Meeting transcription in 2027: eleven feature trends, and what they mean for your data

What meeting note-takers actually shipped between May and October 2026 — bot-free capture, on-device speech models, MCP for agents, chat across meetings, wearables, voiceprints, live translation — with a source for every example, and what each trend changes about where your meetings end up.

This is not a forecast. It is a list of what meeting transcription tools shipped in the last six months, grouped into the directions they share, with a primary source for each example. Where a trend changes where your audio or transcripts go, we say so — that is the question we care about most. Daisy is one of the products in this market, so where we describe it, we say what it does today and what it does not.

Period: May–October 2026, with a few late-2025 roots
Sources: Vendor changelogs and announcements, court filings coverage, regulator guidance
Not covered: Accuracy rankings — we only publish those with raw output on the same audio

1. The meeting bot is going away

Recording from your own computer — system audio plus microphone, no participant joining the call — has become the default rather than the niche. Fathom added a bot-free mode on Mac in April; Wispr launched a bot-free notetaker on Mac in August and on Windows in September; Superwhisper launched Meetings in October and called bots “invasive”; Notion records from its desktop app with /meet. What it changes: no vendor in the participant list, but bot-free does not mean local — many of these still upload the audio. Fathom — what's new. Wispr Notetaker. BetaKit — Superwhisper and Cohere.

2. Speech recognition moves onto the device

Small open speech models now run fast enough on a laptop for real meetings. NVIDIA's Parakeet v3 covers 25 European languages and has a Core ML port; Alibaba released Qwen3-ASR with open weights in January; Cohere released Transcribe under Apache 2.0 in March, and Superwhisper runs it locally. Apple's own SpeechAnalyzer on macOS 26 and iOS 26 is a system-level option. What it changes: “where does the audio become text?” now has a realistic answer of “on your computer” without giving up quality. Parakeet v3 Core ML. Qwen3-ASR. Cohere Transcribe. MacStories — SpeechAnalyzer.

3. Your meetings become a data source for AI agents

The Model Context Protocol (MCP) lets Claude, ChatGPT, Cursor and other agents read meeting history directly. Otter expanded its MCP server in April, Fathom added MCP for ChatGPT and Claude, Wispr and Grain shipped MCP in August, and Notion 3.8 added MCP for outside agents with admin controls in October. What it changes: whatever an agent reads from your meetings is sent to that agent's provider. A local MCP server keeps the archive on your computer, but the excerpts the agent reads still leave it. Otter, April 28. Grain release notes. Notion 3.8.

4. From notes to “company memory”

Vendors now pitch meetings as a knowledge layer for agents, and the money followed. Superhuman (formerly Grammarly) acquired Fathom in September to give its agents meeting context; Otter raised $320M in October around a “conversational knowledge engine”; Granola raised $125M in March around shared spaces and APIs. What it changes: more of your organisation's conversations end up in one searchable cloud index — useful, and a bigger target. Superhuman acquires Fathom. Otter, October 20. Tech.eu — Granola Series C.

5. Ask across every meeting, with saved prompts

Chat over the whole archive is now expected, and saved prompts are becoming the template layer: Granola's Recipes, Fathom's shareable account-wide prompts, Otter's rebuilt AI Chat, Wispr's questions across your meeting history. What it changes: a question across all meetings sends excerpts from many meetings at once, so it matters which model answers and whether you can see what was sent. Granola — Recipes. Fathom — what's new.

6. Before, during and after the call

Tools now cover the whole meeting: Otter's Work Mode writes pre-meeting briefs and drafts CRM updates and follow-ups for review, and its Live Assist shows private prompts during the call; Wispr catches you up on what you missed; Fathom shows live summaries. What it changes: drafts that go into a CRM or an email are only as good as the transcript — and they take meeting content into more systems. Otter, October 27. Wispr Notetaker.

7. Dictation and meeting notes merge

Dictation apps added meetings — Wispr in August, Superwhisper in October — and meeting apps added dictation, such as Granola's voice typing beta. Voice commands that rewrite, translate or answer inside any text field are standard: Wispr calls it Command Mode. What it changes: one app hears both your meetings and everything you dictate, so the privacy model has to cover both. Wispr — Command Mode. Granola — voice typing.

8. In-person capture moves to the watch and the phone

Dedicated pendants are losing ground: Meta bought Limitless in December 2025 and pendant sales stopped. Granola launched an Apple Watch app in July — “the best wearable is one you already have” — while Plaud keeps a niche with its $179 NotePin S. What it changes: conversations away from the desk get recorded far more often, including people who never joined a call. Granola for Apple Watch. BetaNews — Meta acquires Limitless. TechCrunch — Plaud NotePin S.

9. Consent becomes a legal question, not a courtesy

Most claims in the Otter privacy litigation survived dismissal in August, including wiretapping claims under California law. A lawsuit filed against Granola in July targets bot-free capture itself, alleging audio was streamed to a third-party transcription vendor during calls. The EU AI Act's transparency obligations under Article 50 apply from August 2, 2026. What it changes: “nobody saw a bot” is not consent. Expect tools to add ways to tell participants they are being recorded — and expect where the audio is processed to matter in court. Baird Holm — Otter litigation. PPC Land — Granola lawsuit. Goodwin — EU AI Act Article 50.

10. Knowing who spoke, by voice, across meetings

Grain's Voice Signatures, shipped in August, recognise speakers by voice across meetings, including in-person ones. What it changes: a voiceprint is biometric data in several jurisdictions. Whether it is stored on your device or on a vendor's server is the whole question. Grain — Voice Signatures.

11. Live translation, free tiers and Windows

Google Meet expanded real-time speech translation to 63 languages in October; note-takers themselves have barely started on it. Pricing is moving towards generous free tiers and paying with a subscription you already have — Notion 3.8 lets a ChatGPT plan pay for Notion AI. And Mac-first tools are reaching Windows: Wispr in September, Otter for Desktop in April, the open-source Meetily with a Windows installer. Google Workspace — Meet translation. Notion 3.8. Meetily releases.

How to choose a meeting tool for 2027

Where does the audio become text — on your computer or on a server? Bot-free does not answer this.
When an agent or a chat reads your meetings, can you see which excerpts are sent, and to whom?
Where are speaker voiceprints kept, if the tool has them?
Can you pick the model for summaries and questions, including a local one?
Is the archive in files you own, or in a workspace you rent?
How does the tool help you tell participants they are recorded?

Where Daisy stands

Daisy records without a bot, transcribes on your Mac, lets you choose the summarizer — including a local model or Apple Intelligence — writes meetings as files into a folder you choose, does dictation and voice notes in the same app, and exposes a local MCP server bound to 127.0.0.1. It is open source under Apache 2.0, and the verify-local page shows how to test all of this with the network off. It has no team workspace, no live translation and no live assist during calls. An iPhone and Apple Watch app is coming, and a Windows app is in development.