,

AI Meeting Note Takers Compared: What They Miss

Three containers of different shapes each filled to a different level, representing three ways of limiting the same service

AI meeting note takers all promise the same thing and ration completely different resources. Otter limits transcription minutes. Fireflies gives unlimited transcription but limits storage and AI credits. Fathom gives unlimited recording, transcription and summaries on its free plan and limits feature depth instead.

That’s why free-tier comparisons mislead so badly. Three “free” plans, and the amount of actual work you can do on them differs by an order of magnitude depending on whether your bottleneck is meeting length, history, or how much analysis you want afterwards.

Key takeaways

  • Fathom’s free plan is the most generous on the core job — unlimited recordings, transcripts and basic summaries.
  • Otter’s free plan caps meetings at 30 minutes and 300 minutes a month, which most weekly calls exceed.
  • Fireflies caps storage, not transcription — 400 minutes on free means your history disappears fast.
  • Accuracy is the shared weak point: crosstalk, accents and jargon degrade every one of them.
  • Recording consent law is your problem, not the vendor’s. Some jurisdictions require everyone’s permission.

What Do the Main Note Takers Cost?

Prices cluster tightly at $15–20 a seat annually. The plans they buy do not.

Tool Free plan Entry paid (annual) What’s rationed
Fathom Unlimited recordings, transcripts, basic summaries Premium $16/user ($20 monthly) Advanced AI features
Fireflies Unlimited transcription · 400 min storage · 20 AI credits Pro $10/seat ($18 monthly) Storage + AI credits
Otter 300 min/month · 30 min per meeting Pro $8.33/user ($16.99 monthly) Transcription minutes
Read off each vendor’s pricing page on 11 August 2026. Fireflies Business is $19/seat annual; Otter Business $19.99/user annual; Fathom Team $15/user annual with a two-user minimum.

The 30-minute cap is the one that catches people. On Otter’s free plan a standard hour-long client call is cut in half, and the half you lose is usually the half with the decisions in it.

Which Free Plan Actually Gets You Through a Month?

Take a realistic week: four meetings, averaging 45 minutes. That’s 180 minutes a week, roughly 780 minutes a month.

Tool Does the free plan cover it? Where it stops
Fathom Yes Nowhere on volume; advanced summaries and Ask limited
Fireflies Partly Transcribes all of it, stores only the first 400 minutes
Otter No Runs out at 300 min, and truncates every call at 30 min
Our arithmetic against each vendor’s published limits. A heavier meeting load widens the gap rather than narrowing it.

If you only ever test free plans for a week, all three feel identical. The differences appear in week three, which is exactly when switching has become annoying. This is the same free-tier-wall problem we mapped across categories in the free AI stack.

How Accurate Are They Really?

Good on a clear one-to-one call and noticeably worse in the conditions real meetings happen in. Every tool in this category shares the same weaknesses, because they share the same underlying speech recognition challenges.

Condition Effect on the transcript What to do
Two people talking over each other Words dropped, speakers merged Chair the call; one voice at a time
Strong or less-common accents Higher error rate, silent substitutions Check names and numbers manually
Industry jargon, product names Confidently wrong words Add a custom vocabulary where supported
Poor microphone or speakerphone Whole passages unusable Headsets matter more than the tool choice
Numbers, prices, dates Plausible but wrong digits Never action a figure without checking
The last row is the one that costs money. A summary that says “£4,500” when the call said “£14,500” reads perfectly.

The summary layer inherits every transcription error and adds its own. It compresses, and compression drops caveats — “we could do that if the client approves” becomes “we will do that”. Read summaries as prompts for your memory, not as a record.

Do You Need Permission to Record?

Often yes, and this is the part vendors leave to you. Recording law varies by jurisdiction: some require only one participant’s consent, others require everyone’s. In the US that distinction runs state by state, and the stricter rule generally applies when participants are in different states.

The practical answer is simple and costs nothing: announce it. Say at the top of the call that you’re recording and taking AI notes, and let people object. Most bots also join visibly with a name, which helps but is not the same as asking.

If you serve EU customers there’s a second layer. The transparency obligations that took effect on 2 August 2026 cover AI systems interacting with people, and disclosure is the recurring theme — worth reading alongside the AI disclosure rules for small businesses. Consent for recording and disclosure of AI use are separate duties, and you may owe both.

Which One Should You Pick?

Match the tool to the constraint that will actually bite you first.

If you… Pick Because
Want the best free option Fathom Unlimited recordings and transcripts at $0
Need long searchable history Fireflies paid Storage is the thing it rations; paid removes it
Transcribe uploads and interviews Otter Built around transcription rather than meetings
Want CRM sync and deal tracking Fathom Business Field sync and deal view are on that tier
Run mostly short internal standups Any free plan Every free tier covers 15-minute calls
Handle sensitive client matters Reconsider entirely The transcript is a discoverable record
That last row is a real decision, not a hedge. Legal, HR and medical conversations create records with consequences.

What Do They Not Replace?

Deciding what mattered. A note taker captures everything said, which is not the same as capturing what was agreed — and the gap between those two is where follow-through lives.

The businesses that get value from these tools all do one extra thing: someone spends two minutes after the call turning the summary into owned actions with dates. Skip that and you’ve automated the production of documents nobody reads, which is a classic automation mistake — measuring output instead of outcome.

How Do You Get More Out of the Transcript?

The default output of every one of these tools is a wall of text and a generic summary. That’s the least useful form the information can take, and it’s why so many teams stop opening them after a month. Three habits change the return substantially.

First, ask the tool a question rather than reading the summary. All three have some form of conversational query over past calls. “What did we promise on delivery dates?” returns something actionable; scrolling a transcript does not.

Second, standardise your meetings slightly. Say “decision:” or “action:” out loud before the important bits. It feels stilted for two meetings and then becomes normal, and it gives the summariser a signal far stronger than anything it can infer from tone.

Habit Time cost What it changes
Say “action” or “decision” aloud Zero Summaries pick up commitments reliably
Query instead of reading 30 seconds Turns an archive into an answer
Two-minute post-call triage 2 min Owners and dates attached while fresh
Push actions into your task tool Automated Stops the notes being a dead end
Delete recordings on a schedule One setting Limits how much history exists to leak
The last row is easy to skip and worth doing. A four-year archive of client calls is a liability nobody decided to create.

Third, connect the output to where work actually happens. A summary that lands in your task manager gets acted on; one that sits in the note taker gets forgotten. That’s a five-minute automation on any of the platforms in Zapier, Make or n8n.

Frequently Asked Questions

What is the best free AI meeting note taker?

Fathom, on the core job. Its free plan includes unlimited recordings, transcriptions and basic AI summaries, where Otter’s free plan caps you at 300 minutes a month with a 30-minute limit per meeting and Fireflies stores only 400 minutes.

How much do AI note takers cost?

Entry paid plans cluster around $8–16 per user per month on annual billing: Otter Pro $8.33, Fireflies Pro $10, Fathom Premium $16. Monthly billing runs roughly $17–20. Business tiers are $19–25 per seat annually.

How accurate are AI meeting transcripts?

Good on clear one-to-one audio, noticeably worse with crosstalk, strong accents, jargon or speakerphones. Numbers and dates are the highest-risk errors because a wrong figure reads perfectly. Verify any price or date before acting on it.

Often yes. Some jurisdictions require only one participant’s consent and others require everyone’s, and the stricter rule generally applies when participants are in different places. Announcing the recording at the start of the call is the simple, safe practice.

Is Otter or Fireflies better?

They ration different things. Otter limits transcription minutes, so it suits lighter or upload-based use. Fireflies gives unlimited transcription but limits storage and AI credits, so it suits high meeting volume where you don’t need a long archive on the free tier.

Can AI note takers join any meeting?

The major tools join the main video platforms, usually as a visible participant. Dial-in phone calls, in-person meetings and locked-down corporate calls are where coverage gets patchy, so check against the platforms you actually use.

The Bottom Line

Pick by the limit that will bite you, not by the price. If you take four hour-long calls a week, Otter’s free plan fails in week one and Fathom’s doesn’t. If you need to search a year of conversations, storage is your constraint and Fireflies’ paid tier is the fix.

Then treat every transcript as a draft. Check the numbers, check the names, and spend two minutes converting the summary into actions with owners. The tool captures the meeting; it does not do the thinking about what the meeting was for.

Announce that you’re recording. It costs one sentence, resolves most of the legal question, and is the same courtesy you’d want. More tool comparisons in our comparisons hub, or see whether you already pay for this twice.

Want one practical automation you can set up in 15 minutes, twice a month? Join the free newsletter.

Sources

All plan limits and prices read off each vendor’s own pricing page on 11 August 2026. The monthly-usage scenario is our own arithmetic against those published limits. Accuracy observations describe well-documented limitations of automatic speech recognition rather than a controlled test, and recording-consent law varies by jurisdiction — check your local rules.