Verified against Gemini · 2026-07-24
Turn an uploaded meeting or presentation recording into timestamped notes and action items
A native multimodal prompt for Gemini's video understanding that returns timestamped highlights, decisions, and action items — including what was shown on screen, not just said out loud — instead of a flat transcript of everything that happened.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Watch the video I've uploaded above. This is a 40-minute recording of our Q3 planning meeting, with a shared-screen slide deck for parts of it, roughly 40 minutes. Don't just transcribe it — extract the structure, and use both what's said and what's visually shown, since a screen share, a slide, or a whiteboard visible on camera can carry information the audio track alone never states out loud. OUTPUT 1. A timestamped list of the 10 most important moments (format: [mm:ss] — what happens/what's said), covering: budget decisions, headcount discussion, and anything about the Q4 launch date 2. Every decision made or commitment stated on camera, with the timestamp and who said it if identifiable — if speakers aren't visually distinguishable and names aren't stated, label them Speaker 1, Speaker 2, etc. consistently rather than guessing at identities. 3. Every action item mentioned, with an owner if named and any deadline mentioned. If an item is discussed but no owner is ever assigned on camera, say "no owner assigned" rather than inferring one from context. 4. Anything shown on screen — a slide, a number on a dashboard, a document — that adds information not said out loud, with its timestamp. 5. Anything said that contradicts something said earlier in the same video, with both timestamps. 6. If the audio is unclear or a section is inaudible, mark it as [unclear ~mm:ss] rather than guessing at words that sound plausible. Skip filler, small talk, and restated points — only budget decisions, headcount discussion, and anything about the Q4 launch date matters for this pass. Priya is the one sharing her screen for most of the meeting; Marcus joins by voice only, no camera. TONE AND ENERGY, WHEN IT'S SIGNALING SOMETHING If someone's tone shifts noticeably — a hesitant answer to a direct question, a laugh that reads as deflection rather than humor, a long pause before responding — note that as context next to the relevant point rather than only transcribing the words, but keep this to moments that genuinely change how a decision or commitment should be read, not a running commentary on everyone's mood throughout. IF THE MEETING GOES OFF-TOPIC AND BACK Note where the conversation genuinely leaves budget decisions, headcount discussion, and anything about the Q4 launch date and returns, rather than silently stitching the before-and-after together as if the detour never happened — a decision made just before a long tangent can get logically separated from the discussion that led to it if the notes paper over the gap. REVISIT NOTE If a point made earlier in the video gets revisited and changed later without either side explicitly acknowledging the change, flag that too, separately from a stated contradiction — a quiet drift in position across a long meeting is a different and subtler thing than someone directly contradicting an earlier statement.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Gemini processes uploaded video as sampled frames plus audio together, not audio routed through a separate speech-to-text pass with the picture discarded, so it can anchor output to visual timestamps and pull information straight off a shared screen or slide that the audio track never states in words — a plain transcript of the audio alone would simply miss a number visible only on a dashboard someone screen-shared without narrating it. Asking for structure instead of a transcript matters because a flat transcript of a 40-minute meeting is barely more useful than the video itself; naming focus_areas up front stops the model from treating every restated point as equally important and padding the output with small talk. The consistent Speaker N labeling instruction targets a real risk in multi-person recordings where names are mentioned inconsistently or not at all — guessing an identity from voice alone and stating it as fact can misattribute a commitment to the wrong person, which is a worse error than an honest, consistently-labeled unknown. The [unclear ~mm:ss] instruction directly targets a specific failure mode: video with noisy audio, cross-talk, or overlapping speakers gets transcribed with confident-sounding words that were never actually said, and marking those spans instead of guessing keeps the notes trustworthy enough to act on rather than requiring a full re-watch to verify. Asking for tone and energy shifts to be noted alongside the relevant point, rather than transcribed as neutral words, matters because the same sentence — 'sure, that works' — reads completely differently said flatly versus said after a long pause and a half-laugh, and a flat transcript necessarily discards that signal even though a human in the room would have registered it instantly and factored it into how seriously to take the agreement. The revisit-tracking instruction targets a specific and easy-to-miss pattern in long meetings: a position stated at minute 8 quietly shifts by minute 34 without anyone on the call explicitly flagging the change, and unless notes are built to actively watch for that drift, the summary ends up reporting only the earlier or only the later position as if it were the meeting's single stance throughout, silently erasing the fact that anything moved at all.
What you get back
[04:12] — Priya shares her screen showing a revised Q4 launch timeline slide moving the date from Nov 3 to Nov 17; no verbal objection recorded. [11:40] — Decision: engineering headcount request tabled until next planning cycle (stated by Marcus, voice only). [unclear ~22:15] — cross-talk, cannot confirm who raised the budget concern. Visual-only info: the slide at [04:12] also shows a budget line item ($42K) never mentioned aloud in the recording. Action items: Priya to send revised launch timeline by Friday — no owner assigned for the follow-up budget review mentioned at [26:03].
Verified against
Gemini Gemini 3 Pro · 2026-07-24
Changelog
- 2026-07-24 — Initial publish, verified against Gemini 3 Pro on a 40-minute screen-shared planning meeting recording.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
