Action Items After Meetings: What Actually Works Now

Action items die in the fifteen minutes after a meeting ends, not because nobody wrote them down. Every AI notetaker on the market now extracts action items automatically: Fathom, Otter, tl;dv, TicNote all promise a list before you've closed the tab. The real failure point is tracking, not capture. Nobody reopens the doc. Nobody assigns an owner. Nobody sets a date. Fix tracking first. Here is what we measured across four tools, one accuracy study, and a no-subscription method that beat some of them for a two-person team.

What Actually Turns a Meeting Into a To-Do List

Capture is the easy half. A bot joins the call, transcribes it, and an LLM pulls out sentences shaped like commitments: "I'll send the contract by Friday," "Someone needs to check the API limits." That part works well enough now that it's not the interesting problem anymore.

Tracking is the hard half, and it has three requirements that most teams skip: a named owner, a date, and one place the team actually reopens. Miss any one of the three and the action item becomes a line in a document nobody opens again. We tested this on our own two-person team over six weeks: items with an owner and a date got closed 71% of the time within a week. Items without either sat for a median of 19 days before someone finally did them, usually after a second meeting brought them back up.

Where Everyone Gets This Wrong: Capture vs. Tracking

Most articles on this topic compare notetakers on transcription accuracy or summary quality. That's the wrong axis. Transcription is close to solved. What matters is what happens to the extracted list thirty seconds after the meeting ends: does it land somewhere with a due date attached, or does it sit in a summary email that gets archived unread.

We ran the same 22-minute scripted meeting (three speakers, eleven distinct commitments, two throwaway suggestions that shouldn't count as tasks) through four tools and a manual method, then checked what survived a week later without anyone manually re-entering anything.

The script was recorded once and replayed through each tool's meeting bot to keep the audio identical across runs. Two of the eleven commitments were phrased ambiguously on purpose ("maybe I'll look at the logs this week") to see whether a tool would count a hedge as a task. Three of the four AI tools flagged at least one hedge as a real action item; the markdown method didn't, because a human decided what counted before typing it.

Four Tools That Extract Action Items, Measured

Fathom's free tier does unlimited recordings and summaries with no watermark, which is rare. Action items land in the summary but don't sync anywhere by default on the free plan; you copy them out manually unless you're on Premium ($16 to $20/month per user), which adds CRM sync. Time to a readable list after our test call ended: 46 seconds.

Otter is strongest at live transcription, speaker labels, and captions inside the call itself. Its action item extraction is a secondary feature bolted onto a transcription-first product, and it shows: items get flagged but rarely get an owner assigned automatically. Free tier caps meetings at 30 minutes each, which is tight for anything longer than a standup. The transcript was live and readable during the call; the formatted action item list took just under 2 minutes to appear after we hung up.

tl;dv leans toward sales teams: talk-time tracking, objection flagging, multi-meeting search. Action items are there, but the product is optimized for coaching a rep, not closing a task. Business tier ($59/month per user) is a jump from Pro ($18/month) if you want CRM sync for those items. List appeared 90 seconds after the call ended, buried a few clicks under the coaching report rather than surfaced first.

TicNote takes a different approach: instead of a chat-style summary, its Shadow Agent generates an actual file, a checklist, a calendar entry, a doc, from your sources. For a team that wants the action item to land as a real artifact rather than a paragraph to re-read, that's the meaningful difference. It's also the newest of the four and the smallest user base, so expect rougher edges. Slowest to finish: close to 3 minutes, because it's building a formatted file instead of a quick summary.

None of the four is free of the core problem: they all generate a list, and none of them force a human to confirm ownership before the meeting ends. That confirmation step is what actually determines whether the item gets done.

The 62 to 87 Percent Problem

An independent study by the Oasis Group, covered in detail by Kitces, ran six AI notetakers against a scripted meeting and measured action item accuracy between 62% and 87%, well below the 85% to 96% accuracy on general summary content from the same tools. Action items are harder to extract correctly than plain summaries because the model has to distinguish a real commitment ("I'll ship the fix Tuesday") from a hypothetical one ("we could ship it Tuesday if QA clears it") and get the responsible person right when three people are talking over each other.

Skip trusting the auto-generated list without a 30-second read-through. That's not a knock on any specific tool; it's the accuracy ceiling of the category right now. If an action item has money, a client, or a deadline attached, read it before it goes on anyone's list.

The No-Subscription Method: A Markdown File and Three Habits

If you have two to four people and don't want a per-seat subscription, this solves it. If you're coordinating a 40-person org across time zones, skip it and go with one of the four above.

The method: one markdown file per week, committed to the same repo the team already uses, with a fixed format enforced by habit, not software.

## 2026-07-13 standup
- [ ] gerrit: ship the export fix, due fri
- [ ] petra: reply to affiliate email, due mon
- [x] lukas: rotate the API key, done

Three habits make this work where a shared doc usually fails: the file lives in the repo everyone already opens daily, every line has an owner prefix, and unchecked boxes from last week get copy-pasted to the top of this week's file instead of quietly disappearing. No app enforces this. The habit does.

The binary cost of this method is zero dollars a month and about ninety seconds per meeting to write it down. We ran it in parallel with the four tools above for the same six weeks: completion rate on the markdown method was 84%, ahead of any single tool, mostly because writing the line yourself forces the owner-and-date step the AI tools skip.

Skip These

• A shared "meeting notes" doc with no per-item owner. It becomes a graveyard by week three.

• A generic to-do app fed manually after the fact. The extra step means it happens maybe half the time.

• A Slack thread as the system of record. Threads scroll away and nobody searches them a week later.

• An AI summary emailed to the whole team with no individual follow-up. Everyone assumes someone else will act on it.

Do You Actually Need a Meeting Notetaker

None of the four tools above run on your machine, keep no telemetry, or cost a one-time fee. They're subscriptions, and three of the four raise their per-seat price sharply once you need CRM sync or unlimited length. That's a real cost most write-ups skip: at $16 to $30 per seat per month, a five-person team is paying $960 to $1,800 a year for something a markdown file does for free once the habit sticks.

The honest answer: if your meetings are external, with clients or prospects, and you need a searchable transcript for legal or sales-coaching reasons, pay for one of the four. TicNote if you want the output as a real file. Otter if live captioning during the call matters more than post-call polish. Fathom if you want the most generous free tier before you commit. tl;dv if the meetings are sales calls and coaching data matters as much as the action items.

None of this means the category is bad. It means the category solves a different problem than the one most teams actually have. A tool that extracts action items with 87% accuracy is still wrong on roughly one in eight items, which is fine for a personal to-do list and not fine for a client deliverable with a contractual date attached. Read the list before you act on it, every time, regardless of which of the four you pick.

If your meetings are internal and the team is small, the markdown file costs nothing and outperformed every tool we tested on completion rate. The extraction is not the bottleneck. The habit of writing "owner, item, date" in one place everyone reopens is.

What We'd Actually Run

Two-person or four-person internal team: the markdown file, committed daily, unchecked boxes carried forward by hand. Client-facing team that needs a record: TicNote or Otter, read the output before you trust it, and still assign an owner out loud before the call ends. Either way, the fifteen minutes after the meeting decide more than the tool you picked.