Published July 19, 2026
Here is a number that should change how you pick a meeting notetaker. In a Kolmogorov Law survey of 500 US workers published this week, 33.4% said AI notetakers have sat in on their meetings. Of those, only 34.7% said they were always asked permission first. A quarter, 25.1%, were never asked at all. And 18.8% found out a recording existed only after the fact, which is a polite way of saying they were recorded in secret and learned about it later.
That is not a feature gap. That is a trust problem, and it now has legal teeth: the same report notes a federal class action against Otter.ai alleging it recorded conversations without all-party consent, which California law requires. Bloomberg flagged the same consent problem at the end of June. So this ranking does something our note-taking apps roundup deliberately does not: it ranks Granola, Otter, Fireflies, and Fathom on privacy and consent behavior first, and note quality second. One thing before we start. This is an assessment of how each tool is documented to behave, in vendor materials, court filings, and reporting, not a claim that we wiretapped our own meetings four different ways to check.
How we ranked them
Three criteria, weighted in this order:
- Capture method and disclosure. Does the tool join calls as a visible bot that everyone can see, capture audio locally with no bot at all, or record in ways participants may not notice? Visibility is a consent mechanism. Invisibility is a data-handling advantage that puts the disclosure burden on you. Both models can be used well; only one fails silently.
- Where the recording lives and who controls it. Local capture versus cloud storage, and whether admins can set retention and recording policies across a team.
- Track record. Litigation, regulatory attention, and documented complaints count against a tool here, whatever its marketing says.
The ranking
Granola
★★★★☆4.5 Best data handlingGranola takes the opposite architectural bet from everyone else on this list. Instead of sending a bot to join your call, it records locally on your device with no visible meeting bot, transcribing what your machine already hears. No third wheel in the participant list, no recording sitting on a vendor's meeting bot infrastructure. Investors bought the argument: a $125 million Series C in March 2026 valued the company at $1.5 billion, and the product has grown up around that core with Granola Chat, an MCP server, and an enterprise API that gives IT actual integration points rather than a consumer app wearing a suit.
The honest caveat, and the reason this is a 4.5 and not a 5: local capture solves the data question and does nothing for the disclosure question. A bot in the participant list is an imperfect consent notice, but it is a notice. Granola provides none automatically, so a Granola user who says nothing is exactly the "found out after the fact" scenario from the survey. First place here assumes you announce it. Read the checklist below and mean it.
Fathom
★★★★☆4.0 Best built-in disclosureFathom's bot joins the meeting where everyone can see it, and that visibility does real consent work for you: nobody on a Fathom-recorded call can honestly claim they had no way of knowing. For client-facing meetings, where a surprise recording is a relationship-ending discovery, that built-in disclosure is worth more than any summary feature.
The tradeoff is the mirror image of Granola's. Your recordings live in Fathom's cloud rather than on your device, so the data-handling question shifts from "who was told" to "who holds it and for how long," and that answer depends on your plan and your admin settings rather than on architecture. Second place: strongest default disclosure, weaker data custody story than local capture.
Fireflies.ai
★★★☆☆3.5Fireflies works the same visible-bot way as Fathom, and its admin controls over recording and sharing are a genuine plus for a company trying to set one policy instead of forty personal ones. The reason it sits third is scale of exposure rather than any single documented sin. Fireflies bots are among the most common uninvited guests in other people's meetings, the classic case of a notetaker auto-joining every calendar event its owner ever touched, including the ones they declined. That is a configuration problem, and it is fixable in settings. It is also exactly how an organization ends up an accidental data point in that 25.1% who were never asked. If you deploy Fireflies, turning off auto-join for external meetings is the first thing to do, before anyone writes a prompt.
Otter.ai
★★☆☆☆2.5Otter is the veteran of this category and, on pure transcription pedigree, has earned its place in plenty of rankings, including ours. This is not one of them. The federal class action alleging Otter recorded conversations without all-party consent under California law is exactly the failure mode this article exists to rank, and an active suit on that specific question puts Otter last on a privacy-weighted list almost by definition.
To be fair about what an allegation is: a class action is a claim, not a verdict, and Otter disputes plenty. But a privacy-first buyer choosing among four tools does not need to wait for a judgment to prefer the three that are not currently defending their consent practices in federal court. If your team is on Otter today, the practical move is not necessarily to rip it out. It is to lock down auto-join and recording defaults at the admin level, this week.
At a glance
| Tool | Capture | Participants see it? | Biggest privacy risk |
|---|---|---|---|
| Granola | Local, on-device | No, disclosure is on you | Silent capture if the user never announces it |
| Fathom | Cloud, visible bot | Yes | Vendor-held recordings, retention set by plan |
| Fireflies | Cloud, visible bot | Yes | Auto-join spraying bots into meetings uninvited |
| Otter | Cloud, bot or in-app | When the bot is used | Active federal class action over consent |
The consent checklist
Whatever tool wins your evaluation, the survey numbers say the real failure is human, not technical. Only about a third of exposed workers were always asked first. Fixing that costs one sentence per meeting.
- Announce before recording, every meeting. "I use an AI notetaker, is everyone OK with that?" Eleven words. Say them even when your tool has a visible bot, and especially when it does not.
- Get affirmative agreement with external parties. California and several other states require all-party consent for private conversations. Silence from a client is not consent, and the Otter suit shows what the downside looks like.
- Kill auto-join for meetings you don't own. The uninvited bot in someone else's call is how most of these horror stories start.
- Offer to pause. When a conversation turns personal or sensitive, stop the capture and say you did. People remember.
- Know your retention settings. A recording that auto-deletes in 30 days is a smaller liability than one that lives forever. Set the policy at the admin level so it does not depend on individual diligence.
Verdict
Granola wins on architecture: local capture with no vendor bot in the call is the strongest data-handling posture in the category, and its enterprise API means IT can actually govern it. But it wins conditionally, because its invisibility only stays ethical if you disclose out loud. Fathom is the pick if you want the tool itself to do the disclosing. Fireflies is workable with its auto-join reined in. And Otter, whatever the courts eventually decide, is the one we cannot rank above tools that are not being sued over the exact behavior this list measures. For features, pricing, and note quality across the wider category, our full note-taking roundup has that covered. This page had one job: consent. Most of the industry is still failing it, roughly two meetings out of three.