Granola Review: The Bot-Free Meeting Notepad That Actually Replaced My Scratchpad

What it’s actually good at

The bot-free capture is the feature that quietly changes the texture of every call you take. Granola pulls audio straight from your computer’s system output, so no bot joins the meeting and nobody on the other end has any way of knowing you’re recording. That’s the single biggest gap between Granola and Fathom, Fireflies, tl;dv, or Otter.ai, all of which still walk into the call as a visible participant. Sounds like a small thing until you’ve sat through a discovery call where the prospect spent the first three minutes asking what “Fireflies Notetaker” was and whether they could turn it off. With Granola, the question never comes up. It’s running on your laptop the way a voice memo would.

The hybrid workflow took a few calls to click, and then I couldn’t unclick it. You get the usual meeting-recorder machinery plus a live scratchpad you can type into during the call, and afterwards the AI fills in the surrounding context. Your enhanced “scratchpad notes” land in black; the auto-generated stuff fills in around it in light gray. The practical effect is that the finished doc reflects what you thought mattered, not what a generic summarizer thought mattered. You jot the observations, the priorities, the reactions that actually hit you in the moment; the assistant fills in the rest. So the synopsis carries your emphasis instead of flattening every meeting into the same shape. It’s the first AI notetaker I’ve used that solves the “yes, but it missed the only thing I actually cared about” problem.

The 2026 expansion is what bumps the score up. This isn’t the prosumer Mac app it was eighteen months ago. With the Series C, Granola shipped Spaces (team workspaces with granular access controls) plus a personal API and an enterprise API for piping meeting context into broader AI workflows, an MCP server in February 2026, and a customer list that now includes Vanta, Gusto, Asana, Cursor, Lovable, Decagon, and Mistral AI. The MCP server is the part that matters most for daily life. It lets you yank Granola meeting context straight into Claude or any other MCP-compatible tool, so if you’re drafting a follow-up proposal in Claude and want to reference what was said across three discovery calls, you query it inline without leaving the doc. Pair that with the cross-meeting chat and you stop using Granola like a transcript service and start using it like a memory layer.

The pricing rework is, honestly, the unsung win of the year. The lineup runs from a capable free plan up through a $14/user/month Business plan and a $35+/user/month Enterprise tier. Enterprise comes with the SOC 2 Type 2 certification Granola finished in July 2025, and they hit compliance in three months instead of the usual 12 to 18 because the architecture deletes audio the moment it’s been transcribed. That’s a real engineering decision showing up as a real pricing decision: there’s no bot infrastructure to pay for, so Granola can undercut the bot-based competitors while shipping a tighter privacy story.

Where it lets you down

The 25-note free plan cap is the first wall every new user hits, and the framing is misleading enough that it deserves a mention up front. Basic limits your meeting history to 25 notes total. Not 25 a month. Twenty-five, ever. You hit it and you’re either deleting history or upgrading. Five client calls a week gets you there in five weeks. Ten a week and you’re done in under three. The free plan is a real evaluation tier, but it’s not the “use it forever for free” tier that the word “Basic” implies. Plan around that or get surprised.

Speaker ID is the next rough edge, and it’s the reason Granola isn’t a 95. Transcription lands at 90 to 95 percent in clean audio with one speaker at a time, but multi-speaker accuracy drops noticeably in calls with three or more participants. Speaker identification holds up in 1:1s and small calls, then falls apart in bigger rooms, and Otter still leads on raw accuracy and attribution. If your day is 1:1 client calls and small team syncs, you’ll barely notice. If your day is six-person product meetings where you need to know who said the controversial thing, you’ll be fixing attributions manually, or you’ll be using Otter.

The “no audio playback” thing is a feature, not a bug, but be clear-eyed about it. Granola captures text only. It transcribes in real time and never stores audio or video. For a lot of people on the other end of the call, that’s the entire reason this works: they don’t want their voice or their face sitting on someone else’s server. CEO Chris Pedregal has been blunt that the value is in useful notes, not in retaining audio. The tradeoff is real: you can’t pull up the file and verify a quote, and you can’t hear the emotion in someone’s voice. For most knowledge work that’s the right call. For journalism, depositions, or anything where tone is half the data, look elsewhere.

The last thing to flag is what Granola has quietly become. The company isn’t pitching itself as a notetaker anymore; the current framing is “enterprise AI context layer.” Translation: Granola wants to be the central store for your team’s meeting data, with APIs and MCP wiring that pipe it into Claude, ChatGPT, and whatever else your team is using. That’s mostly upside if you’re a power user. It also means the parts of the product that still feel rough (speaker ID, post-meeting workflow automation, the lack of native Google Doc sync) are exactly the parts the company isn’t optimizing for right now.

Should you pay for it?

If you take meetings for a living, yes. The 2026 lineup is three tiers: Basic (free) with AI meeting notes and limited history, Business at $14 per user per month with unlimited history and integrations to Notion, HubSpot, Slack, and Zapier, and Enterprise at $35 per user per month with SSO, API access, and team-wide opt-out of model training. Spend a week on Basic to make sure the bot-free workflow clicks for you (it might not, especially if you don’t type notes during calls) and then jump straight to Business once you’ve decided. Don’t try to live on Basic forever; the 25-note cap will catch you.

Business is the tier the product is built around. For $14 a seat you get the full team experience: shared folders, native CRM sync to HubSpot, Affinity, and Attio, and the integrations you need to build real institutional memory across customer calls, hiring loops, and pipeline reviews. If you’re running a team of three to fifty people who all live in customer calls, this is the configuration. It’s tuned for teams in exactly that range running frequent meetings and wanting notes to flow into existing tools automatically.

Enterprise is for the specific situations Business can’t cover. At $35 a seat it adds governance for compliance-heavy orgs or large deployments: SSO with Okta, Azure AD, or Google Workspace, organization-wide model training opt-out enforced by default, admin controls over meeting link sharing, priority support, usage analytics, and public API access. If you need SSO, top-down AI training opt-out, or you’re in a regulated industry where per-user settings won’t fly with legal, Enterprise earns its keep. Most teams won’t need it; the ones that do will know.

The bottom line

Granola is the AI meeting notepad every other tool is now measured against, and most of them are quietly copying it. The bot-free capture changes how clients perceive the call, the hybrid type-then-enhance workflow changes how present you can be in it, and the MCP plus cross-meeting chat changes what your meeting history is actually for. The speaker ID gap, the missing audio playback, and the stingy free plan are real, but they’re the price of an architecture that’s deliberately simpler and more private than the bot-based competition. If you take customer calls or 1:1s for a living and you’re still letting a bot announce itself in every meeting, switching to Granola is the easiest productivity upgrade you can make this quarter. It’s the one to beat, and it earns the Editors’ Choice.