In August I opened my phone and counted 71 voice memos. I had played back six of them. The other 65 were, apparently, future-me's problem, and future-me was not coping.
Two of those memos were from the same client meeting, recorded eleven minutes apart, because I'd forgotten I'd already made the first one. One was 34 seconds of me swearing at a parking inspector. Four of them I genuinely cannot account for. That is not a filing system. That is a landfill with a microphone attached.
So, naturally, I did the only sensible thing: I ran seven apps side by side. Otter, Granola, Voicenotes, Fireflies, Superwhisper, Notta and Agentsy. Each got a week as my primary tool, and two overlapped slightly because I couldn't get a clean week of meetings for Granola and Fireflies without doubling up. Same phone, same car, same badly ventilated meeting rooms, same accent that transcription engines have been quietly struggling with since 2019.
I judged all seven on six things:
- How fast you can capture. Does it get in the way?
- How good the transcript is on messy speech.
- What it does with the transcript afterwards.
- How it handles a real 45-minute meeting.
- What the free tier actually gets you.
- Whether you'd still be using it a month later.
That last one, I'll be honest, knocked out more apps than the other five combined.
How they scored
Five stars is best. Nobody scored well across the board, which turned out to be the useful finding of the whole month.
| App | Capture speed | Messy speech | After the transcript | 45-min meeting | Free tier | Still using it |
|---|---|---|---|---|---|---|
| Otter | ★★★☆☆ | ★★★★☆ | ★★☆☆☆ | ★★★★★ | ★★★☆☆ | ★★★★☆ |
| Granola | ★★☆☆☆ | ★★★★☆ | ★★★★☆ | ★★★★★ | ★★☆☆☆ | ★★★★☆ |
| Voicenotes | ★★★★★ | ★★★★☆ | ★★☆☆☆ | ★★☆☆☆ | ★★★★☆ | ★★★☆☆ |
| Fireflies | ★★☆☆☆ | ★★★★☆ | ★★★★☆ | ★★★★★ | ★★☆☆☆ | ★★☆☆☆ |
| Superwhisper | ★★★★★ | ★★★★★ | ★☆☆☆☆ | ★☆☆☆☆ | ★★★☆☆ | ★★★★☆ |
| Notta | ★★★☆☆ | ★★★★★ | ★★☆☆☆ | ★★★★☆ | ★★★☆☆ | ★★☆☆☆ |
| Agentsy | ★★★★★ | ★★★★☆ | ★★★★★ | ★★★★☆ | ★★★★☆ | ★★★★☆ |
A note on the prices below: they're what I paid or was quoted during testing, in the app or on the web. Voice tools change their pricing roughly as often as I change my mind about coriander. Check before you buy.
1. Otter
Otter has been the default answer to "what should I use to record meetings" for about six years, and there's a reason for that. It sends a bot into your calls, records, transcribes live, and hands you a searchable transcript with a summary sitting neatly on top.
What I used it for. Everything on Zoom and Google Meet. Otter's bot joined 43 of my calls that month, including two I'd cancelled, which was a mildly awkward way to discover that a calendar decline doesn't always propagate.
What worked. The live transcript is genuinely good, and having it scroll on a second screen during a call changed how I took notes, which is to say: I stopped. Speaker separation held up fine with four people in the room. Search across months of calls is fast and reliable. If somebody says "you told us in June you'd handle the migration", Otter settles the argument in about nine seconds.
What didn't. Otter stops at the transcript and the summary. It will happily hand me the action items as a bulleted list, and then that list simply sits there, being a list. I still wrote every follow-up email myself. Also, the bot is socially visible: clients notice a third participant called "Otter.ai", and some ask about it, which cost me roughly a minute of throat-clearing per new client.
Who it's actually for. Teams whose day is back-to-back video calls, who mainly need a record rather than a to-do list. If anyone in your business ever says "can you send me the recording", Otter is your answer.
Pricing. Free tier with a monthly transcription allowance and a cap of about 30 minutes per conversation. Pro sits around US$17 a month billed annually, Business around US$30 a seat.
2. Granola
Granola is the anti-bot meeting notes app. Nothing joins your call. It listens through your Mac, watches what you type, and afterwards merges your scrappy typing with what was actually said.
What I used it for. Two weeks of investor and partner calls, the kind where I want to look present and not visibly be taking notes.
What worked. The merge trick is clever, and it produces the best summary output of anything I tested. I typed four fragments in one call, one of them just the word "pricing?", and got back a note that correctly explained the pricing objection the other person had raised. No bot in the participant list is worth more than it sounds. Clients never asked about it, not once.
What didn't. It only exists around meetings. When I finished a call, walked to the car and had the actual thought worth keeping, Granola offered nothing, because Granola lives on the laptop I'd just shut. It's Mac-only as of writing, so half the people I'd recommend it to simply can't use it. And nothing happens after the note. Great notes. Still my problem.
Who it's actually for. Desk-based people on Macs who take their own notes and want them tidied rather than replaced. Consultants and founders, mostly.
Pricing. Limited free allowance of meetings, then around US$18 a month for an individual plan, with team pricing above that.
3. Voicenotes
Voicenotes is the closest thing to what I actually wanted when I started this exercise: a fast, quiet place to talk into, that remembers everything and lets me ask it questions later.
What I used it for. Thinking out loud in the car, mostly. Also a dumping ground for the ideas that turn up in the eleven seconds between the shower and getting dressed.
What worked. Capture is instant. Open, talk, done, and the friction is low enough that I used it 60-odd times in a week without ever deciding to. The transcripts held up on messy speech, including one memo recorded with the window down on the M4. Asking it questions across all my notes worked better than I expected.
What didn't. After the transcript, not much happens. There are rewrite templates, and they're fine, but nothing lands in my email or my calendar. On a 45-minute meeting it's out of its depth, which is fair enough, since it never claimed to be a meeting recorder.
Who it's actually for. People whose problem is genuinely just capture and recall. Writers, researchers, anyone who thinks by talking and wants the archive searchable.
Pricing. Usable free tier, paid plans in the US$10 a month range, with a lifetime option floating around during the test.
4. Fireflies
Fireflies looks like Otter from the outside: a bot that joins, records, transcribes, summarises. The difference is what it does with the data afterwards, which is aimed squarely at a sales team's pipeline.
What I used it for. Two weeks of anything resembling a sales conversation, which, for a founder, is more conversations than I'd like to admit.
What worked. The conversation analytics are the real product here. Talk-to-listen ratio, who mentioned a competitor, which objections keep recurring across calls. CRM sync did exactly what it promised into HubSpot, no fuss, and call notes landed against the right contact. For a sales manager coaching five reps, this is a serious tool.
What didn't. For one person, it's enormous. I spent more time in settings during week one than reading output. The dashboards want a team to mean anything, and I don't have five reps, I have a phone and a car. It also does nothing for the between-meetings capture problem that started this whole exercise.
Who it's actually for. Sales teams. Genuinely, only sales teams, and the bigger the team, the better it gets.
Pricing. Free tier with limited credits, Pro around US$18 a seat a month, Business around US$29, enterprise above that.
5. Superwhisper
Superwhisper is not a note-taking app and doesn't want to be. It's dictation that works inside whatever app you already have open. You hold a key, talk, let go, and your words appear where your cursor is.
What I used it for. Writing. Slack replies, long emails, the first draft of two documents, and a fair chunk of an earlier post on this blog.
What worked. Accuracy was the best of anything I tested, including technical words and Australian place names that normally come out as nonsense. It runs locally, which matters if you handle anything sensitive. Latency is low enough that it feels like typing. I am still using it, which, in this piece, is the highest compliment going.
What didn't. It keeps nothing. No archive, no summary, no tasks, no follow-up. Once the text lands in the box, it's your problem, exactly as if you'd typed it. Comparing it to the others is a touch unfair in both directions, because it's solving a different problem, and solving it well.
Who it's actually for. Anyone who writes a lot at a computer and types slower than they think. Pair it with something else for capture.
Pricing. Free tier with local models, one-off and subscription options for the better models, in the region of US$8 a month or a lifetime licence.
6. Notta
Notta is a transcription utility, and it's very good at that particular job. Upload or record, pick from a long list of languages, get an accurate transcript and translations if you need them.
What I used it for. A 45-minute recorded workshop, two phone calls, and one deliberately hard test: a conversation with a friend switching between English and Malayalam.
What worked. Accuracy across languages was the standout. The bilingual test came back better than anything else I tried, by a wide margin. Export options are generous, so it drops cleanly into whatever you use next.
What didn't. It's a utility, not an assistant. The AI summary exists and is serviceable, but nothing about the product wants to help with the work that follows. By day four I was using it the way I use a PDF converter: open when needed, otherwise forgotten. That's why the "still using it a month later" score is low, and it isn't really a mark against the app.
Who it's actually for. Anyone who needs accurate transcription across languages, particularly for interviews, research or compliance records.
Pricing. Free tier with a monthly minute allowance, paid plans starting around US$9 a month billed annually.
7. Agentsy
Disclosure before anything else: I built this one. Read the next few paragraphs with the appropriate suspicion, then read the honest limitations at the end, which is the part I'd check first if I were you.
Agentsy records meetings and takes voice notes like the others, then keeps going. It drafts the follow-up email in your own account, writes the tasks with the dates you spoke aloud, and proposes the calendar entries.
What I used it for. During the week it was my only tool, I ran it across nine meetings and 40-something voice notes, most of them recorded in a car park.
What worked. The bit after the transcript. I said "email Priya the revised scope and tell her the site visit needs to move to Thursday", and the draft was sitting in my Gmail, in my voice, before I'd even got home. Tasks arrived with due dates I never typed. Capture is fast in the same way Voicenotes is fast, and the three modes cover the awkward moments, including the ones where I couldn't talk and typed instead.
What didn't. iPhone only, right now. If your primary capture device is Android, or you live on a Mac all day, this is not your tool yet, and I'm not going to pretend otherwise. It's a newer product than Otter by years, and the integration list reflects that. On the 45-minute meeting test it did fine, but Otter and Granola are still better at the pure meeting-record job, and Granola's summaries beat mine.
Who it's actually for. You're a solo operator or in a small team, you capture between meetings, and you want the follow-up email drafted in your inbox by the time you're back at your desk.
Pricing. Free plan at US$0 with 20 fully drafted capture cycles a person each month, no card. Paid plans at US$5 and US$19 a seat a month, higher on iOS because of Apple's cut.
So which one
Not a summary. A decision framework, based on what you actually do all day.
If you record every meeting: Otter or Granola. Otter if you want the recording and the searchable archive, Granola if you'd rather no bot appeared in the participant list and you're on a Mac.
If you speak between meetings: Voicenotes or Agentsy. Voicenotes if you want a clean archive of your own thinking. Agentsy if the notes keep turning into work you then have to do anyway.
If you want the note to do the work after: Agentsy. That's the whole reason it exists, and if it isn't drafting things you'd otherwise write yourself, use something cheaper.
If you want dictation into any app: Superwhisper, and it isn't close. Pair it with one of the others for capture, because it deliberately keeps nothing.
If you're in sales: Fireflies, especially with a team to coach. The analytics are worth more than the transcript.
If you need accurate multilingual transcripts: Notta.
Two things I'd tell my August self. First, the app matters less than the habit of talking for sixty seconds straight after the conversation ends, while you still remember what was agreed. Second, test on your own worst audio before you pay for anything. Every one of these demos beautifully in a quiet room. My car is not a quiet room.
For what it's worth, I'm the founder of Agentsy. Which is exactly why I've been careful to say when it doesn't win. If a comparison post that includes the author's own product doesn't say what it's bad at, you shouldn't trust the rest of it.