Best AI meeting note-takers for a small business
The best AI meeting note-taker isn’t the one with the cleverest summary. It’s the one that puts your notes where you actually need them without creating a problem you didn’t have. They all transcribe a call well now, so the real questions are three: where the notes have to land, whether you can stand a bot joining the call to get them there, and who’s allowed to keep and train on the recording once it exists.
Fathom has the most generous free tier and the highest rating in the category, which makes it the default for most people. Fireflies logs straight into your CRM and is the pick for a sales team. Otter is built for live, shared notes, and is also the one being sued over consent. Granola and Circleback take notes without a bot ever joining the call, which is what matters for solo and sensitive work. And for the call you’d never put on someone else’s server, there’s a private route that costs nothing.
The line-up, and where each one’s notes land
Read the last two columns first, because where the notes end up and whether a bot shows up on the call decide more than any feature list.
| Tool | What it really is | Free tier | Paid entry (annual) | Bot joins the call? |
|---|---|---|---|---|
| Fathom | The generous all-rounder | Unlimited recording and transcript | ~$16/user/mo | Yes |
| Fireflies | Notes wired to your CRM | 800 min storage/seat, 20 AI credits/mo | $10/user/mo (CRM on $19 tier) | Yes, or desktop bot-free |
| Otter | Live, shared transcription | 300 min/mo, 30 min/call cap | $8.33/user/mo | Yes (OtterPilot) |
| Granola | A bot-free AI notepad | Free, limited history | $14/user/mo | No |
| Circleback | Bot-free, action-item obsessed | Trial only | ~$20.83/user/mo | No |
Prices are in US dollars, on annual billing, before tax, and this category re-prices and renames its plans constantly, so check each figure on the vendor’s own page before you commit. The split that matters isn’t the price column, it’s the last one, and the two questions underneath it: what your notes have to become, and who else you’re comfortable having in the room.
Accuracy is solved. Decide where the notes go.
Everyone compares these on transcription accuracy, and that stopped being a real difference a while ago. They mostly run the same class of speech model, land somewhere around 85 to 95 percent on clear English audio, and trip on the same things: three people talking over each other, strong accents, a bad line. Choosing on accuracy is like choosing a car on whether the wheels are round.
So decide on the two things that actually differ. First, where the notes have to end up. A note that’s only useful as a deal record wants the tool that writes to your CRM by itself. A note three people edit while the call runs wants the collaborative one. A private file note from a client call wants the tool that keeps it private. Second, whether a bot joining the call is a problem. A visible “Notetaker” in the participant list is nothing on an internal stand-up and awkward on a first sales call or a delicate one-to-one. Answer those two and the category sorts itself faster than any feature grid will.
Fathom: the most generous free tier, and the top rating
If you don’t know where to start, start with Fathom, because for a lot of small businesses the free plan is the entire answer. It joins your Zoom, Meet or Teams calls, records and transcribes them, and has the summary and action list ready about thirty seconds after you hang up. The free tier is the most generous here: unlimited recordings, unlimited transcription, unlimited storage. What’s held back for paid plans is the good stuff on top, the AI-generated action items and the advanced summary formats.
It’s also the best-liked tool in the category by a distance, rated 5.0 on G2 across more than 6,000 reviews, named to G2’s 2026 Best Software awards, and HubSpot’s most-used app of 2025. When you want the paid features, Premium is about $16 a user a month on annual billing, and the CRM sync to HubSpot and Salesforce sits higher up on the Business tier at around $25, so cost it there if the CRM is why you’re buying. One name-check worth making: the note-taker is the independent company that raised a $17M round in 2024, not the unrelated survey tool also called Fathom that got acquired in 2026. Same word, different product.
Fireflies: the one that logs to your CRM
Buy Fireflies when the notes have to become CRM records, not just documents. It’s the sales team’s pick, because it writes call summaries and action items into HubSpot or Salesforce and adds the conversation intelligence a sales manager actually reviews: talk-time ratios, sentiment, topic trackers. It handles 100-plus languages, the widest support here, so it’s the sensible choice for a team spread across countries, and Fireflies pitches itself as the preferred platform for CIOs across the Fortune 500, which tells you it scales past anything you’ll need.
The free plan gives you 800 minutes of storage per seat (a running total, not a monthly refill), 20 AI credits a month and the AskFred assistant. Pro at $10 a user a month on annual billing adds unlimited transcription and custom vocabulary; Business at $19 brings the CRM sync, conversation intelligence and unlimited storage that are the actual reason to buy it. Two catches to price in. The AI features run on a monthly credit allowance, so heavy users pay above the sticker once the credits run out. And the CRM logging you came for is gated to that $19 Business tier, so budget $19, not $10. The bot joins as a visible participant by default, though a bot-free desktop mode exists if that matters.
Otter: live shared notes, and the cautionary tale
Otter is the pick when the value is watching the transcript scroll in real time and editing notes together during the call, rather than reading a tidy summary after it. Its OtterPilot joins from your calendar, and that live, collaborative transcript is what it does better than anything else here. It’s the tool for a team that treats the meeting doc as a shared workspace while the meeting is still happening.
Watch the minute caps, because they bite early. The free plan is 300 minutes a month with a 30-minute limit per conversation, so a single 45-minute call loses its last quarter. Pro is $8.33 a user a month on annual billing ($16.99 monthly) for 1,200 minutes and a 90-minute cap; Business is $19.99 for unlimited minutes with a four-hour cap per call. Otter holds SOC 2 Type II and added HIPAA with a signed agreement on Enterprise in 2025, which matters for health providers, and it transcribes six languages, four of them mature. But Otter is also the tool carrying the category’s biggest warning label, and it’s worth its own section below rather than a footnote here.
Granola and Circleback: no bot in the room
Granola is the pick when you don’t want a bot in the call at all. It runs on your own Mac or Windows machine and listens to your computer’s audio directly, so nobody sees a “Notetaker” join, and it deletes the audio once it’s transcribed rather than keeping a recording. It also works differently: you jot rough notes during the call the way you always have, and Granola fleshes them out afterwards using the full transcript, so the notes come back shaped around what you thought mattered instead of a generic summary you have to trim. The free plan limits how much history you can open; Business is $14 a user a month, billed monthly with no annual discount, and adds unlimited history plus Notion, Slack, HubSpot and Zapier. It runs on macOS, Windows and iOS, no Android yet, and it’s SOC 2 and GDPR but not HIPAA. It’s also the tool with the most momentum: a $125M round in March 2026 at a $1.5B valuation, on the back of which it’s pushing hard into team and enterprise features, so if you’re a solo user, watch that its attention doesn’t drift upmarket and away from you.
Circleback is the under-covered one, and the sharper pick if what you actually want out of a meeting is the follow-through. It’s bot-free the same way Granola is, capturing audio on your machine, but it’s built around action items: it pulls out who agreed to do what and pushes those into your task manager or CRM automatically, which is the part most note-takers do limply. At about $20.83 a user a month it’s dearer than Granola and has no free tier beyond a trial, and it stumbles on overlapping speakers like the rest. For a consultant or agency that lives on client calls and drops balls between them, it’s worth the look the listicles don’t give it.
No audio means no going back. The bot-free tools keep the transcript and the notes, not the recording. That’s the privacy win and the limitation in one line: you can re-read what was said, but you can’t re-listen to how it was said, and there’s no tape to check the day the transcript garbles a name or a figure. For most meetings that’s a fair trade. For the one call a year you might need to prove word for word, it isn’t.
The bot that joins meetings you didn’t invite it to
Here’s the risk nobody puts in the feature table. These tools connect to your calendar once, then send a bot to record every meeting on it by default, and that default has caused real damage. The story every operator should know: in September 2024 a VC firm ran Otter on a Zoom pitch, the humans left, and Otter kept transcribing the investors’ private post-call debrief, their words about “cooked metrics” and strategic failures, then auto-emailed the full transcript to the founder they’d just met. The deal died. Nobody chose to send that; the tool did, because it was set to.
It’s not one bad tool. Otter is now the subject of a consolidated privacy class action in California (In re Otter.AI Privacy Litigation) alleging its Notetaker recorded people without all-party consent and trained on the results. Fireflies was hit with a separate suit in late 2025 over capturing a voiceprint without consent. And Read.ai, a copilot that markets itself as joining everything, was formally banned by a university’s IT department for auto-joining meetings without consent, with a nasty wrinkle: when it emails a summary to outside participants, anyone who clicks the link can spin up their own account with calendar access, and the bot spreads.
The fix takes two minutes and you should do it the day you install any of these. In the tool’s settings, switch the notetaker from “join every meeting” to “meetings I manually select”, and if there’s an “external meetings” toggle, turn auto-join off for anything outside your own company. Then keep the human habit that no setting replaces: say “I’ve got a notetaker on this, all good?” at the top of the call, before it joins, every time. A bot in the participant list makes it obvious the moment you’ve skipped that, which is exactly why you shouldn’t.
Who’s training on your meetings
The quieter question, and the one that ties this to owning your wider AI setup: once your meeting is transcribed, does the vendor train its own AI on it? Mostly yes, by default, and you have to opt out. Otter’s own policy has it training its models on de-identified recordings and transcripts, which is the practice the lawsuit is about. Fathom’s marketing implies it doesn’t, but its policy reserves the right to train in-house models on de-identified data unless you turn it off in settings, and Granola does the same. The one clean exception is Fireflies, whose terms commit across every plan not to train any AI on your content.
The phrase to distrust is “we don’t train on your data”. It almost always means the third-party model, OpenAI’s or Anthropic’s, doesn’t, which the big labs already promise on their business tiers anyway. It says nothing about whether the note-taker vendor trains its own model on you, which is the part that matters. If your calls carry anything confidential, either pick the tool with the binding no-train commitment, dig into the settings and switch training off on day one, or use the private route below and skip the question entirely.
The note-taker you might already own
Before you add a subscription, check what’s inside the video tool you already pay for, because the built-in note-takers have quietly caught up. Zoom’s AI Companion is included in paid Zoom Workplace plans (and sold standalone from about $8.33 a user a month), and it now takes notes on Teams and Meet calls too, not just Zoom, though on those it joins as a visible bot. Microsoft Teams has intelligent recap, but it needs a Teams Premium or Microsoft 365 Copilot licence, so it’s only “free” if you already hold one. Google Meet’s Gemini “take notes for me” was enterprise-only, and in mid-2026 Google extended it to the paid consumer AI Pro and Ultra plans, so a solo operator on one of those now gets it across platforms and in-person meetings too.
They’re fine for internal meetings and improving fast. The limit is that each is tied to one ecosystem and still does less than the dedicated tools: weaker summaries, no CRM logging, and a note-taker that stops at the edge of its own platform. The rule is the one the 2026 shortlist applies to every category: start on what you already own, and add a dedicated tool the day you hit a wall you can name.
Ask your meetings from the chat model you already pay for
The newest useful layer isn’t a note-taker feature at all. It’s asking the chat model you already pay for about everything said across your meetings. Fireflies was the first meeting tool in Anthropic’s Claude connector directory: connect it once and you can ask Claude “what were the main objections in this week’s sales calls?” or “list every action item assigned to me this month” and it reads across your transcripts to answer. It runs on an MCP, a standard plug that lets an AI reach into another app, the same mechanism our other comparisons cover.
Most tools also ship their own in-app version: Fireflies’ AskFred, Granola’s chat, Fathom’s Ask Fathom, all letting you question a meeting in plain language instead of scrolling a transcript. This is the rung where the note-taker stops being a filing cabinet you never reopen and becomes something you interrogate, and it leans on the tool most small businesses already have, the $20 chat model doing the asking.
The private route, when no cloud tool will do
Every tool on this page sends your meeting audio to a vendor’s servers, and most train on it. That’s fine for most calls and wrong for a few: the board conversation, the legal matter, the client who assumes they’re only talking to you. For those, skip the cloud entirely and run transcription on your own machine, free, with nothing leaving the laptop and no bot in the call. The full build is in how to run private AI transcription on your own laptop, which is the private alternative to everything here and where the do-it-yourself summariser lives too. It’s more setup than clicking install on Fathom, and for genuinely sensitive work it’s the honest answer.
Recording law, which in Australia depends on your state
Keeping the file on your own laptop doesn’t change the law on recording the conversation, and in Australia that law is set state by state, which trips up businesses working across the country. In Victoria, Queensland and the Northern Territory you can generally record a private conversation you’re a party to. In New South Wales, Western Australia, South Australia, Tasmania and the ACT you can’t, even as a participant, without the others’ consent, and WA is the strictest of the lot. A single call can pull in the law of every state a participant is sitting in, so the safe rule is the simple one: assume the strictest applies, tell everyone there’s a notetaker running and get a yes before it starts, and don’t forward the transcript to anyone who wasn’t on the call, because circulating it can be a separate offence even in the permissive states. The United States works the other way round, one-party consent by federal default with about a dozen two-party states like California, so if you record across borders, check where the other person is, not just where you are. None of this is legal advice; if a recording could ever matter in a dispute, get some.
Should you even be recording this one?
One honest check before you automate your whole calendar: a note-taker on every meeting can quietly make the meeting problem worse. When the summary writes itself, it gets easier to hold the call that should have been a message, and easier to send a bot instead of showing up, which is how you end up with the now-real spectacle of a Zoom room holding six people and ten notetakers. If a bot attending in your place would capture everything you needed, you probably didn’t need to be there, and the fix is one fewer meeting, not a better transcript.
There’s a solo-versus-team split underneath the whole category, too. The real payoff of these tools is shared memory: notes the whole team can search, a record that survives someone leaving, a handover that doesn’t depend on one person’s recollection. A business of one gets little of that. For a solo operator the honest case is narrow, being present in a client call while it’s captured for you, and it only pays above a few of those a week. Below that, jotting the three things that matter beats installing anything. And know that people behave differently on the record: most soften or perform once a notetaker is visible, so for the conversation where you need candour, the best tool might be no tool.
So, which one should you pick?
Answer the destination question and the pick mostly makes itself. Want it working, free, today, and you’re most people? Fathom free. Running a sales team where notes have to hit the CRM? Fireflies Business at $19. Want the transcript live and shared while the call runs? Otter, with the consent settings locked down first. Don’t want a bot in the room, or you take sensitive and client calls? Granola for notes, Circleback if action items are the point, or Jamie if you’re in Europe. Already living inside Zoom, Teams or Google? Try the built-in before you pay. And for the calls too sensitive for any cloud, the local route.
Then spend your first hour on the settings that decide whether it’s worth having. Turn off join-every-meeting so the bot isn’t gate-crashing calls it shouldn’t. Switch training off, or pick the tool that never had it on. Wire the integration so notes land where you’ll see them, in your CRM or Slack or Notion, not in an app you forget to open. And if the notes you keep losing are from phone calls rather than meetings, a note-taker won’t help and an AI receptionist will. A free plan set up properly, with consent asked and training off, beats a paid one left on its defaults every time.
Questions people ask
- What's the best AI meeting note-taker for a small business?
- It depends on where the notes need to end up. For most people Fathom is the default, because its free plan is the most generous in the category and it's the highest rated. Pick Fireflies if the notes have to become CRM records for a sales team, Otter if you want live shared transcription during the call, and Granola or Circleback if you don't want a bot joining at all. Decide the destination first and the tool mostly picks itself.
- Is there a genuinely free AI note-taker?
- Yes, and Fathom's is the most generous: unlimited recording, transcription and storage, with the AI action items and advanced summaries reserved for paid plans. Otter's free plan gives 300 minutes a month with a 30-minute cap per call, Fireflies gives 800 minutes of storage per seat (a running total, not a monthly refill), and Granola is free with a limited history of past meetings. For a lot of small businesses a free plan is the whole answer.
- What's the difference between Fathom and Fireflies?
- Fathom is the generous all-rounder for individuals and small teams: the best free plan and the top rating, built around fast, clean summaries. Fireflies is built for sales teams: it logs calls into HubSpot or Salesforce, adds conversation intelligence like talk-time and sentiment, and supports 100-plus languages, though the CRM logging you'd buy it for sits on its $19 Business tier, not the $10 one. If your notes need to live in a CRM, Fireflies. If they just need to be good and free, Fathom.
- Which AI note-taker doesn't join the meeting as a bot?
- Granola and Circleback. Both run on your own Mac or Windows machine and capture the audio directly, so nobody sees a 'Notetaker' join the call, and Granola deletes the recording once it's transcribed. Jamie works the same bot-free way and is popular in Europe. Fireflies has a bot-free desktop mode too. And for a call you'd never put on any vendor's server, you can run the whole thing locally instead.
- Do I still need a note-taker if I use Zoom, Google Meet or Teams?
- Maybe not. Zoom AI Companion is included in paid Zoom plans, Microsoft Teams has intelligent recap if you hold a Teams Premium or Copilot licence, and Google Meet's Gemini notes now work on paid Google AI Pro and Ultra plans. They're fine for internal meetings. Buy a dedicated tool when you hit a wall the built-in can't clear: CRM logging, capturing calls across different platforms, or a genuinely better summary.
- Are AI meeting notes private enough for sensitive calls?
- Every cloud note-taker sends your meeting audio to a vendor's servers, and most of them train their own AI on it by default. That's fine for most calls and wrong for some. Otter's auto-joining notetaker is the subject of a consolidated privacy class action over recording people who didn't agree. For confidential conversations, use a bot-free tool that doesn't retain audio, or run transcription locally on your own machine so nothing leaves it. And tell people you're recording, every time.
- Do AI note-takers train their AI on my meetings?
- Most do, by default, and you have to opt out. Otter trains its models on de-identified recordings and transcripts, which is central to the lawsuit against it. Fathom and Granola also train their own models on de-identified data unless you turn it off in settings. The exception is Fireflies, whose terms commit across every plan not to train any AI on your content. 'We don't train on your data' usually means the third-party model like OpenAI's doesn't; read it carefully, because the vendor often still trains its own.