MultimodalImage, audio & video
Google launches Keep Live, a Gemini voice feature that turns spoken brain dumps into structured notes
Google has begun rolling out Keep Live, which uses Gemini audio models to turn loosely spoken thoughts into organized lists and notes in Google Keep, including updates to existing lists, alongside companion voice features in Gmail and Docs.
QuietHow it’s measured
YouTube video · 130 views · 3 likes
Reach blends public engagement (views, likes, comments, shares, stars) from the platforms where the story or its original source lives: YouTube, Hacker News, GitHub, Substack and similar. A video and the article written from its transcript share one score, taken from the video. Platforms with no data are left out rather than counted as zero, and the age of a story never lowers its score.
AT Pulse research deskPublished · 4 min read · AI-assisted analysisKeep Live takes unstructured speech and produces a structured note or checklist, with an inline preview after a few seconds. It can add new items to a list that already exists instead of creating a duplicate, and reports say it also accepts voice edits such as checking items off. [3] [6] [7]
Keep Live is the quick-capture member of a three-part voice launch. Gmail Live answers spoken questions about the inbox. Docs Live helps draft documents and, with permission, can draw on Gmail, Drive, Chat and the web. Google says only that the features use Gemini audio models, and the exact model version is unconfirmed. [3] [5] [8] [9]
Access is narrow, and launch reports disagree on its scope. The feature is English-only. Google lists the AI Plus, Pro and Ultra plans, but TechCrunch names only Pro and Ultra. Outlets also disagree on whether iOS is included at launch, and Workspace business accounts have no release date. [3] [5] [6] [7]
Google's description is the only account of how well Keep Live organizes speech; no independent accuracy tests exist.
Read the full assessment
Implication: voice-to-structured-edit is spreading across productivity apps, but teams should pilot with human review before relying on it.
Status: research done; the dossier follows.
Executive brief
The most consequential gap is that Google hasn't published any evidence of how well Keep Live works. Google announced it on 3 September 2026. Keep Live listens to a spoken "brain dump" and turns it into structured lists and notes in Google Keep (Google). It is the note-taking member of a trio with Gmail Live and Docs Live. Reports at launch disagreed on which plans and devices get it. The linked post and video are commentary with paid-community promotion. They repeat launch coverage and add usage tips, not test results.
What changed and event timeline
Google previews the feature at I/O
Google showed voice "Live" features for Gmail, Docs and Keep at its developer conference before releasing them ().
Gmail Live, Docs Live and Keep Live launch
Google says they began rolling out that week, are built on Gemini audio models and are "coming soon" to Workspace business customers (;).
3–5 Sep 2026: Reports split on who gets access
9to5Google updated its story to add the AI Plus plan. TechCrunch and Chrome Unboxed list only Pro and Ultra. Several outlets say Keep Live is Android-only at first (;).
A commentary post and video promote "brain dump" use
Both restate launch coverage, warn against treating Keep Live like dictation, and promote a paid community (;).
Capabilities and access
- What it does: You speak loosely and Keep returns a structured note or checklist. An inline preview appears after a few seconds (9to5Google).
- Works with existing notes: It can add an item to a list you already have instead of creating a duplicate (Chrome Unboxed).
- Model: Google says only "Gemini audio models." GCN names "Gemini 3.5 Live." No exact version is confirmed.
Read the full section
- What it does: You speak loosely and Keep returns a structured note or checklist. An inline preview appears after a few seconds (9to5Google).
- Works with existing notes: It can add an item to a list you already have instead of creating a duplicate (Chrome Unboxed). 9to5Google says you can also edit by voice, for example checking items off.
- Model: Google says only "Gemini audio models." GCN names "Gemini 3.5 Live." No exact version is confirmed.
- Access: English only. Workspace business accounts have no date (TechCrunch).
Technical analysis for researchers and developers
- What is documented: Speech goes in, the model infers what you mean, and it writes or changes structured content in the app.
- What is not documented:
- Whether audio is processed on the device or in the cloud
Read the full section
- What is documented: Speech goes in, the model infers what you mean, and it writes or changes structured content in the app. That includes formatting items as a checklist and merging them into a list that already exists (Chrome Unboxed).
- What is not documented:
- Whether audio is processed on the device or in the cloud
- How long audio is kept
- Latency targets
- Any API for developers
- Any model card or evaluation for this feature
- Implication: Nobody can reproduce or benchmark it. The only observable behaviour is the preview-then-confirm step, which suggests users are expected to review results.
- Developer takeaway: Builders should treat "merge into existing note" as an untested retrieval-and-edit step that may hit the wrong list.
Claims and evidence
- Turns rambling speech into structured notes: Google's own description (Google).
- Plan tier: Google's post lists Plus, Pro and Ultra. TechCrunch says Pro and Ultra only. 9to5Google's update sided with Google.
- Platform: TechCrunch says iOS and Android. 9to5Google, Chrome Unboxed and GCN say Android-only.
Read the full section
- Turns rambling speech into structured notes: Google's own description (Google). No independent test of accuracy or speed was found, and the video offers none either (video).
- Plan tier: Google's post lists Plus, Pro and Ultra. TechCrunch says Pro and Ultra only. 9to5Google's update sided with Google.
- Platform: TechCrunch says iOS and Android. 9to5Google, Chrome Unboxed and GCN say Android-only.
- "Saves time": This is the video's own claim, with no measurement behind it (video).
Context and prior work
- Keep has offered voice-to-text for years. The video argues that a raw transcript leaves messy thoughts messy, and that Keep Live's real change is organising them (video).
- The Live trio splits voice by app: Gmail Live answers questions about your inbox, and Docs Live helps draft documents, pulling from Gmail, Drive, Chat and the web if you allow it (alternativeto; video).
Read the full section
- Keep has offered voice-to-text for years. The video argues that a raw transcript leaves messy thoughts messy, and that Keep Live's real change is organising them (video).
- The Live trio splits voice by app: Gmail Live answers questions about your inbox, and Docs Live helps draft documents, pulling from Gmail, Drive, Chat and the web if you allow it (alternativeto; video).
Limitations, safety and contested findings
- Limited access: Keep Live is gated by plan, English only, and possibly Android-only. The video itself mentions these limits (video).
- Privacy: The only privacy statement found covers Docs Live, which pulls from other apps only with your permission.
- Reliability: The video advises reading every output before relying on it.
Read the full section
- Limited access: Keep Live is gated by plan, English only, and possibly Android-only. The video itself mentions these limits (video).
- Privacy: The only privacy statement found covers Docs Live, which pulls from other apps only with your permission. Nothing comparable was found for Keep Live audio (video; BetaNews).
- Reliability: The video advises reading every output before relying on it. Its talk of a closing "window" for early adopters is a sales pitch, not evidence (video).
- Conflicting reports: The plan and platform conflicts above are unresolved across the reviewed sources.
Business and practitioner implications
- Who can use it now: Only consumer AI subscribers. Workspace teams have no date, so don't plan workflows around it yet (Google).
- Possible use: Capturing follow-ups right after a meeting, as long as someone reviews the result.
- Small pilot: Run short single-topic brain dumps and log which items were missed or added to the wrong list (video).
Read the full section
- Who can use it now: Only consumer AI subscribers. Workspace teams have no date, so don't plan workflows around it yet (Google).
- Possible use: Capturing follow-ups right after a meeting, as long as someone reviews the result.
- Small pilot: Run short single-topic brain dumps and log which items were missed or added to the wrong list (video).
- For product teams: "Voice in, structured edit out" is becoming a standard pattern in productivity apps. Competitors will need to show measurable accuracy, not just demos.
The source trail.
Sources (10)
Google Keep AI Note Taking Fixes the Part of Notes Nobody Does
Feed excerpt only. A transcript of the video linked from this page is supplied (linked video auto captions; https://www.youtube.com/watch?v=zJqg-dbkmtg); automatic text may contain errors.
reddit.comLinked video: Google Keep AI Note Taking Fixes the Part of Notes Nobody Does
linked video auto captions
www.youtube.com