Back to Blog

Why Meeting Action Items Never Get Done

Emily Watson9 min

There's a specific kind of silence that happens about four days after a meeting. Someone asks "hey, did you send that over?" and the person on the other end has no memory of agreeing to send anything. They didn't forget on purpose. The commitment just never survived the trip from the call to their actual to-do list.

This happens constantly, and it's not really a writing problem. Search "how to write action items" and you'll get a dozen guides telling you to use an owner, a verb, and a deadline. That advice is correct and also mostly irrelevant, because the failure point isn't the sentence structure. It's what happens to that sentence after the meeting ends.

The action items are fine. The routing is the problem.

Picture the last meeting recap you actually read closely. It probably lived in one of three places: a shared doc, an email, or a Slack message that got buried within an hour. In all three cases, the action item existed — clearly written, correctly assigned — and still didn't get done, because nothing about its location made it anyone's job to look at it again.

Compare that to a task that shows up as a card in someone's project board, or a message that pings them directly with their name attached. Same content, wildly different completion rate. The five-part formula (owner, verb, deliverable, deadline, context) makes an action item legible. It does nothing to make it visible three days later when the person is heads-down on something else entirely.

This is worth sitting with, because most "how to write better action items" content quietly assumes the writing is the bottleneck. In teams that run 15-20 meetings a week, it isn't. The bottleneck is that action items get captured in a format optimized for note-taking and then expected to function as a task management system, which it was never built to be.

A formula that actually holds up

The formula itself is still worth using — it just isn't the whole fix. A commitment that survives contact with a real week looks like this:

  • Owner — one person, not "the team" or "whoever has time"
  • Verb + deliverable — "send the updated proposal," not "follow up on proposal"
  • Deadline — a real date, not "soon" or "before next meeting" (which quietly means nothing if the next meeting slips)
  • Context — one line on why it matters, so it survives being read out of order weeks later

The context line is the one people skip most often, and it's the one that saves the most time. "Sarah: send revised MSA to legal by Thursday" is fine. "Sarah: send revised MSA to legal by Thursday — client wants to sign before their fiscal year closes" tells whoever picks it up in a rushed inbox scan exactly why it can't slide. Six months from now, when someone searches your meeting history wondering why that deal moved fast, that line is the answer.

What AI actually gets right about detecting action items — and where it still trips

This is the part most vendor blog posts gloss over, so it's worth being straightforward about it. Meetbook (and every AI notetaker like it) listens for commitment language in a transcript, cross-references who's speaking using speaker labels, and pulls out a candidate list of action items, decisions, and takeaways using an LLM pass over the call. It's genuinely good at the clean cases: "I'll get you the deck by Friday," said by one identified speaker to another, with a clear deliverable. That's a solved problem now.

Where it still needs a human glance:

Implicit commitments. "I'll take a look" is not the same speech act as "I'll fix it by Friday," but both sound like commitments to a model trained to catch commitment language. Good systems flag the softer ones as lower-confidence rather than silently promoting them to hard action items with a fake deadline attached.

Overlapping speech and crosstalk. When two people talk over each other — common in the last five minutes of a call when everyone's trying to wrap up — attribution gets shakier. Speaker diarization handles most of this well, but the meeting's messiest thirty seconds are usually also its most action-item-dense, since that's when people are rushing to confirm next steps before hanging up.

Sarcasm and hedged language. "Sure, I'll get right on that" said in a flat, defeated tone reads identically to a real commitment on a transcript. Tone carries information text doesn't, and even strong models miss it sometimes.

Decisions disguised as questions. "Should we just go with option B?" followed by silence and a topic change often means the room agreed without anyone saying "yes." A transcript alone won't always catch that as a decision.

None of this means AI-detected action items are unreliable — it means they're a strong first draft, not a finished one. The teams that get the most value treat the auto-generated list as something to skim and confirm in ten seconds at the end of a call, not something to blindly trust or blindly distrust. Meetbook surfaces its detected action items, decisions, and takeaways as an editable list right after the call ends, specifically so that ten-second confirmation is possible before anything gets routed downstream.

The real fix: get action items out of the notes document

Here's the part that actually moves completion rates. An action item trapped inside a meeting recap doc competes with everything else in that person's day for attention exactly once — when they read the recap, if they read it. An action item routed to where the person already works competes for attention every time they open that tool.

That means:

  • Sales follow-ups go to the CRM, attached to the deal record, not left in a transcript nobody reopens. When a rep is prepping for the next call with that account, the last meeting's open items are sitting right there on the opportunity.
  • Team commitments go to Slack, posted to the relevant channel or DM'd to the owner directly, so the person sees it in the tool they already check forty times a day instead of a doc they check once.
  • Cross-functional items go where the work actually happens — a Notion page the team already references, an email thread the client is already watching.

This is the difference between "documenting" an action item and actually assigning it. Meetbook pushes detected action items to Slack, CRM records (Salesforce, HubSpot), Notion, and email automatically after each call, which sounds like a small mechanical step but is the entire reason the action item graveyard problem exists in the first place — items get written down beautifully and then routed nowhere.

There's a second layer to this too: stale action items don't just need a home, they need to be findable months later. "What did we agree to do about the pricing objection in that call with Acme back in March?" is a question a semantic search over your meeting history can answer in seconds — a keyword search for "pricing" across a folder of docs usually can't, because nobody phrased it that way at the time.

Tracking without becoming a nag

Once action items are routed correctly, the remaining problem is stale ones — items that got assigned, then quietly died because the deadline passed and nobody circled back. A few habits fix most of this:

  1. Open with open items, not a fresh agenda. The first two minutes of a recurring meeting should be a quick pass through what's still outstanding from last time, not a jump straight into new business.
  2. Use three states, not two. Done and not-done hides the difference between "haven't started" and "blocked waiting on someone else." Add a blocked status and you'll actually see where things are stuck.
  3. Roll over, don't re-write. An item that slips a deadline should carry its history forward — how many times has this moved? — rather than getting silently recreated as if it's new. Three rollovers on the same item is a signal worth raising out loud.
  4. Review weekly, not per-meeting. Chasing every action item after every call is exhausting and ignorable. A five-minute weekly sweep across everything open catches what's actually falling through.

The template, if you want to steal it

OwnerCommitmentDeadlineContextStatus
PriyaSend revised pricing to legalThu 3pmClient needs signed contract before quarter closeIn progress
MarcusConfirm API rate limits with engFri EODBlocking the integration timeline discussionBlocked — waiting on eng
DanaShare updated onboarding deckMonNew hire starts next weekNot started

Nothing exotic here — the value isn't the columns, it's actually keeping status current and reviewing it on a rhythm. A perfectly formatted table nobody opens is exactly as useless as a recap doc nobody rereads.

The honest version of this advice is that most teams don't have an action-item-writing problem. They have an action-item-abandonment problem, and it happens in the gap between the meeting ending and the work resuming. Fix where the item lives, not just how it's phrased, and the completion rate takes care of itself.


Further reading: see our guide on how to choose an AI meeting notetaker for a broader look at transcription and reporting accuracy, or compare notetakers directly in Otter.ai alternatives and Fireflies alternatives.

Stop taking notes. Start free today.

ZoomVisaUberFedExeBayCoca-ColaZoomVisaUberFedExeBayCoca-Cola