Granola Review (2026): It's Not a Notetaker, and That's the Whole Point

Short answer - should you use Granola?

• Use it if you type during meetings anyway. The whole design assumes you're an active participant with opinions, and it's better than anything else at merging what you meant with what was said.
• Use it if you need one tool across macOS, Windows, iPhone and Android, plus an Apple Watch app, which is four more platforms than I ship.
• Use it if your notes need to land in HubSpot, Attio, Affinity, Notion or Slack without you doing data entry. This is genuinely the product, not a checkbox.
• Skip it if what you actually want is a searchable record of everything anyone said for the last two years. It'll do it on Business, but you're buying a notepad and using it as a filing cabinet.
• Skip it if the audio can't leave the building. There's no on-device mode and no bring-your-own-key, and all data sits on AWS in the US with no EU residency option (granola.ai docs, 1 September 2026).

Bias: I make Humla, which competes with Granola, and I'd rather you downloaded mine. Every price, quote and date below is off Granola's own live pages on 1 September 2026, and they're linked so you can go check.

I have to review the category before I can review the product

Here's the thing that makes Granola annoying to review: it refuses to be the thing everyone compares it to.

Go look at granola.ai/ai-note-taker, which is a page on their own marketing site, with a headline that reads "More than just a standard notetaker" and a sentence underneath it that says, and I quote, "Most AI notetakers record what was said and call it done. Granola is an AI notepad."

They didn't back into this framing under pressure from a competitor. They coined it. My own glossary entry for notetaker vs notepad credits them with coining it, which is a slightly funny sentence to have on my own website.

And a notepad and a notetaker are different machines with different jobs:

• A notetaker exists so a meeting becomes a record. Success is coverage. Did it catch everything, can I find it later, can I prove what was agreed.
• A notepad exists so a meeting becomes your thinking. Success is fidelity to intent. Did the output reflect what I actually cared about in the room.

You can build one app that does both, badly. Granola picked one and went at it with unusual discipline, and that discipline is what generates every single thing people complain about in reviews. Which is why I want to do the pricing next, and then immediately argue with the pricing.

What it costs

From granola.ai/pricing and the billing docs, 1 September 2026:

Billing is per workspace, so you can genuinely be on Basic in your personal workspace and Business at work. Startups and students get 12 months of Business free, which is a nice touch that I'm mildly annoyed I can't afford to copy.

The 30-day thing, which everyone reads wrong

Every "Granola alternatives" listicle - including the one I wrote in July, where I called the free tier "capped" and moved briskly on - treats the 30-day limit as the usual freemium screw-tightening. Give them a taste, take it away, upsell.

I went looking for the actual number, because the plan for this post literally had a line asking whether Granola states it anywhere in digits, and they do: notes older than 30 days are "not deleted - we keep them stored - but they are not accessible in Granola," and a CSV export of everything, including the older stuff, is available on the free plan and gets emailed to you.

Sit with that for a second. They keep your notes. They don't show them to you. They'll hand you the whole lot as a file whenever you ask.

That is not a data hostage situation. That's a product saying the last month is the working set, and everything before it is archive, and archive is the paid product. Which is exactly what you'd expect from a notepad. Nobody buys a paper notebook and complains that page 400 requires buying a second notebook.

Do I think it's also good business? Obviously. It's a beautifully aligned upsell, the kind you sketch on a whiteboard and then can't believe you get to keep. But "commercially convenient" and "philosophically consistent" aren't mutually exclusive, and I've spent enough time being cynical about competitors' pricing to notice when the cynical reading is just lazier than the real one.

The setting that proves it

Okay, this is the part of the review where I stop being balanced and start being a fan for a paragraph.

Granola will delete your transcripts on a timer. Settings, Preferences, Data & sharing: pick 1 day, 1 week, 1 month, 3 months, 6 months or 1 year. It's available to individual users, not just Enterprise. There's a one-week cooldown before the first deletion actually happens, so a misclick at 11pm doesn't eat your quarter.

And here's the load-bearing detail: it deletes the transcript and keeps the notes.

The docs are refreshingly grim about the consequences. You can't regenerate notes for that meeting afterwards. Chat quality "will be degraded" because Chat is reading the transcript. Deletion is permanent and irreversible, in a <Warning> box, in bold.

Think about what shipping that feature requires you to believe. You have to believe the transcript is scaffolding - the thing that helped produce the note, disposable once the note exists. A tool that thought of itself as a record of what was said could never offer this, because the record IS the product; deleting it is deleting the thing you sold. Otter does not have this switch. I went and looked.

I don't have this switch either, incidentally. Humla keeps the transcript because Humla thinks the transcript is evidence - it's the thing my chat feature searches and the thing you re-read when a client remembers a deadline differently than you do. Two apps, opposite instincts, both defensible. Granola's is the braver one.

(There's an "off" option too, and I'd bet a reasonable sum that "off" is what almost everybody runs, because approximately nobody opens Preferences. The default matters more than the feature. But the feature existing tells you what the company thinks it's building, and that's what I'm reading it for.)

Where the thesis starts wobbling

Right. I've been nice for four sections. Let me be a competitor for one.

If Granola is a notepad and the archive is somebody else's problem, then what exactly is all this?

• Chat across meetings - ask questions spanning your whole history.
• An MCP server so Claude and ChatGPT can query your notes. Hosted endpoint, you sign in through the browser.
• A REST API with keys, for programmatic access to notes, transcripts and summaries.
• People and Companies, which automatically builds a directory of humans out of your meetings and calendar.
• Pre-meeting briefs, generated overnight, pulling "open threads" from prior conversations.
• Unlimited history, sold as the headline reason to pay.

That is a corpus. Every one of those features is only worth building if there's a big searchable pile of past meetings sitting behind it, and several of them are only worth paying for if the pile goes back years. Their own marketing page pitches "ask questions across your last 20 meetings" while the pricing page next door sells unlimited history as the upgrade, and I can't fully reconcile those two sentences.

I don't think this is hypocrisy. I think it's what happens to every successful tool: you nail one job, your users get deep enough into it that they start needing the adjacent jobs, and the roadmap follows them. Notepad-that-remembers-everything is a perfectly good product. It's just no longer cleanly a notepad, and the "we're not a notetaker" page starts doing more marketing work than descriptive work.

Watch this space, basically. If the 30-day free tier ever becomes "unlimited but ad-supported" or the transcript auto-deletion feature quietly loses its 1-day option, that's the thesis formally ending.

Anyway. Back to reviewing the software that exists today rather than the one I'm speculating about.

What's genuinely good, including one thing I got beaten to

No bot, confirmed in the docs, not just the marketing. "There is no meeting bot - Granola runs only on your computer and uses your system audio and microphone." It captures the mic and the system output at the endpoint, exactly like Humla does, and no phantom participant appears in anyone's Zoom sidebar.

The audio is not kept. "It does not record or save audio or video at any point during the call, so there's no way to access audio from your meetings." Audio goes to their transcription provider, comes back as text, and that's the end of it. This is a real privacy property and a real limitation in the same sentence: you cannot re-listen, ever, and if the transcription mangled a name you're out of luck.

The note quality is the best in the category. I'm not going to pretend to have run a rigorous eval - I haven't, and anyone who tells you they have with n=6 meetings is selling something - but the augment-what-you-typed approach produces output that reads like a person was present, because a person was. When you write "!! pricing objection" mid-call, the summary knows that mattered. Mine works the same way, and I still think theirs is better tuned.

And then there's the thing I have to hand them.

A week ago I published a post about the notetaker lawsuits whose central argument was something I called the Account Holder: every consent switch, opt-out and privacy setting in this entire category belongs to the one person who pressed record, and never to the other people in the room. The client. The candidate. The person who dialled in for nine minutes. Structurally unrepresented, all of them.

Granola has shipped transparency features. Two of them. An automated message posted into the meeting chat when transcription starts ("Hey, I'm using www.granola.ai to transcribe and take notes"), and a Granola Watermark, which is a virtual camera that puts a visible badge on your video for the entire call. Available to individuals on the free and Business plans, deployable fleet-wide by admins over MDM, and on Enterprise there's an org-wide notice setting too.

That is aimed squarely at the hole I spent 3,000 words describing. It's not a complete answer - the Account Holder still switches it on, the other people still don't get a vote, and a watermark is notice rather than consent - but it's a real, shipped, non-trivial attempt at telling the other people in the room, and it's more than I've built. Which is nothing.

(Granola is also a defendant in Chamberlain v. Granola, Inc., No. 3:26-cv-07926, N.D. Cal., filed 30 July 2026 - unproven, denied, and covered in that other post. I won't relitigate it inside a product review, except to say the transparency features cut against the theory of the complaint rather than for it.)

What to actually know before you commit

The unglamorous list, all from their docs on 1 September 2026:

On training: opt-out is available on every plan, not just Enterprise, which is worth saying clearly because my own /vs/granola page claimed otherwise until this morning. Enterprise gets it enforced org-wide and on by default, which is a different and smaller thing.

The combined-audio one deserves a beat, because it's the sort of thing you only discover in a bad way. Granola captures everything your Mac is playing, mixed. Humla keeps mic and system as two separate streams end to end, which is why it can tell you-versus-them apart structurally rather than by inference. Neither approach is wrong; one of them puts your Spotify in the meeting minutes.

And I'll flag a small documentation snag I hit while checking retention: the security FAQ says configurable transcript auto-deletion is an Enterprise feature, while the dedicated auto-deletion page walks individual users through setting it themselves. The specific page is almost certainly the current one and the FAQ is lagging behind a shipped feature. I mention it only because if you're evaluating on the FAQ alone you'll conclude you can't have something you can have.

So: Granola or Humla

The split, thumb visibly off the scale:

Pick Granola if you're a normal professional with a laptop, a phone and a CRM, who wants the notes to appear in the tools where work happens, and who is fine with a US cloud holding the notes. That's most people, and it's not a compromise, it's the correct answer for that person.

Pick Humla if the meeting contains material that genuinely cannot leave your machine, or you want to read the code, or you're paying for a team and $5 versus $14 a seat times twelve months times everyone is real money.

I lose this comparison on convenience and win it on control, which has been the shape of every comparison I've written and will probably be the shape of the next one. If you want the full grid there's the side-by-side and the switching guide, both of which go deeper than a review has room for.

Verdict

Granola is the most coherent product in this category. Not the most featureful, not the most private, and increasingly not the most focused - but the one where you can look at any given decision, including the annoying ones, and reconstruct the belief behind it. The 30-day limit, the disposable transcripts, the refusal to import a file: those all fall out of one idea held firmly.

My worry for them isn't a competitor. It's that the idea is getting expensive to hold. Every enterprise deal wants the corpus, every corpus wants retention, and retention is the notetaker's business model wearing the notepad's clothes. The transcript auto-deletion switch is the canary. As long as the 1-day option is in that dropdown, somebody at Granola still means it.

---

Prices, quotes and behaviour checked against granola.ai and docs.granola.ai on 1 September 2026; their privacy policy is effective 24 July 2026. All of it first-hand this time - those domains were blocked when I wrote the MCP post in August. Litigation references are unproven allegations, covered separately in the lawsuits post. Other reviews: Otter.ai.