The meeting happened.Now extract what it decided.
Most meeting notes end as a wall of text nobody reads. The decisions blur into the discussion. The action items lose their owners or due dates. The follow-up emails never go out. By the next meeting, half the team can’t remember what was actually agreed.
The Meeting Intelligence System is one master prompt + five focused prompts that turn any transcript into five structured artifacts: summary, decisions, action items, follow-up emails, and a standup-ready status roll-up. Honest by design — every missing owner or date gets flagged as UNCONFIRMED instead of silently invented. Paste, run, ship. One-time $49.
Recordings without rigor are just a longer thing to ignore.
Otter, Fathom, Fireflies, Granola, and the AI-summary feature built into every videoconference tool will faithfully record your meeting and produce a summary. The summaries are usually fine. The problem is what they leave you with: a wall of text that conflates discussion with decision, action items without owners, and a “to-do” list that’s really a wishlist with no due dates.
Forward that to a manager who wasn’t in the room and they’ll bounce it back asking the same three questions every time: What did we actually decide? Who’s doing what by when? What’s blocked? The system makes you answer all three, in a fixed format, with explicit flags when the meeting didn’t answer the question itself.
Generic AI summaries blur the two. The hour you spent debating Option A vs Option B reads as if you decided one of them. The system separates Decisions (with rationale + what each one closes off) from Open / undecided (debated but not resolved).
Generic summaries silently fill in the most likely owner. That's the fastest way to get a calendar full of nobody-actually-committed work. The system marks items without a clear owner or due date as UNCONFIRMED ⚠ and surfaces them in an Owner watch line.
One meeting summary is a note. Twenty meeting summaries with consistent structure is intelligence. The “What Changed” diff prompt compares last meeting's action items vs this meeting's — Advanced / Slipped / Now overdue / New / Dropped.
Clear about the lane. No inflated promises.
- A paste-and-go prompt system that runs on top of any transcript — Otter, Fathom, Fireflies, Granola, hand-typed notes, even YouTube interview transcripts.
- One master prompt that produces all five artifacts in one pass, five focused prompts for single-artifact runs, plus the “What Changed” diff for recurring meetings.
- Model-agnostic — works in Claude, ChatGPT, Gemini, or whatever model your team already uses.
- Designed to be forward-able: an UNCONFIRMED flag on every missing owner / date, “Open / undecided” list separated from Decisions, and a final “Needs clarification” section for ambiguity.
- Customizable — the prompts ship as editable templates with safe-to-edit and load-bearing sections clearly marked.
- Zero install, zero API keys, zero subscription.
- A recording / transcription platform. Otter, Fathom, Fireflies, Granola, Zoom AI Companion, Google Meet Gemini all do that work.
- A meeting scheduler. Calendly, SavvyCal, Clockwise do that work.
- A project management tool. Asana, Linear, ClickUp, Notion store the action items long-term.
- A subscription. One-time $49 with 12 months of template updates.
- A guarantee that the meeting was productive. The system structures the output honestly; if the meeting didn't decide anything, the Decisions list will be short.
- A replacement for a good note-taker. It removes the structuring work, not the listening work.
One paste. Five structured outputs.
Every transcript produces the same five artifacts in the same fixed format. Same format every time is what makes the output searchable, diff-able, and forward-able later.
What the meeting was about, the headline outcome, and the single most important next step. Three lines that a manager who wasn't in the room can read in fifteen seconds and know whether to drill in.
Every decision actually made — not every topic discussed. Each one gets the decision itself, the stated rationale, and what it closes off. Debated-but-not-resolved items go in a separate “Open / undecided” list so the record doesn't quietly turn discussion into decision.
Every commitment as a checkbox, starting with a verb, with the owner and due date as stated in the meeting. Items without a clear owner or due date are marked UNCONFIRMED with a ⚠ flag and surfaced in an “Owner watch” line — never silently filled in.
A recap email to all attendees (subject + body under 150 words, plain and human, decisions and action items embedded, with the “reply if I got anything wrong” line). Plus one short nudge email per action-item owner stating their item and due date and asking for a quick confirm.
A tight standup-ready summary for a manager who wasn't in the room. Moved forward, blocked / at risk (with the blocker named and who owns clearing it), needs a decision, and an owner watch noting anyone overloaded or any UNCONFIRMED items.
What keeps the output honest enough to forward.
The five artifacts are the surface; the three rules are the IP. Every prompt enforces them, and the tuning guide is explicit that they’re the load-bearing sections — edit the format blocks all you want, but leave these alone or the output stops being trustworthy.
Never invent owners, dates, decisions, or numbers. If the transcript is ambiguous on something important, list it under “Needs clarification” at the very end — don't smooth it over. The whole product breaks if this rule slips; everything downstream depends on the output being a faithful record, not a polished narrative.
Any action item without an owner or a due date in the actual meeting gets UNCONFIRMED with a ⚠ flag and surfaces in the “Owner watch” line. Decisions that were debated but not resolved go in a separate “Open / undecided” list, not the Decisions list. The output errors honestly on the side of flagging.
Quote sparingly; summarize in plain language. The output is meant to be read in seconds by someone who wasn't there, not parsed in minutes. If a quote is load-bearing for the decision (a specific commitment, a number called out), keep it short and quoted; otherwise summarize.
What the action-items artifact actually looks like.
Below is an abridged sample of the Action Items artifact (one of the five), generated by the focused Action Items prompt. Note the explicit UNCONFIRMED flags and the Owner Watch line — these are what survive being forwarded to a manager who wasn’t in the room.
[ ] Send the refreshed pricing slide to Q3 board pack
Owner: Maria · Due: Friday May 30
[ ] Confirm with legal whether the new EU clause needs re-papering
Owner: James · Due: Wed May 28
[ ] Migrate the staging vector DB to the new region
Owner: UNCONFIRMED · Due: UNCONFIRMED · ⚠ raised as a need; not assigned in meeting
[ ] Draft the Q3 OKRs reflecting the new growth target
Owner: Priya · Due: UNCONFIRMED · ⚠ owner committed; date not stated
─────────────────────────────────
Owner watch: Maria has 3 items this week; vector-DB migration has no owner — needs assignment before next standup.The same artifact would be invisible in a generic AI summary — most tools would silently assign the vector-DB migration to whoever spoke about it last, give Priya’s OKRs a confident-but-invented due date, and bury the Owner Watch line. That’s the gap the system closes.
Paste, focus, or diff.
One system, three paths depending on what you need from a specific meeting. The recurring + diff path is where the compounding value shows up.
Open Claude or ChatGPT, paste the master prompt, then paste your transcript directly under it. First five artifacts in under a minute. Zero setup, any model that handles a meeting-sized transcript.
Need only the Action Items? The Decisions? The follow-up emails? Run the matching focused prompt instead of the full master. Useful when you're triaging a long meeting and only need one slice right now.
For weekly / biweekly meetings, run the master prompt every time AND save each meeting's Action Items output. The “What Changed” diff prompt compares last time vs this time — Advanced / Slipped / Now overdue / New this time / Dropped. The compounding pattern that turns a meeting series into actual intelligence.
The integrity moat.
Exactly what you get for $49, and what you don’t.
- One master prompt (all five artifacts in one pass) + five focused prompts (one per artifact).
- "What Changed" diff prompt for recurring meetings.
- Fixed output format per artifact, with safe-to-edit and load-bearing sections clearly marked.
- Tuning guide for long transcripts, recurring meetings, and custom format extensions.
- Model-agnostic — works in Claude, ChatGPT, Gemini, or whatever your team uses.
- 12 months of template updates as the prompts evolve.
- A transcription platform. Use Otter, Fathom, Fireflies, Granola, or your videoconference's built-in AI.
- A meeting scheduler. Use Calendly, SavvyCal, Clockwise.
- A project management tool. The action items need a home in Asana / Linear / ClickUp / Notion / your CRM.
- An API integration. The system is paste-and-go; if you want auto-pull from your transcription tool, wire it yourself.
- A subscription. One-time $49 with 12 months of updates.
- A guarantee the meeting was productive. The system structures the output honestly; the meeting still has to make decisions.
Pairs naturally with the Claude Project Blueprint Library ($69) — drop the master prompt into a Claude Project’s instructions, load past meeting outputs into the knowledge base, and the “What Changed” diff runs across the whole meeting series with zero re-pasting.
For agency teams running client status meetings, pair with the Agency Operators Skills Pack ($89) — the weekly-status-report-generator skill consumes the meeting-intelligence Status Roll-up directly and turns it into the client-facing weekly update.
For founder operators running board / leadership meetings, pair with the Solo Founder Skills Pack ($79) — the board-update and weekly-cadence skills take the meeting-intelligence outputs as input and turn them into the investor-facing artifacts.
The questions operators actually ask before deploying.
One master prompt that runs all five artifacts in a single pass, five focused prompts that run any single artifact alone (Action Items, Decision Log, Follow-up Emails, Status Roll-up, “What Changed” diff), a fixed output format for each artifact so results look the same every time, and a tuning guide for long transcripts, recurring meetings, and custom format extensions. The whole system is paste-and-go — no install, no API keys, no setup.
Any model that handles a meeting-sized transcript in one pass — Claude (Sonnet 4.6 or Opus 4.7 recommended for long meetings), GPT-5.x, Gemini, or whatever your team already uses. The prompts are model-agnostic. For long transcripts (60+ minutes), use a model with a 200K+ context window so the whole thing fits in one pass; the tuning guide covers the chunking pattern if you have to split.
Those are recording / transcription platforms that auto-attach to your meetings, capture audio, and produce a transcript plus a summary. This is the analyst layer that runs on top of any transcript regardless of where it came from — including hand-typed notes, a YouTube interview transcript, a Slack huddle export, or your existing Otter / Fathom / Fireflies / Granola output. The hard rules (UNCONFIRMED flags, evidence-only extraction, structured artifacts) are designed to produce notes you can forward to a manager without re-reading the transcript first. Many users run both: the recording tool captures, this system structures.
No — that's the load-bearing rule. The master prompt explicitly forbids inventing owners, dates, decisions, or numbers; any action item without a clear owner or due date gets marked UNCONFIRMED with a ⚠ flag instead of being silently filled in. Same for decisions that were debated but not resolved (they go in a separate “Open / undecided” list). The whole point is that the output has to survive being forwarded to people who weren't in the room — and that requires honesty about what was actually agreed vs what was discussed.
Yes — and it's where most of the compounding value shows up. Paste last meeting's Action Items output and this meeting's transcript, and the diff prompt produces a clean delta: Advanced (items that progressed or completed), Slipped / still open (items that didn't move), Now overdue (action items past their stated due date), New this time, and Dropped. Recurring meetings where you run the diff every week start to surface chronic-blocker patterns and people-overload patterns that don't show up in a single meeting in isolation.
Still works. The master prompt is written for both clean Otter / Fireflies-style transcripts AND rough hand-typed notes — anything that contains the substance of what was discussed. Rough notes will produce more UNCONFIRMED flags (because the source has fewer explicit owner / date assignments) but the system errors honestly on the side of flagging rather than guessing. The output is still structured, still searchable, still forward-able.
Yes — the prompts ship as templates, not locked logic. Edit any of the format blocks to add a “risk” tag, a Jira-ready format, a custom field for your CRM, or whatever your team needs. The only thing to leave intact is the “never guess / flag UNCONFIRMED” rules — that discipline is what keeps the output honest. The tuning notes walk through the safe-to-edit vs load-bearing sections.
30-day no-questions refund. Run the master prompt on your next three meetings. If the output isn't clean enough to forward to someone who wasn't in the room — without you re-reading the transcript first to spot-check — email and we refund. Refunds across the catalog are countable on one hand; for meeting tooling specifically, the value is usually obvious on the first meeting.
Stop forwarding rough notes.
Start forwarding decisions.
One master prompt + five focused prompts + a diff for recurring meetings. Five structured artifacts per transcript. No install, no API keys, no subscription. Works in Claude, ChatGPT, or Gemini.
Sold by RedHub AI LLC · Secured by Stripe · redhub.ai