Best AI Note Taker for Meetings in 2026 (We Tested Them)

There is no single best AI note taker, and the fork that decides it for you is whether a bot may join your call. If you want transcripts that are instantly shareable across a team, deep CRM and workplace integrations, and a recorder that can sit in calls you skip, pick Otter or Fireflies. If you would rather nothing extra appeared in the participant grid, pick Granola for clean personal notes, or Routines when the notes need to stay as files on your own Mac.

This is a first-person, we-tested rundown of those four. We will be generous about what each one does well, because they are all genuinely good at something. One disclosure up front: we build Routines, one of the options below. We have tried to keep the rest of this as fair as we would want someone to be about our own app.

Prices move constantly and vary by plan, so the table stays on capture and storage. Everything below is the first-hand version: what each tool does well, where it stops fitting, and what the recording felt like in the room. If you sit through more than a couple of calls a day, the right one buys back the twenty minutes you used to spend writing up notes and chasing what everyone agreed to.

The bot-joins-your-call problem

Before the list, the single biggest fork in the road. A large share of meeting assistants work by sending a bot into your call as a visible participant. You have seen it, a tile that says "Otter.ai" or "Fireflies Notetaker" sitting in the Zoom grid, recording on behalf of someone in the room.

That model has real upsides. The bot captures audio server-side, so it does not matter what device you are on, and the transcript is instantly shareable with people who were not even there. For sales teams and recruiters who live in scheduled calls, that is exactly what they want.

It also has real downsides. The bot is visible to everyone, which changes the room, some external guests are uncomfortable being recorded by a third-party participant, and a few security-conscious companies block unknown bots outright. And whatever that bot hears is processed and stored on a vendor's servers by default.

The alternative is local capture: an app on your own machine records the system audio and your microphone directly, with no extra attendee in the call. Granola popularized this approach, and it is the model Routines uses too. Neither one is universally better. They are answers to different priorities.

The three ways a tool gets your audio

You can transcribe a meeting without a bot in it, and the reason is plumbing. Three designs sit behind everything in this category.

A joined meeting bot. The vendor runs a participant account that dials into your call and records from inside it. The Fireflies AI Notetaker, in its own documentation, "joins your meeting as an AI Notetaker ... like any other participant" (Fireflies), and Otter's Notetaker joins Zoom, Google Meet, and Microsoft Teams the same way (Otter). Capture runs on the vendor's servers, which is why a bot can sit in a call you skip.

Local system-audio capture. An app on your Mac reads your microphone and the system audio output directly, the way a screen recorder does, and never touches the meeting platform. Granola's help pages state it flatly: "No bot joins your meeting". Routines captures the same way and splits the sources, recording your mic and the system audio as separate streams on the machine (meetings feature page).

Native platform hooks. The conferencing platform records and transcribes itself, so nothing extra appears in the participant list because the vendor is the platform. None of the four tools below use this route.

Mode matters more than vendor here. Otter is bot-based by default and also offers to "Record bot-free on desktop", which moves that session into the second design, so check which mode you will actually run rather than which camp the logo belongs to. One caution: the capture mode decides who is visible in the call, not where the audio is processed. A local recorder can still upload audio to a cloud transcription service, which the security section below breaks down.

How we judged each tool

We looked at four things that actually change your day: capture quality (does the transcript get the words and the speakers right), the notes themselves (are the summaries and action items usable without heavy editing), where your data goes (cloud by default or local), and the meeting experience (does a bot join, and how much setup is involved). Prices change constantly and vary by plan, so we are deliberately not quoting numbers here. Check each vendor's current pricing directly.

Granola

Granola earned its reputation by getting one thing right: it captures your meeting without a bot in the call. It listens to your Mac's audio locally and pairs the transcript with whatever you jot down yourself, then cleans your rough notes into something structured afterward. The result feels less like a transcription dump and more like the notes you wish you had taken.

What we liked: the no-bot capture is genuinely calmer than watching a recorder tile sit in the grid, and the note-enhancement step is excellent for people who already take sparse notes and want them tidied up. It is a polished, focused product.

Where it might not fit you: it is built around that personal-capture workflow, and your notes and transcripts still sync to Granola's cloud. If your priority is that meeting data never leaves your machine at all, that is worth checking against your own requirements. If you want the paper trail, our audit of where Granola's audio and notes go is built from its own privacy policy, terms, and help pages. We wrote a fuller side-by-side in our Granola comparison if you are weighing the two directly, and a longer verdict in our Granola AI review.

Otter

Otter is one of the most established names in the category, and for good reason. Its live transcription is fast and readable, it handles shared team workspaces well, and its meeting assistant can join calls to capture audio and even surface summaries and action items automatically. For teams that want a searchable, shareable record of every conversation, Otter is a safe, mature choice.

What we liked: real-time transcript quality is strong, the collaboration features are well thought out, and the integrations into common workplace tools are broad.

Where it might not fit you: the assistant typically joins as a participant, and your transcripts live in Otter's cloud as the default. If a visible bot or cloud storage is a problem for your meetings, that is the trade-off to weigh.

Fireflies

Fireflies sits firmly in the team-and-CRM camp. It deploys a notetaker bot to your calls, transcribes them, and then pushes summaries and action items into the tools your team already runs, with particularly strong hooks into sales and CRM workflows. If your meetings feed a pipeline, Fireflies is built for exactly that motion.

What we liked: the integration breadth is impressive, the automatic summaries are solid, and the searchable archive across an organization's calls is a real asset for teams that need shared recall.

Where it might not fit you: same fork as Otter, a bot joins the call and the data is cloud-hosted by design. That is the right call for a sales org and the wrong one for a confidential one-on-one with an external party.

Routines

Full disclosure again: this is ours. Routines is a macOS app that stores everything locally, and meeting notes are one of several things it does. It records your microphone and your system audio as two separate streams, so you get a live, speaker-separated transcript that distinguishes you from everyone else in the Zoom, Teams, or Webex call, with cross-channel echoes filtered out. No bot joins the meeting, because the recording happens on your Mac.

When you stop recording, your chosen AI turns the full transcript, attendee list, and duration into a one-page summary with key points, decisions, and action items, stored right next to the transcript. If you would rather that wrap-up arrived on its own, the meeting notes routine runs at the end of the workday and turns the meetings you recorded into follow-up notes with owners on each action item. You can also ask the live transcript questions mid-meeting ("what did I miss?", "draft a follow-up") and even paste a screenshot of a shared slide into the chat so the model sees the actual pixels.

What makes it different from everything above is where the output lands: your notes, transcripts, and summaries are plain markdown and a local SQLite database on your Mac, and they stay there unless you switch on Cloud Sync yourself. The meeting audio leaves too. It streams to Deepgram for transcription, on the key we manage if you are on Pro or in the free trial, and on your own Deepgram key on the Free plan. The AI comes with your account, or you bring your own Anthropic, OpenAI, or OpenRouter key, or sign in with your existing ChatGPT account, and there is a 7-day trial to test it on real calls.

Where it might not fit you: it is macOS-only, and because capture is local, it is built around you being in the meeting rather than dispatching a bot to a call you are skipping. If you need a server-side bot to attend calls without you, a cloud tool is the better answer. We go deeper on the meeting workflow on the meetings feature page and on the broader approach in our writeup on AI meeting notes.

So which one is the best AI note taker?

There is no single winner, and any honest list should say so.

• Choose a bot-based cloud tool like Otter or Fireflies if you want transcripts that are instantly shareable across a team, deep CRM and workplace integrations, and the ability to record calls you are not even on.
• Choose Granola if you want clean, bot-free personal notes on your Mac and you are comfortable with cloud sync for the convenience.
• Choose Routines if your priority is meeting notes that live as files on your own machine, captured without a bot in the call, with the freedom to bring your own model. The audio still goes to Deepgram to be transcribed, on our key on Pro or the trial and on your own key on Free.

The clearest way to decide is to run a few real meetings through whichever one matches your priorities, then look at how much editing the notes need and how the recording felt in the room.

Are AI notetakers legal?

Recording law decides this, not the software, and the standard changes at the state line.

Start with the federal floor, which is one-party consent. Under 18 U.S.C. § 2511(2)(d), a person not acting under color of law may intercept a communication where "such person is a party to the communication or where one of the parties to the communication has given prior consent to such interception", unless the interception is "for the purpose of committing any criminal or tortious act". Sitting in your own meeting clears that bar. States are free to demand more, and several do.

The ten states below are the most-searched ones, not a full map of the country. At least nine further jurisdictions demand more than the federal floor: Maryland, Massachusetts, Montana, New Hampshire and Washington require all-party consent, and Delaware, Nevada, Oregon and Connecticut impose all-party rules in specific situations. Check your own state before you record.

Four of these rows carry nuance that the table above flattens.

California does not cover every conversation. § 632 reaches a "confidential communication", meaning one where a party has an objectively reasonable expectation that it is not being overheard or recorded, so conversations in public or without that expectation fall outside the section. Violations also create civil liability under § 637.2.

Illinois rebuilt its statute after the courts struck the old one down in 2014 for criminalizing recording regardless of any privacy expectation. The legislature amended 5/14-2 the same year to reach only a "private conversation", one where the parties reasonably expect privacy, while keeping all-party consent for those. It also added a subsection permitting the recording of police performing their duties in public.

Michigan is the row we cannot clean up honestly. The statutory text is facially all-party: it bars using a device to eavesdrop on a private conversation "without the consent of all parties thereto", with no participant carve-out on the face of the section. Secondary legal sources commonly report that Michigan courts have read the companion definition of eavesdropping as covering only a third party listening in on other people's conversation, which would leave a participant needing only their own consent. We did not verify that case law against the opinion text, so we are not relying on it here. Treat Michigan as unsettled.

Georgia is the one statute above we could not pull from an official state source. Georgia's legislature site does not publish free full codified statute text, so the link goes to Justia's copy of the Code section rather than a state government page.

None of these statutes turn on whether the recorder shows up in the participant list. A silent app on your Mac and a visible bot tile are judged by the same consent test, so choosing the calmer capture mode does not lower the consent bar. Statute text checked 2026-08-10. This is general information, not legal advice.

What is the most secure AI note taker?

There is no honest way to crown one, because "secure" bundles several questions that these vendors answer differently and publish in different places. Here is what each says about itself.

Four things in that table are worth reading closely, because they are where the marketing and the policy tend to drift apart.

A no-training promise can mean two different things. Otter's page says "No customer data will be used to train or improve our AI Service Provider(s)' artificial intelligence models/algorithms", which is a promise about OpenAI-class vendors, while Otter separately de-identifies user data to train its own transcription and summarization models. One promise covers the model vendors Otter uses, the other covers Otter's own models, and only the first appears in that sentence. Granola states its own default plainly: "Granola trains on your anonymized data ... You can opt out of this in your Settings" (Granola security page). Fireflies goes the other way with "We don't train on it by default unlike other AI companies", and Routines states it "does not use your recordings, transcripts, notes or prompts to train models". We read Granola's policies line by line in our audit of its data handling, and ran the same exercise across Otter's published policies and Fireflies' security pages.

Bounded retention and unbounded retention are both defensible, but you should know which you signed. Fireflies publishes a floor rather than a ceiling: "data is saved for at least 12 months", with individual meeting deletion purging immediately and account deletion clearing everything within 30 days (Fireflies FAQ). Granola keeps notes indefinitely by default until you or an admin configures otherwise. Otter documents the Trash window, "Otter automatically deletes conversations from the trash after 30 days", but the page we read gives no default retention period for conversations you have not deleted.

Audio and transcripts are separate assets. Granola deletes one and keeps the other: "Once transcription is complete, the audio is deleted from our systems ... We do not retain audio recordings." A vendor can be strict about raw audio and relaxed about the transcript, which is usually the more sensitive artifact anyway, since it is searchable.

Self-reported is not the same as auditable. Every SOC 2 claim above comes from the vendor's own page. Granola cites "SOC 2 Type II certification", Fireflies says "SOC 2 Type II Compliant" (Fireflies), and Otter announced that it "successfully completed its SOC 2 Type II audit and received a SOC 2 Type II report" (Otter), on a post whose dateline and page metadata disagree by a year, a discrepancy we could not resolve from the source. Ask any of them for the report itself. Our own row is the blunt one. Routines has no SOC 2 status to point at, so if a certification is a procurement requirement where you work, that is a real gap on our side rather than a footnote. The rest of the audits live in the transparency directory.

Does Apple have an AI note taker?

Partly, and the native features are worth checking before you pay for anything.

Notes on Mac records and transcribes inside the note itself. Apple's guide describes recording "audio, pause, and resume recording, all from within your note", and says you can "view, search in, and copy live audio transcripts" by double-clicking the recording (Apple Support, checked 2026-08-10). The same page limits transcription to "Mac computers with M1 or later" and to a device language set to English, Spanish, Portuguese, Italian, French, German, Japanese, Korean, Simplified Chinese, or Traditional Chinese.

Summarizing that transcript is a separate feature that needs Apple Intelligence switched on. Open the note, "Click the recording, then click Summary" (Apple Support, checked 2026-08-10), which Apple lists as requiring the latest version of macOS, an M1 or later Mac, and describes as available in beta in specific languages and regions. Writing Tools handles the rest of your text with Summary, Key Points, List, and Table options across Notes, Mail, and most places you type, and Apple flags that it is "Not available on all Mac models or in all languages or regions" (Apple Support, checked 2026-08-10). Apple's own product page raises the bar for some features to "M3 and later and at least 12GB of unified memory" (apple.com, checked 2026-08-10).

The native path stops short of meeting notes. Apple's transcription page does not mention speaker labels, attendee lists, or joining a conference call, so a Notes recording gives you one undifferentiated transcript of whatever the Mac heard rather than notes that separate who said what.

Which Mac tools are genuinely offline

"Runs on your Mac" and "processes on your Mac" are different claims, and the gap between them is where most privacy marketing lives.

• Apple Notes is hybrid by design. Apple describes on-device processing "so it is aware of your personal information without collecting your personal information", with requests that need a larger model routed to Private Cloud Compute, where "Your data is never stored. It is only used to fulfill your requests." (Apple Support, checked 2026-08-10)
• Granola is hybrid. It transcribes in real time on the desktop without storing raw audio, then sends audio to third-party cloud transcription and uses cloud models to summarize. The output lands off the machine: "Notes are stored in our US-hosted AWS Virtual Private Cloud. They are encrypted at rest and in transit." (Granola)
• Otter is cloud-processed. "Otter uses AWS S3 storage and enables AWS SSE (Server Side Encryption) on data (S3 buckets)", and the page we read offers no on-device or offline option. (Otter)
• Fireflies is cloud-processed by default. Enterprise customers can add "Private Storage on Enterprise to store your meeting data at your preferred location", which relocates the bucket rather than the processing. (Fireflies)
• Routines keeps notes as markdown files on the Mac, and its transcription defaults to the cloud. Our own homepage says "Meeting audio goes to Deepgram (Whisper as fallback) to be transcribed", and the privacy page adds that "An offline transcription option is also available." Cloud transcription is the default, and that is worth saying plainly rather than leaving in the small print. The meetings feature page covers how capture and transcription fit together.

If your Mac question is really about dictation rather than meetings, we compared those tools separately in the best voice to text app for Mac.

Is there a free AI note taker?

All four have something free. What differs is the currency the limit is denominated in: minutes, storage, history, or feature count.

Last checked 2026-08-10. These pages move constantly, so treat the table as a snapshot and open the pricing page before you commit to anything.

FAQ

What is the best AI note taker for meetings?

It depends on how you meet. Granola is excellent for solo capture without a bot, Otter and Fireflies are strong for shared team transcripts and integrations, and Routines fits people who want meeting notes kept as files on their own Mac, captured by a local app rather than a bot that joins the call.

Do AI note takers join my call as a bot?

Many do. Otter and Fireflies-class tools typically add a meeting assistant or bot as a visible participant to capture audio. Granola and Routines record locally on your Mac instead, so no extra attendee appears in the call.

Are AI note takers private?

It varies widely. Most cloud tools store your transcripts on their servers. Routines keeps notes, transcripts, and summaries as markdown and SQLite on your Mac, while meeting audio streams to Deepgram for transcription, on the key we manage if you are on Pro or in the free trial, and on your own Deepgram key on the Free plan.

Do I need a subscription to use an AI note taker?

Not always. Several tools offer free tiers with limits, and Routines has a permanent Free plan with up to 3 active routines plus a one-time €5 of AI credit. A 7-day Pro trial lifts that to unlimited routines and included transcription, and you can also bring your own API key or sign in with your existing ChatGPT account.

Can I transcribe a meeting without a bot joining the call?

Yes. Three capture designs sit behind this category. A joined meeting bot dials into the call as a participant and records on the vendor's servers, which is how Otter's Notetaker and the Fireflies AI Notetaker work by default. Local system-audio capture runs as an app on your Mac that reads your microphone and the system audio output directly, so no extra attendee appears in the grid; Granola and Routines both capture this way, and Otter offers a bot-free desktop recording mode as well. Native platform hooks are the notes features built into Zoom, Teams or Meet, where the conferencing vendor records itself and no third party sits in the call. Dropping the bot changes who is visible in the meeting, not necessarily where the audio is processed, because a local recorder can still upload the audio to a cloud transcription service.

Are AI notetakers legal?

In the United States, yes, with a consent rule that changes at the state line. The federal Wiretap Act, 18 U.S.C. § 2511(2)(d), sets a one-party consent floor: a private person may record a communication they are a party to, or where one party has given prior consent, unless the recording is made for the purpose of committing a criminal or tortious act. States may be stricter. California, Florida, Pennsylvania and Illinois require the consent of all parties. Other all-party states not listed here include Maryland, Massachusetts, Montana, New Hampshire and Washington. Texas, New York, Ohio, Georgia and North Carolina follow the one-party standard. Michigan's statute reads as all-party consent on its face. California and Illinois limit their statutes to conversations where the parties reasonably expect privacy. These rules turn on consent, not on whether the recorder appears in the participant list, so a silent local recorder faces the same test as a visible bot. This is general information, not legal advice.

What is the most secure AI note taker?

No honest ranking exists, because security here collapses several separate questions. Compare four things on the vendors' own pages. First, what a no-training promise actually covers: Otter states that no customer data trains its third-party AI providers' models, while Otter itself de-identifies user data to train its own models, Fireflies says it does not train on customer data by default, Granola trains on anonymized data by default with an opt out in Settings, and Routines states it does not use recordings, transcripts, notes or prompts to train models. Second, whether retention is bounded: Fireflies publishes a floor of at least 12 months, Granola keeps notes indefinitely unless someone sets a policy, and Otter purges deleted conversations from Trash after 30 days without stating a window for active transcripts. Third, that audio and transcripts are separate assets, since Granola deletes audio right after transcription but keeps the notes. Fourth, whether a SOC 2 report is self-reported or one you can actually read. Granola, Otter and Fireflies each claim SOC 2 Type II on their own pages, and Routines does not state a SOC 2 status.

Does Apple have an AI note taker?

Partly. Notes on Mac records audio and transcribes it inside the note, and you can view, search in and copy the live transcript by double-clicking the recording. Apple lists this as requiring a Mac with M1 or later and the device language set to English, Spanish, Portuguese, Italian, French, German, Japanese, Korean, Simplified Chinese or Traditional Chinese. Summarizing that transcript is a separate feature. You open the note, click the recording, then click Summary, and Apple lists the requirements as the latest version of macOS, Apple Intelligence turned on and a Mac with M1 or later, described as available in beta in specific languages and regions. Writing Tools adds Summary, Key Points, List and Table options across Notes, Mail and most places you type. Apple's transcription page does not mention speaker labels or attendee lists, so a native Notes recording gives you one undifferentiated transcript rather than meeting notes that separate who said what.

Is there a free AI note taker?

Several, and the limits come in different currencies. Granola's Basic plan is free and includes AI meeting notes and AI chat, but meeting history is limited and the pricing page gives no number. Otter's free Basic plan gives 300 transcription minutes per user per month, caps each conversation at 30 minutes, allows 3 lifetime file imports and 20 AI chat queries a month, and keeps only your 25 most recent conversations. Fireflies has a free forever plan with unlimited transcription and AI summaries, capped by 400 minutes of storage per team, 20 AI credits and a 2 hour maximum recording, with no transcript or recording downloads. Routines has a free plan with up to 3 active routines and a one-time €5 of AI credit, plus a 7-day Pro trial that adds unlimited routines and 50 hours a month of included transcription before dropping back to Free. Last checked 2026-08-10.

Where do AI note taker transcripts get stored?

Granola's notes sit in its US-hosted AWS Virtual Private Cloud, encrypted at rest and in transit. Otter stores conversations in its own cloud on AWS S3 with server-side encryption enabled. Fireflies is cloud-hosted by default, and Enterprise customers can add Private Storage to choose the location, which relocates the bucket rather than the processing. Routines keeps notes, transcripts and summaries as markdown and a SQLite database on your Mac, copied to your Routines account only if you turn on Cloud Sync and Backup, and the meeting audio itself streams to Deepgram to be transcribed.

Try the no-bot, local option

If the part that pulled you in was meeting notes that stay as files on your Mac and no bot in the call, that is the whole point of how Routines records. Run it against your next few real meetings and see how the speaker-separated transcript and the one-page summary hold up.

Download Routines and record your next meeting locally, on a 7-day trial, with your own model.