claims under federal wiretap law and California's privacy statute. In August 2026 a federal judge let several of those claims proceed, including the federal Electronic Communications Privacy Act claim and the California Invasion of Privacy Act claim, while dismissing others. It is still an ongoing case, not a final verdict, but the lesson for everyday users is clear.
Many jurisdictions require everyone on a call to consent to being recorded. So build a habit: tell people at the start, or put a line in the invite ("an AI assistant will take notes"). Skip it entirely for sensitive conversations such as HR matters, legal advice, or anything confidential. I'm not a lawyer, so check your own company's policy.
Step 3: Decide what it drafts and what it sends
Split the assistant's jobs into two buckets.
Let it draft: a summary, a list of decisions, action items with suggested owners and dates, and a follow-up email for you to review.
Don't let it send or commit: emails to clients, calendar invites on your behalf, or task assignments to other people, at least at first. A wrong action item that lands in a colleague's task list erodes trust fast. A wrong draft that sits in your own folder costs you ten seconds to fix.
Step 4: Review summaries like a skeptic
Build a two-minute review into your day. For each summary, check three things: Are the decisions what was actually decided? Does each action item have a real owner? Is any number, date, or name wrong? Speech recognition stumbles on names and jargon, and summaries tend to sound confident even when they're mistaken.
If you catch the same error twice, fix the cause. Add jargon and names to a custom vocabulary if the tool supports it, and keep meetings from having people talk over each other.
After two weeks, check one metric: how many action items from summaries actually got done? If that number is climbing, widen the assistant's access. If not, tighten the review step before adding features.
The takeaway: an AI meeting assistant earns its place by being boring and accurate. Connect little, disclose clearly, draft but don't send, and review everything. Try it for one recurring meeting this week.
Sources: Fisher Phillips — New Lawsuit Highlights Concerns About AI Notetakers · In re Otter.AI Privacy Litigation, N.D. Cal. (Aug. 2026) · Zoom AI Companion note taking · Google Workspace AI note taking