You finish a call, the transcript lands in your notetaker, and three weeks later you cannot remember which meeting held the decision you need. So you scroll. Here is the verdict before the details: Granola, Otter and Fathom all transcribe accurately, but they are not the same product. What separates them is not transcript quality. It is whether you can ask one question across every past meeting and get a real answer, and what each one does with your recordings after the call ends.
Most comparisons rank these tools on summary accuracy and speaker labels. That is the easy part, and all three are good at it. The thing that actually decides whether a notetaker still earns its place six months from now is memory: can it recall what you decided, not just store the words you said? On that axis the three pull apart fast. Price tells you almost nothing about which one recalls best, and the privacy defaults are not what their marketing pages suggest.
The short answer: pick by recall and privacy default, not by transcript quality
Want a notetaker that never adds a visible bot to your call? Granola is the clear pick. It records system audio from your own device instead of joining as a participant. Want the strongest history when you record a high volume of meetings? Otter is built around that, with unlimited conversation history on its paid tiers. Want real unlimited free recording? Fathom gives you unlimited recording and transcription on its $0 plan, where Otter caps you at 300 minutes a month. The catch sits underneath all three: each keeps your meeting memory inside its own app, and each trains on your data by default unless you go switch it off.
Granola: bot-free capture, but training is on by default
Granola does not send a bot into your meeting. It captures the audio playing through your computer, transcribes it to text, then writes structured notes, so other participants never see an extra attendee join the call. You start the recording yourself before each meeting; it does not auto-join. That design is the whole appeal for anyone who finds it awkward when an AI notetaker shows up in the participant list, and it sidesteps the all-party-consent problem that meeting bots create.
Pricing, as of June 2026: a free Basic tier at $0 with limited meeting history, a Business tier at $14 per user per month with unlimited history plus AI chat across your meetings, and an Enterprise tier at $35 per user per month. Several independent reviews peg the free history cap at roughly 25 notes total rather than a rolling window, so heavy users hit the wall in about five weeks. Business adds integrations to Notion, Slack, HubSpot and Zapier, and the AI chat over your past meetings, the closest thing Granola has to cross-meeting memory, lives on the paid tiers.
Now the part the homepage does not lead with. Per Granola's own privacy policy, Granola trains its models on your anonymized meeting data by default on the Basic and Business plans. You can opt out, but each person has to flip the switch themselves in settings; the organization-wide opt-out is gated behind the $35 Enterprise tier. Granola does not let third parties like OpenAI or Anthropic train on your data, which is a meaningful limit, but the bot-free design does not mean training-free.
Otter: built around your conversation history, dogged by a lawsuit
Otter treats your back catalogue of meetings as the product. You can search by keyword across conversations and ask questions of your meeting history, which earns its keep when you need a decision from months ago. The free Basic plan caps you at 300 transcription minutes a month, keeps only your 25 most recent conversations, and allows just 3 file imports for the life of the account. The long memory that makes Otter worth it really lives on the paid tiers, where history is unlimited.
Paid pricing, as of June 2026: Pro at $16.99 per user per month, dropping to $8.33 on annual billing; Business at $30 per user per month, dropping to $19.99 on annual billing; Enterprise on custom pricing. Otter uses a meeting bot that can auto-join your calls. That convenience is also where its biggest controversy lives, and it matters if your meetings touch anything sensitive.
On privacy, read the fine print. Otter's policy states it trains models on de-identified user recordings and transcripts. In August 2025 a federal class action, Brewer v. Otter.ai, was filed in the Northern District of California by Justin Brewer, who was not even an Otter user. As NPR reported, the suit alleges OtterPilot recorded private conversations without the consent of all participants and used those recordings to train its AI; it cites the Electronic Communications Privacy Act and California's all-party-consent recording law. Otter says it de-identifies data before training. If meeting confidentiality is a hard requirement, weigh that openly.
All three notetakers store your meeting memory and train on it at the same time, by default. Those are two very different deals, and the opt-out switch is never the one you find first.
Fathom: the only real unlimited free recording, but watch the AI summary cap
Fathom's pitch is the free tier, and on raw recording it delivers. At $0 you get unlimited recordings and transcriptions plus search across your calls, with no monthly minutes cap. That is a real edge over Otter's 300-minute free ceiling. The catch most listicles skip: the free plan limits advanced AI summaries to 5 calls per month, after which you drop to a basic chronological template. The conversational assistant that lets you query across calls also sits on the paid plans.
Paid pricing, as of June 2026: Premium for individuals at $20 per month, or $16 per month billed annually. Team starts at $19 per user per month, or $15 annually, and Business runs $34 per user per month, or $25 annually, both with a two-user minimum, plus a custom Enterprise tier. Fathom uses a meeting bot to record by default, though it now offers a bot-free capture mode in beta on Mac. On training, Fathom says it uses de-identified customer data to improve its own models, with an opt-out in settings, and it contractually bars sub-processors like Anthropic, OpenAI and Google from training on your data.
Side by side: bot, price, memory and training default
Here is the comparison that matters once you stop arguing about transcript accuracy. All prices are as of June 2026 and change often, so confirm on each vendor's own pricing page before you buy.
| Dimension | Granola | Otter | Fathom |
|---|---|---|---|
| Records via | Bot-free, captures device audio | Meeting bot, can auto-join | Meeting bot (bot-free beta on Mac) |
| Free tier limit | Limited history (~25 notes) | 300 min/mo, last 25 chats, 3 lifetime imports | Unlimited recording; 5 AI summaries/mo |
| Paid entry price | $14/user/mo (Business) | $8.33/user/mo (Pro, annual; $16.99 monthly) | $16/mo (Premium, annual; $20 monthly) |
| Cross-meeting memory | AI chat over history (paid) | Keyword search + unlimited history (paid) | Search + AI assistant across calls (paid) |
| Trains AI on your meetings | Yes, default; opt-out per user | Yes, de-identified by default | Yes, de-identified; opt-out in settings |
| Best for | No-bot capture, manual control | High meeting volume, history search | Free unlimited recording |
Before you standardise a team on any of these, check the default training setting and the participant-consent behaviour, not just the price. Those two settings, not the summary quality, are what land companies in front of a lawyer.
Match the tool to the meeting, not the meeting to the tool
The cleanest way to choose is by how you actually meet. If you run external sales or client calls where a visible bot reads as intrusive, Granola's device-audio capture keeps the call feeling human, and the $14 Business tier buys you the AI chat over history you will want once notes pile up. If you are an internal team running back-to-back calls and your real need is finding the one decision buried three months deep, Otter's unlimited paid history and keyword search across every conversation is the strongest fit, with the lawsuit and training default as the cost of entry. If you are a solo operator or a small team watching spend, Fathom's free unlimited recording is hard to beat, as long as you can live inside the 5-summaries-a-month ceiling or pay $16 a year for Premium to lift it.
One detail that gets lost in feature checklists: cross-meeting recall is only as deep as the history the plan keeps. Otter's free tier holds your last 25 conversations, so a question about a call from two months ago may simply have nothing to search. Granola's free history cap means older notes can age out before you think to ask. Fathom keeps the recordings, but on the free plan the conversational assistant that would answer across them is paywalled. The promise of asking your meetings a question is real on every paid tier and thin on every free one.
What most comparisons will not tell you: the memory is trapped in the notetaker
Here is the gap none of the three closes. Your meeting memory lives inside the notetaker that recorded it. Open ChatGPT or Claude or Gemini to draft the follow-up, write the spec, or prep the next call, and that assistant knows nothing about what you decided unless you copy and paste it back in. You captured the memory in one app and you need it in another, and the two do not talk. Cross-meeting search is genuine progress, but it still only works inside the tool that happened to be running that day.
So the honest answer to which one remembers is this: each remembers inside its own walls. Otter searches Otter. Granola chats over Granola. Fathom queries Fathom. The decision you made six months ago is findable, but only if you return to the specific product that recorded it, and only if it survived that product's free-tier history cap. None of these tools makes the memory portable, and none lets the AI you actually write in reach back into it.
Where MemX fits: meeting memory your AI can actually use
Meeting memory should not be locked inside a single notetaker. MemX is a private-by-architecture memory layer that sits next to the AI assistant you already use, so the context you capture in one place is recalled everywhere you work. Keep recording in whichever notetaker fits your call style, then let your assistant pull the decisions, owners and deadlines into the next draft without you pasting transcripts back and forth.
On the privacy point all three notetakers stumble over, MemX takes the opposite default. It is built with per-user isolation and encryption at rest, with CMEK, and your memory is not used to train models. That is the reverse of a notetaker that learns from your meetings unless you find the off switch: capture in any tool, recall everywhere, without handing your conversations over as training data.
Frequently asked questions
01Which is best for searching past meetings?
Otter is built around searching your conversation history by keyword, with unlimited history on its paid tiers. Granola's AI chat and Fathom's assistant also search across calls but sit on paid plans. As of June 2026, confirm current limits on each vendor's pricing page.
02Does Granola put a bot in my meeting?
No. Granola is bot-free. It captures the audio playing through your own device and transcribes it, so no extra participant appears in the call. You start the recording manually before each meeting; it does not auto-join like Otter or Fathom can.
03Does Otter train its AI on my meetings?
Yes, by default. Otter's policy states it trains models on de-identified recordings and transcripts. A 2025 class action, Brewer v. Otter.ai, challenged its consent practices. Check your account settings and the privacy policy before recording sensitive calls.
04Which notetaker has the best free tier?
Fathom, for raw volume: its $0 plan includes unlimited recording and transcription with no minutes cap, though advanced AI summaries are limited to 5 calls a month. Otter's free plan caps you at 300 minutes and 25 conversations. Verify as of June 2026.
05Can these notetakers feed my ChatGPT or Claude?
Not natively. Granola, Otter and Fathom each keep meeting memory inside their own app, so your AI assistant only sees it if you paste it in. A separate memory layer like MemX carries that context across the tools you actually work in.
The takeaway
All three transcribe well, so stop choosing on that. Pick Granola if you want no visible bot and you will manage recording manually, Otter if you live in your meeting history and need the strongest search, Fathom if you want genuinely unlimited free recording. Then accept two real limits: each one trains on your data by default until you opt out, and the memory stays trapped in whichever notetaker recorded the call. To make a meeting useful inside the AI you draft and plan in, you need a memory layer that travels with you, not another silo.
