What IT should check in a meeting recorder for Zoom, Teams and Google Meet

How does the recorder capture on each platform?

The first question to ask any vendor is whether it uses a bot, a desktop app or the platform's API on Zoom, Teams and Google Meet, because each method behaves differently and carries different security implications. A bot joins the call as a visible participant and works regardless of which machine the user is on. A desktop app records locally from the user's own device, which avoids bot-lobby problems but depends on the endpoint. Some vendors offer both, which gives IT a fallback if one method is blocked.

Notta covers all three platforms with both a bot and the Notta Desktop app, which makes it easier to standardise on when your organisation meets on Teams but clients and partners use Zoom or Google Meet; Fireflies and Read AI also cover all three platforms both ways. Otter lists a desktop app as well, though its bot-free mode is unverified. Granola takes the opposite approach: it has no bot at all and captures via a desktop app. The platforms' built-in AI — Microsoft Copilot, Google Gemini and Zoom AI Companion — is governed in the admin console you already run, but each is strongest on its home platform only, so a Teams-only deployment of Copilot leaves Zoom and Meet meetings uncovered. For a deeper comparison of these architectures, see built-in vs cloud vs local meeting transcription.

Capture methods by tool (Zoom, Teams and Google Meet)

How will Teams, Meet and Zoom treat its bot?

Teams now holds detected bots in the lobby by default, and Google Meet risk-queues suspicious bots before admitting them, so a recorder that works today can silently break tomorrow. Ask each vendor whether it is part of Microsoft's Bot Identification Program, which labels known bots so they can be admitted automatically, and how users admit the bot when it is held or queued.

This matters for the approval decision in both directions: a bot that gets lobby-held in every external meeting creates support tickets, while a bot that joins unnoticed creates consent problems. Desktop-app capture sidesteps bot detection entirely because nothing joins the call — one reason Notta, which offers both a bot and the Notta Desktop app, is easier to roll out across mixed-platform organisations than a bot-only tool. If your concern is stopping unapproved tools rather than approving one, see how to block AI note takers from joining meetings.

Is SSO and SCIM on the plan you'll actually buy?

Identity controls are often not available below the Enterprise tier, and the seat minimums vary enough to change the shortlist. Otter requires a 100-user licence for SSO and SCIM. Notta, Granola and Fireflies keep SSO and SCIM on their Enterprise plans, with Notta's Enterprise available from 51 seats. Fellow lists SSO (Okta, OneLogin) on its Business plan, and Read AI places SSO and SCIM on Enterprise+.

The practical check is to price the plan you will actually buy, not the plan the vendor advertises, and to confirm the seat minimum in writing before rollout. A tool that looks cheaper per seat can become the most expensive option if SSO forces you onto Enterprise for 200 seats when you needed 40. For ranked picks on identity and access controls specifically, see meeting transcription tools with access controls.

Identity controls by tool and plan

What can admins enforce?

The question is not what users can configure but what admins can force. Check whether the tool supports auto-join rules that distinguish external from internal meetings, recording permissions, sharing defaults, the ability to disable integrations, and IP restriction.

Notta is a useful benchmark here: it offers IP restriction and the ability to disable third-party integrations, alongside auto-deletion on Enterprise — controls that let IT lock the tool down rather than trust individual settings. Fellow offers advanced recording permissions on Enterprise, and Granola supports workspace-wide sharing controls on Enterprise. In hands-on checks, Fathom exposed separate toggles for external, internal and ad hoc meetings, and tl;dv supported rule-based filtering, while Fellow combined bot and bot-free modes with custom rules and admin policies.

Whatever the vendor, ask for a test tenant and verify that an admin — not a user — can stop the bot joining external meetings. That single control prevents the most common incident: an AI note taker recording a client call it was never invited to.

Where does the audio go, and for how long?

Ask four data-handling questions: does the vendor store the audio or only the transcript, what retention controls exist, can you opt out of AI training on your meetings, and where the data is processed and stored. Retention controls in particular are frequently Enterprise-gated — Notta offers auto-deletion on Enterprise, Otter and Fireflies offer custom retention on Enterprise, and Read AI only on Enterprise+.

Residency also varies: Otter stores on AWS US West, Granola on AWS US, Fireflies offers Private Storage on Enterprise, and Notta uses AWS but does not publish its regions — worth confirming with the vendor if residency is a hard requirement. Granola's desktop-only capture keeps audio local by design, which reduces the upload surface but shifts responsibility to the endpoint. For each shortlisted tool, read the trust centre, note which subprocessors are listed, and check whether the SOC 2 report requires an NDA.

Data handling by tool

Which certifications and contracts are available?

Most established vendors in this category hold SOC 2 Type II, but the contracts differ more than the badges. ISO 27001 is held by Notta and Plaud among the tools reviewed. If you operate in a regulated industry, check whether a HIPAA BAA is available and on which plan: Otter offers it as an Enterprise add-on, Fireflies includes it with Private Storage on Enterprise, Fellow lists it on Business, and Read AI supports HIPAA on Enterprise+. A standard data processing agreement (DPA) should be available from every vendor you shortlist.

Certifications are table stakes; the differentiator is whether the plan you are buying includes them. A SOC 2 badge on the trust centre tells you nothing if the retention and admin controls that matter to your auditors sit two tiers up. For the deeper checks regulated teams should run — BAAs, record-keeping obligations and vendor risk — see transcription software for regulated teams.

How does it handle consent and notices?

Consent works differently depending on the capture method. Bots are visible participants, which functions as a form of notice, but the notice text is usually the vendor's, not yours. Desktop modes shift responsibility to the user: Read AI's local capture and Microsoft's Capture app both rely on the person recording to obtain consent, which is fine for a disciplined sales team and risky for a large organisation.

Ask two questions: can admins force a notice or a recording announcement, and does the tool distinguish internal from external meetings in its consent behaviour? A tool that lets IT set the policy — rather than leaving it to each user — will survive a legal review far better. Pair this with your jurisdiction's recording-consent rules before rollout, and check the questions raised by the first major AI meeting assistant lawsuit around consent, training and retention.

What will it cost per seat once security features are included?

The advertised price is rarely the price you will pay, because the security features above often sit on Enterprise tiers. Fireflies jumps from $19 to $39 per seat for Enterprise, Granola from $14 to $35, and Read AI's security-inclusive plan is Enterprise+ at $29.75. Notta's Business plan is $16.67 per seat with Enterprise pricing custom and available from 51 seats — a lower entry point than most, though the SSO and admin controls land on Enterprise like its competitors. Otter's Business plan is $19.99 annual with Enterprise custom, and remember the 100-user minimum for SSO.

Compare the platforms' own AI too: Microsoft Copilot Business is $18 per user (promo pricing) and Google Workspace Business Standard is $14, but each covers only its home platform. For a full framework on comparing real team costs rather than list prices, see the real team cost of AI meeting tools.

Pricing by tool (USD, checked 29 October 2026)