The Meeting AI Analyser Blog

Local Whisper transcription, Claude AI meeting notes, and the case against cloud meeting bots.

Guide

Microsoft Teams is blocking AI notetaker bots: how to still get meeting notes in 2026

Published 11 Aug 2026 · 7 min read

If your AI notetaker has stopped showing up in Microsoft Teams meetings, it is probably not broken — it is being detected and held in the lobby. Since a change Microsoft announced in March 2026 (Message Center notice MC1251206), Teams identifies external third-party meeting bots, flags them as unverified, and requires the organiser to admit them explicitly. From August 2026, administrators can switch a tenant-wide policy to block detected bots outright. The durable fix is not a different bot: it is a tool that captures the audio on your own machine and never joins the call as a participant.

Why is my AI notetaker stuck in the Microsoft Teams lobby?

An AI notetaker gets stuck in the Microsoft Teams lobby because Teams now classifies external third-party meeting bots as a distinct participant type and screens them on join. Microsoft's stated reasoning is blunt: bots may access meetings "without the knowledge or consent of the meeting organizer or the hosting tenant." Detected bots land in a lobby group labelled as suspected threats, and there is deliberately no one-click admit for them.

Practically, that means your Otter Assistant, Fireflies "Fred" or tl;dv recorder now depends on a human noticing a prompt and approving it, every single meeting. If the organiser is not you, you have no control over that at all.

What exactly did Microsoft change in Teams in 2026?

Microsoft added a Teams meeting policy called "Manage external bots and their access to meetings", exposed in the Teams Admin Center and via the ExternalBotAccessMode attribute of Set-CsTeamsMeetingPolicy. It has three values: AllowAllBots (no detection), RequireApprovalWhenDetected (lobby approval — the current default), and BlockDetectedBots (refused at the door). The dates, per Microsoft's own communications and roadmap entry 566201:

  • 13 March 2026 — announced in Message Center notice MC1251206.
  • Mid-May 2026 — Targeted Release: bots labelled and routed to the lobby.
  • Early-to-mid June 2026 — general availability of lobby approval as the default.
  • August 2026BlockDetectedBots begins rolling out across desktop, web, Mac, iOS and Android, for worldwide and GCC tenants.

Two exclusions matter: the policy targets external third-party bots, so it does not block Microsoft 365 Copilot, and it excludes bots registered as Entra ID applications inside your own tenant. The older CAPTCHA verification policy is being retired in favour of this framework. So this is not a temporary glitch to wait out — it is the new default posture, and it only gets stricter.

Are Zoom and Google Meet blocking notetaker bots too?

Zoom and Google Meet have moved in the same direction, though less aggressively than Teams. Zoom updated its settings back in October 2024 to let organisations block third-party AI bot domains from joining meetings, and Zoom's own AI Companion requires host initiation with a visible indicator. Google Meet in 2026 flags third-party notetaker bots as a potential risk and routes higher-risk join requests into a separate screening queue, while Workspace admins can restrict OAuth-connected notetakers outright.

The direction of travel is consistent across all three platforms: a recorder that joins as a guest is now treated as an unknown third party by default. Part of the pressure is legal — the consolidated class action In re Otter.AI Privacy Litigation (case 5:25-cv-06911, Northern District of California) had a motion-to-dismiss hearing on 20 May 2026, with plaintiffs arguing that a bot's consent model fails in all-party consent states.

Does "bot-free" also mean my meeting audio stays private?

No — and this is the distinction most "best bot-free notetaker" listicles quietly skip. "Bot-free" describes only how the audio is captured: from your device instead of a guest participant. It says nothing about where the audio is transcribed. Several popular bot-free tools capture locally and then stream that audio straight to a cloud transcription provider.

Granola is the clearest example: it runs on your computer and listens to microphone and system audio, then passes that audio to cloud transcription subprocessors (AssemblyAI and Deepgram are listed) while the meeting is happening. The audio is deleted after transcription and the notes sit in a US-hosted AWS VPC — reasonable engineering, and Granola is SOC 2 Type 2 certified. But if the reason your IT team blocked bots was "meeting content must not reach a third party", a bot-free cloud tool does not actually solve their problem. It just makes the violation invisible.

If that distinction matters to you, the question to ask a vendor is not "do you send a bot?" but "does the audio leave my machine, and does the transcript leave my tenant?"

Which meeting note tools do not send a bot into the call?

Tools that do not send a bot into the call fall into three groups: local-first desktop capture, cloud-backed desktop capture, and browser extensions that read native captions. The table below separates the two things that get conflated — bot presence and audio destination. Prices are approximate, as of August 2026.

Tool Joins as a bot? Where audio is transcribed Survives BlockDetectedBots Price model
Meeting AI Analyser No On your PC (Whisper, local) Yes €29 one-time
Otter.ai Yes (Otter Assistant) Otter cloud No ~$17/user/mo
Fireflies.ai Yes ("Fred") Fireflies cloud No ~$18/user/mo
tl;dv Yes tl;dv cloud No ~$25/user/mo
Fathom Yes by default; bot-free desktop capture in beta (Mac) Fathom cloud Only in the bot-free mode Free tier / ~$20/user/mo
Granola No Cloud subprocessors (AssemblyAI, Deepgram) Yes Subscription
Tactiq No (Chrome extension) Tactiq cloud Yes ~$12/user/mo

Only the first row keeps the audio on the machine. Meeting AI Analyser captures system audio and microphone through the Windows WASAPI stack, transcribes locally with Whisper in 90+ languages, and sends only the resulting text to Claude for summaries, translation and action items — through your own Claude plan or Anthropic API key. Nothing joins the meeting, so there is no bot for Teams to detect, label or block. For the full feature-by-feature view, see the comparison on the homepage.

Will IT block a desktop capture app as well?

A desktop capture app is governed by a different control than a meeting bot, so the Teams policy does not apply to it — but that is not the same as blanket approval. ExternalBotAccessMode operates on participants joining a meeting; an application recording your own machine's audio output never joins, so there is nothing for Teams to screen. Endpoint policy, software allowlisting and your organisation's recording rules still apply, and they should.

The honest framing for a conversation with IT is this: the tool removes the third-party data flow they were worried about, and it removes the undisclosed participant. It does not remove your obligation to follow the recording policy. Being able to point at open source code on GitHub and say "here is exactly what leaves the machine — the text transcript, to an Anthropic endpoint we already have a contract with, or nothing at all if AI analysis is off" is a much shorter security review than defending a SaaS recorder.

Do I still have to tell participants that I am transcribing?

Yes. Removing the bot removes a technical disclosure, not a legal one. Consent rules follow the participants' location, not the architecture of your tool: over a dozen US states, including California, Florida and Pennsylvania, require all-party consent, and the safest practice in a mixed-location call is to apply the strictest rule. In the EU, a meeting transcript is personal data of every participant under GDPR, and from 2 August 2026 Article 50 of the AI Act adds transparency obligations on deployers of AI systems, with worker monitoring treated as high-risk under Annex III.

What local transcription genuinely changes is the compliance surface: no international transfer to assess, no additional sub-processor to document, no third-party retention period to audit. We wrote about that in more detail in why local meeting transcription matters for privacy in 2026. Say out loud that you are taking AI-assisted notes. It costs you one sentence.

What does this cost compared to a per-seat subscription?

Switching away from bot-based notetakers is usually cheaper, not more expensive. Cloud meeting assistants bill roughly $12 to $25 per user per month, indefinitely, and the AI analysis is metered on their side. Meeting AI Analyser is a one-time €29 Pro license valid on up to 3 machines with free updates for life, after a 7-day free trial. The AI side runs on the Claude subscription or Anthropic API key you already control, so there is no per-minute analysis fee — the mechanics of that are covered in how to use your Claude subscription as a live meeting assistant.

The trade-off, honestly

Local transcription is not free of downsides. Whisper needs roughly 4 GB of free RAM and a modern multi-core CPU, and a GPU helps if you want the larger models; a cloud service offloads all of that. Meeting AI Analyser is Windows 10 and 11 only, so Mac and Linux users are out. There is no shared web inbox of past meetings for your whole team to browse, no CRM auto-sync, and setup asks a little more of you than clicking "connect calendar" — you pick a mode and, for the AI features, install the Claude Code CLI or paste an API key.

What you get in exchange is a tool that cannot be blocked in the lobby, cannot be labelled a suspected threat, does not appear in anyone's participant list, does not put your meeting audio on someone else's infrastructure, and does not add a monthly line item. If the reason you are reading this is that Teams stopped admitting your notetaker, that set of properties is the actual fix.

Try Meeting AI Analyser free for 7 days →

Privacy

Why local meeting transcription matters for privacy in 2026

Published 28 Feb 2026 · Updated 17 May 2026 · 6 min read

Every modern meeting transcription tool — Otter.ai, Fireflies, Tactiq, tl;dv, Read.ai — sends your audio to its servers. That's not a side effect; it's the architecture. The transcription model runs in their cloud, the recording is stored in their account, and the AI summary is generated on their infrastructure. Convenient, yes. Private, no.

What you actually upload when you "just take notes"

When a bot like Fireflies joins a Zoom call, every word your team, your client, your CFO or your lawyer says is streamed to a third party. The transcript, the audio, and increasingly the AI-extracted "key topics" sit in a SaaS account governed by terms you almost certainly didn't read. Most providers reserve the right to use anonymised data to improve their models. In regulated industries — healthcare, legal, finance, defence — this alone is a deal-breaker.

Why "encrypted in transit" isn't enough

Vendors love to say "your data is encrypted in transit and at rest." Both statements are true. Neither answers the real question: who can read it? The vendor's employees, under their access policies, can. Their sub-processors can. A court order can compel disclosure. A breach can leak it. The only architecture that defeats all four risks at once is the one where the audio never leaves your machine.

The local-first alternative

Meeting AI Analyser runs OpenAI's Whisper model directly on your Windows PC. System audio is captured via WASAPI, transcribed in memory, and discarded. Only the resulting text is sent to Claude for analysis — and even that step can be turned off for fully offline use. There is no bot in the call, no cloud upload, no third-party recording.

GDPR, NIS2, and the practical compliance argument

Under GDPR, a meeting recording is personal data of every participant. Sending it to a US-based SaaS provider triggers a transfer assessment, a DPA, and consent management. Doing the transcription on the host's own device removes the transfer entirely — the data simply never crosses a border. For NIS2-regulated organisations the same logic applies: minimise the attack surface by keeping sensitive content on-prem.

The trade-off, honestly

Local transcription needs a few GB of RAM and a modern CPU. A GPU helps with larger Whisper models. In exchange you get: no monthly bill per user, no bot to disclose to participants, no audio sitting in someone else's S3 bucket. For most professional meetings on a modern laptop, the trade-off is no trade-off at all.

Try Meeting AI Analyser free for 7 days →

Comparison

Meeting AI Analyser vs Otter.ai, Fireflies, Tactiq and tl;dv

Published 28 Feb 2026 · Updated 17 May 2026 · 8 min read

If you're researching meeting AI tools in 2026 you'll meet the same four names everywhere: Otter.ai, Fireflies.ai, Tactiq and tl;dv. They share an architecture (cloud transcription) and a pricing model (per-user monthly subscription). Meeting AI Analyser was built on the opposite premises. Here's how they compare, feature by feature.

Otter.ai

Otter is the incumbent. Strong web UI, decent English transcription, polished mobile apps. Weaknesses: audio is uploaded and stored on Otter's servers, free tier is capped at 300 minutes/month, the "Otter Assistant" bot literally joins your meetings (participants see it), and pricing starts at $17/user/month for the Pro plan with limits that the average power user hits in a week.

Fireflies.ai

Fireflies sends a recorder bot ("Fred") into every meeting. It's the deepest integration with CRMs (HubSpot, Salesforce), which is great for sales teams. The flip side: the bot is visible to every participant, recording is cloud-hosted by default, and the Pro tier ($18/user/month) is where most of the AI value sits.

Tactiq

Tactiq is a Chrome extension that taps Google Meet's native captions. Clever — no bot — but it only works with Meet, only in Chrome, depends entirely on Google's captioning quality, and ships transcripts to Tactiq's cloud for AI features. $12/user/month.

tl;dv

Closest in spirit to Fireflies: a bot joins, records, transcribes, summarises. Strong free tier for personal use, but the AI features and integrations live behind $25/user/month, and again — the audio is theirs.

Meeting AI Analyser

One installer for Windows. No bot. Audio captured locally via WASAPI from any source (Teams, Zoom, Meet, Webex, Discord, browser tab, in-person via mic). Transcription runs locally with Whisper in 90+ languages. Analysis runs through your own Claude subscription via the Claude Code CLI — so the AI cost is whatever you already pay Anthropic, with no per-minute markup. One-time €29 for the Pro license, valid on up to 3 machines, free updates for life.

When each one wins

  • Otter if you need a polished web inbox of past meetings and don't care about privacy.
  • Fireflies if your sales team lives in HubSpot/Salesforce and tolerates the bot.
  • Tactiq if 100% of your meetings happen in Google Meet and you only want highlights.
  • tl;dv if you want a generous free tier and don't mind the bot.
  • Meeting AI Analyser if you want zero cloud audio, no bot, a flat one-time price, and Claude AI quality on the analysis side.

For the full feature table, see the comparison on the homepage.

Productivity

How to get the most out of AI-generated meeting notes

Published 28 Feb 2026 · Updated 17 May 2026 · 5 min read

A 4000-word AI transcript of your hour-long meeting is not productivity — it's homework. The real win shows up when you bend the AI from "summariser" to "operations assistant". Here's how teams using Meeting AI Analyser with Claude squeeze actual leverage out of their meetings.

1. Decide what you want before the call starts

Claude works dramatically better when you tell it the role it should play. Before the meeting starts, type one line into the chat panel: "You are my technical PM. Track decisions, owners, deadlines and any RFC numbers mentioned. Ignore small talk." Or for sales: "You're listening as if you were my sales manager. Track objections, commitments and next steps. Flag any pricing mentioned." The summary you get every 60 seconds is shaped by this single instruction.

2. Use conversation memory for recurring meetings

The new Conversation Memory feature lets you resume a previous Claude session. For weekly syncs, picking the previous week's conversation means Claude already knows the project, the people and the open threads. By week 4 the summaries are noticeably sharper than any cloud tool's "first-time" output.

3. Translate, don't just transcribe

If you join a meeting in a language you only partly speak — German with a French team, English with a Tokyo office — ask Claude in your native language. Whisper transcribes in the original, Claude translates and summarises in yours. You stay in the conversation instead of mentally translating in real time.

4. Decode jargon on demand

Sitting in a technical review and someone drops "we'll handle it via the OIDC discovery endpoint"? Ask Claude "explain that last paragraph for a non-specialist". The transcript is right there; the answer lands in two seconds. This is the single most underused feature.

5. End every meeting with three artefacts, not one

Before closing the app, ask Claude for three things in sequence: (a) a 5-line executive summary, (b) a decisions / action items / owners table, (c) a draft email to send to anyone who missed the call. That's 30 seconds of work and turns a meeting into a published outcome.

Try Meeting AI Analyser free for 7 days →

Claude

How to use your Claude subscription as a live meeting assistant

Published 9 Aug 2026 · 6 min read

If you already pay for Claude Pro, Team or Enterprise, you are sitting on an underused asset: a top-tier model you could be running live during every meeting, for no extra per-minute cost. Most meeting tools ignore this and bill you a second subscription to use their AI on their cloud. Here is how to turn the Claude plan you already have (or a pay-as-you-go Anthropic API key) into a real-time meeting co-pilot on Windows.

Why route meetings through your own Claude, not another SaaS

Cloud meeting tools charge $12 to $25 per user per month, and the AI summary is generated on their servers from audio you uploaded. Running the analysis through your own Claude account flips both problems: the transcript is processed under your Anthropic account (not a third party's), and the cost is whatever your subscription already covers, with no per-minute markup. You are not renting an AI you already pay for.

Two ways to connect: subscription or API key

There are two supported modes, and you pick whichever fits:

  • Claude subscription via the Claude Code CLI: install the CLI, sign in with your Pro/Team/Enterprise plan, and the app uses it for analysis. Cost is covered by your subscription, no per-minute fees.
  • Your own Anthropic API key: paste a key in the app settings, pay-as-you-go, no CLI or subscription required. Ideal if you would rather pay per token than hold a subscription.

Transcription runs locally with Whisper either way, so it works even before you connect any AI at all.

The setup, end to end

Download Meeting AI Analyser, install it, and choose your mode in settings. Launch it, then open Teams, Zoom, Meet, Webex or any app playing audio. The app captures system audio via WASAPI, transcribes locally with Whisper, and streams only the text transcript to Claude for a structured summary every 60 seconds. No bot joins the call, and the audio never leaves your PC.

What you get during the call

Live transcription in 90+ languages, on-demand translation into your language, jargon explanations when someone drops an acronym, and rolling summaries with decisions, action items and owners. Because it is your own Claude, you can steer it with a one-line instruction ("track objections and pricing" for sales, "capture RFC numbers and decisions" for engineering).

Is it worth it if you only have the API key?

Yes. A one-hour meeting transcript is a few thousand tokens of text (not audio), so a typical meeting costs a few cents of API usage. Compared to a $17 to $25 monthly SaaS subscription, pay-as-you-go on your own key is dramatically cheaper for anyone who does not meet all day.

Try Meeting AI Analyser free for 7 days →

Ready to retire your meeting bot?

7-day free trial. €29 one-time. Local audio, Claude AI brain.

Get Pro - €29 one-time Start Free 7-Day Trial