Fathom AI 評測:免費會議筆記實用,但人名和行動項目需核對
Fathom AI Review: Great Free AI Notes, But Check Every Name
作者實測 Fathom AI,發現會議摘要和數字整理可靠,但轉錄人名、行動項目歸屬及跟進電郵仍會出錯。兩場腳本會議的逐字稿字詞準確率約為 96% 和 92%;Ask Fathom 能指出未有討論的資訊,並連結至錄音中的相關片段。免費方案提供不限次數的錄音、轉錄、儲存及片段;作者建議分享前核對人名,若需要上傳錄音或更準確的逐字稿,可考慮 Otter 或 MeetGeek。
You finish a call, and someone asks, “Wait, who’s sending the proposal?” Nobody took notes because everyone was busy talking. Now you’re scrolling through a recording trying to find the 20 seconds that matter.
That’s the problem AI meeting assistants are built to solve, and Fathom has become one of the most popular. It records Zoom, Google Meet, and Teams calls, then gives you a transcript, a summary, action items, and a follow-up email.
Fathom’s core features are still free, and a lot has changed since my first review. You can now record without a bot joining, use Ask Fathom across your meetings, and connect it to Claude and ChatGPT. In September 2026, it was also acquired by Superhuman, the company formerly known as Grammarly.
To see how good the notes actually are, I ran Fathom through the same scripted meetings I used in my Otter review. That way, I could count every transcription mistake and check whether the summaries matched what was actually said.
In this Fathom review, I’ll show you how it worked, where it impressed me, and where I think it still falls short. I’ll finish the article by comparing it to my top three alternatives: MeetGeek, Otter, and Granola. By the end, you’ll know which tool is right for you.
Verdict
Fathom has one of the most generous free plans I’ve seen in an AI notetaker. Its summaries were clear, it got every number right, and Ask Fathom gave me useful answers without making things up. But it struggled with names, assigned tasks to the wrong people, and generated follow-up emails I’d need to edit before sending. It’s a great starting point, but I’d still check its work.
Pros and Cons
- Unlimited recordings, transcripts, storage, and clips on the free plan.
- Free advanced summaries, action items, and Ask Fathom for five calls per month.
- Clear, organized summaries that got every number right in my tests.
- Ask Fathom admitted when information wasn’t discussed instead of making it up.
- Answers link to exact moments in recordings for easy fact-checking.
- Record meetings on Mac without a bot joining the call.
- Works with Zoom, Google Meet, Microsoft Teams, and Slack Huddles.
- Connects with HubSpot, Salesforce, Slack, Zapier, Notion, Asana, Claude, and ChatGPT.
- Supports transcription in 38 languages.
- Can’t upload existing audio or video files.
- Often misspelled names, sometimes badly.
- Missed entire sentences, including a client’s name.
- Assigned other people’s tasks to me in the summaries and action items.
- AI follow-up emails sounded generic and repeated transcription mistakes.
- If the transcript gets something wrong, Ask Fathom can get it wrong too.
- Adding the notetaker to unscheduled calls took extra setup.
What is Fathom AI?
Fathom is an AI notetaker founded in 2020 by Richard White. It joins your video calls as a bot (or now records without one), transcribes everything, and turns the conversation into notes you can search, share, and act on.
Its free plan, which includes unlimited recording, is a big part of why it caught on. Founder Richard White has said giving away a robust free product drove Fathom’s viral growth. When Superhuman bought it in September 2026, Fathom had more than 400,000 monthly active users.
For now, Fathom still works the same as before. The plan is to connect meeting notes to Superhuman’s email, docs, and AI agents.
What Happens When You Use It
Once you connect your calendar, Fathom finds meetings with a Zoom, Google Meet, or Teams link and can automatically join to take notes.
When the call ends, you get:
- A recording with a full transcript.
- An AI summary organized into a meeting purpose, key takeaways, topics, and next steps.
- AI-generated action items, each linked to the moment it was said.
- A follow-up email you can copy and send.
- Ask Fathom, which is a chat that answers questions about that call or across all your calls.
Bot or Bot-Free
The biggest change since my first review is Fathom 3.0, released in April 2026. It added bot-free capture, which records the meeting from the Fathom desktop app instead of sending a bot into the call.
You can choose bot-free video, audio, or transcript only for each meeting. When I tested it in November 2026, bot-free capture was still a Mac-only beta that worked with Zoom, Google Meet, and Slack Huddles. Check Fathom’s website for current support before you rely on it.
From Notes to Meeting Intelligence
Fathom is also pushing beyond notes.
Paid team and business plans add shared folders, Trackers that flag topics like competitor mentions, CRM field updates, Deal View that brings all meetings together, and AI Scorecards for coaching sales calls. Its Claude and ChatGPT integrations (available on every plan) let you ask about your meetings from inside those chatbots.
Who is Fathom AI Best For?
Here’s who Fathom is best for:
- Freelancers and consultants can record every client call for free, and use Ask Fathom to find what a client said about scope or budget weeks later.
- Sales reps can pull objections and next steps from calls, sync summaries to HubSpot or Salesforce, and automatically update CRM fields on the Business plan.
- Customer success teams can share recordings internally and search across every account conversation.
- Managers can save time catching up on meetings they missed by asking Ask Fathom for decisions instead of watching recordings.
- Recruiters and HR teams can focus on the candidate instead of typing notes, then review the transcript before making decisions.
- Small teams on a tight budget can each record unlimited calls on the free plan, then move to the Team plan when they need a shared call library.
- People who don’t like meeting bots can use bot-free capture on a Mac so nothing extra appears in the call.
Fathom AI Key Features
Here are the key features that stood out to me:
- Unlimited free recording: Unlimited recordings, transcription, storage, and downloads on every plan, including Free.
- AI summaries: Organized into purpose, takeaways, topics, and next steps, with more than 15 templates for different call types.
- AI action items: Pulled from the conversation with timestamps, and you can assign each one to a person.
- Ask Fathom: Ask questions about one call or across your meetings, with linked moments as sources. The free plan only looks back two weeks.
- Follow-up emails: One click writes a recap email from the summary and action items.
- Bot-free capture: Record from the Mac desktop app without a notetaker joining the call.
- Highlights & clips: Mark moments during a call and share short clips by link.
- 38 transcript languages: Including Spanish, French, German, Japanese, Hindi, and Arabic, with translated summaries in six languages on Premium and Team plans.
- CRM sync: Push summaries to HubSpot or Salesforce, and update CRM fields automatically on the Business plan.
- API & MCP server: Bring your meeting data into Claude, ChatGPT, Zapier, Make, or your own tools.
Hands-On Testing: What Fathom AI Produced
To measure Fathom’s accuracy, I reused the two scripted meetings from my Otter review. They’re recorded with AI voices and full of names, prices, percentages, and dates, so I know exactly what was said.
Because Fathom can’t take uploads, I played each recording through my Mac’s speakers into a Google Meet call with Fathom’s notetaker in it. That’s closer to an actual call than a clean upload, but it also means the audio was a little harder for Fathom than it was for Otter. And since every voice came through one microphone, Fathom treated the whole meeting as one speaker: me.
Getting there took some setup.
My Google Meet call wasn’t scheduled, and the notetaker didn’t join on its own. The web app had no way to add it by link, so I added the call to my calendar and installed the desktop app which let me start the meeting with Fathom. The notetaker then waited in Meet’s lobby until I admitted it, and since I was the only attendee, it didn’t start recording until I clicked “Skip & Start Recording Now.”
Test 1: A Three-Person Team Meeting
For my first test, I recorded a three-minute content team meeting between Maya, Dev, and Oliver about traffic numbers, a product launch with uncertain pricing, and an upcoming roundup article.
The script contains roughly 500 words, several deadlines, and names like “Vantora,” “Quillbase,” “Daniel Okafor,” and “MeetGeek.” I also used Fathom’s default Enhanced summary template.

Fathom got most of the roughly 500 words right, a word accuracy of around 96%. The errors came from three places:
- The first sentence, “Okay, let’s get started,” was missing.
- “Should we include tl;dv as well? A few readers have asked about it” became “Should we include?” The rest of that sentence was dropped.
- “Okafor” became “Ockhofer,” and “MeetGeek” became “Meat Geek.”
Every number was correct, including “42,000 visits,” “18%,” “1.9%,” “$29” and “$34,” and every date.
The summary was the strongest part. It had a one-line meeting purpose, four key takeaways, and a breakdown by topic. It even explained why the approval date moved to November 10 (to create a two-day buffer before the launch).

The summary credits me with Maya’s email task (“Janine will email Daniel Ockhofer”). But that isn’t really Fathom’s fault: the whole recording came from my computer, and Maya’s name is never spoken, so Fathom had no way to know someone else said it.
However, the action items are a different story:
- “Rewrite 12 pricing-page meta descriptions… then Janine reviews CTR Nov 18.” The recording says “Oliver reviews the click-through rate on November 18th” out loud, so this was a genuine mistake.
- “Email Daniel… then article ready for review Nov 12.” The article was due for review on Thursday and approval on November 10. Nov 12 is the launch date.
- “Deliver first draft of AI Notetaker Roundup… by Nov 28.” That one was correct.
The follow-up email was a short and friendly team recap. However, it repeated the “Nov 12” mistake and the misspelled names.
My Take
I’d happily read this summary instead of rewatching the meeting, and every number in it was right. But I’d review the action items and follow-up email before sending them. Two of three action items had a wrong owner or date, and the email repeated those mistakes.
If everyone on your team joins from their own computer, Fathom can tell who’s speaking. That should make speaker labels much more accurate than in my test, where everyone shared one microphone.
Test 2: A Noisy Client Call
Next, I played a 1.5-minute client call between Priya (British) and Marco (Italian accent), with café chatter and clinking cups in the background. It covered an email rebrand, a $7,500 budget plus $2,000 for templates, a ten-day legal review, and names like “Elena Kowalczyk.”
Fathom has a Transcription Dictionary to help it recognize names like these, but only team and account admins can access it. I couldn’t test it on my free account.

I started recording late and missed the opening “Hi Marco,” so I excluded it. Of the remaining 307 words, Fathom got the majority right at around 92%.
The background noise didn’t seem to cause much trouble, but names did:
- “Elena Kowalczyk’s team” became “Alina Cowag 16.”
- “Daniel Okafor” became “Daniel R. Kaffer.”
- “Thank you, Priya” became “Thank you for a year.”
- It dropped “talk through the Vantora rebrand before we send you the proposal,” which is the only sentence naming the client and the project.

The summary still got the important facts right.
It listed the following:
- 12% completion rate.
- 25% target.
- The move from Mailchimp to HubSpot.
- The cut from seven emails to four.
- The budget.
It even added it up correctly: $7,500 + $2,000 = $9,500. It estimated the timeline at about four weeks (two weeks of writing plus ten business days of legal review), which is a reasonable estimate.
It created one correct action item: send the proposal and copy legal. It was assigned to the right person but under the wrong name (“Alina”), and the item left out the Monday deadline.
The follow-up email was the weakest output in my testing. It opened with “Hi there!” because Marco’s name never made it into the summary (my own fault). It said “here’s the proposal” as if one were attached, and told the client: “I’ve cc’d Alina Cowag 16.”
My Take
For my own notes, this summary was useful. Every number was correct, and I could see the plan at a glance.
But I’d be careful using it with clients, since it got some names wrong and left out important details. If I’d sent that email without checking it, my client would have seen their legal lead’s name written as “Alina Cowag 16.”
Test 3: Asking Ask Fathom About the Meeting
For my final test, I asked Ask Fathom the same three questions I asked Otter’s AI Chat about the team meeting from Test 1:
Using the team meeting recording, I asked Ask Fathom the same three questions I’d asked Otter’s AI Chat in my Otter AI Review:
- “List every deadline from this meeting, who owns each one, and the exact date.”
- “Draft a short follow-up email from Maya to Daniel asking him to confirm the Quillbase pricing and send the dashboard screenshots. Include the dates that depend on his reply.”
- A trick question: “What budget did the team approve for the AI notetaker roundup, and why did they decide to drop Notta from it?”

For the deadlines, Ask Fathom listed seven with dates.
Surprisingly, it did a better job of identifying who was responsible for each task than it did in the action items: Dev rewrites the meta descriptions, and Oliver reviews the CTR and owns the roundup. It also correctly turned “Wednesday” into November 25 and “Thursday” into November 26.
Ask Fathom also flagged a problem on its own. It pointed out that a November 10 approval date was earlier than the meeting date and might be a transcript inconsistency. In fact, my script was written for a meeting in early November, so that was a fair catch. However, it still gave Maya’s email task to me.

For the email, Ask Fathom told me there was “no ‘Maya’ in this meeting” and wrote it from me instead. That’s technically correct, since no one says Maya’s name out loud in the recording.
The draft itself was well written. It had both prices, the screenshot request, and the Wednesday and Thursday dates. However, it called November 10 the “launch date” when it’s the approval date.

The trick question was the best answer.
Fathom said no budget was discussed and that Notta “wasn’t mentioned at all.” It listed what the team actually decided and linked to the right moment (2:36).
My Take
Ask Fathom was more reliable than its automatically generated action items. When it didn’t know something, it said so instead of making up an answer. That’s exactly what I want from an AI meeting assistant. It even caught a date issue I hadn’t asked about.
But it’s only as accurate as the transcript. If someone’s name was never mentioned, it can’t know it. And if the transcript gets something wrong, Ask Fathom can get it wrong too.
Results & Output Quality
Across both meetings, Fathom was about 94% accurate. Otter made just 3 mistakes on the same scripts, although I uploaded clean audio files to Otter instead of playing the audio through a speaker and microphone.
Fathom’s mistakes weren’t random. It got every number right and struggled with names, while dropping a few full sentences. So the summaries were reliable for facts and figures, but anything involving people needed a second look.
The summaries were the best output. Both were well structured, quick to scan, and accurate on the details I’d actually return to, like prices, percentages, budgets, and deadlines. I’d share either one in a collaboration tool after fixing the names.
Meanwhile, action items and follow-up emails were weaker. One action item gave me a task that both the recording and Fathom’s own summary assigned to Oliver, and another used the launch date as the review date. The emails were friendly, but they read like templates and carried every name mistake forward. You can expect to edit them every time.
I was most impressed by Ask Fathom. It linked to its sources, flagged a date problem I hadn’t asked about, and refused to invent details for the trick question.
In terms of speed, the only real time cost was getting the notetaker into an unscheduled call.
How to Use Fathom AI
- Sign up and connect your calendar. Create a free account with Google or Microsoft. Fathom will find meetings on your calendar that have a Zoom, Google Meet, or Teams link.
- Install the desktop app. This is where you can see upcoming meetings and add Fathom to calls. You’ll also need it if you want to record without a bot joining.
- Set your recording preferences. Choose whether Fathom records all meetings automatically or just external ones. You can also decide if attendees should receive the summary afterward.
- Record your meeting. Join through the desktop app or let Fathom join automatically. If the notetaker is waiting in the lobby, you’ll need to let it in. Make sure recording has started before continuing.
- Check your results. After the meeting, open Fathom to review the summary, action items, transcript, and follow-up email. You can also use Ask Fathom to ask questions about anything you missed.
Top 3 Fathom AI Alternatives
Here are the best Fathom alternatives I’ve tried that I’d recommend.
MeetGeek
MeetGeek is one of Fathom’s closest competitors. Both join Zoom, Meet, and Teams calls and write summaries, but MeetGeek also offers meeting analytics and Meeting Skills that turn discussions into tasks, CRM updates, and follow-up drafts.
The biggest practical difference is uploads. MeetGeek lets you upload your own audio or video files, even on the free plan, and transcribes 100+ languages compared to Fathom’s 38. That makes it a better choice if you also want to transcribe interviews or calls you’ve already recorded.
But the trade-off is volume. MeetGeek’s free plan stops at three hours of transcription a month, while Fathom’s free recording is unlimited. Paid plans are similar in price.
Choose MeetGeek to upload audio or video and get meeting analytics. Otherwise, choose Fathom for unlimited free recording.
Read my MeetGeek review or visit MeetGeek!
Otter
Otter is the alternative I’d recommend if transcript accuracy matters most. On the exact same scripts I used for Fathom, Otter made only 3 word errors out of 802 words, and its custom vocabulary fixed the names it got wrong. Fathom has no custom vocabulary, so “Okafor” stayed wrong.
Otter also accepts uploaded files and shows a live transcript during the call, which Fathom doesn’t. Both can answer questions about your meetings, and both correctly refused my trick question.
But where Fathom wins is free use. Otter’s free plan caps you at 300 minutes a month and 30 minutes per conversation, while Fathom’s free recording is unlimited.
Choose Otter for more accurate transcripts and uploads. Otherwise, choose Fathom for unlimited free calls.
Read my Otter AI review or visit Otter!
Granola
The final Fathom alternative I’d recommend is Granola, which takes a different approach.
Instead of a bot recording everything, it’s an AI notepad that captures your computer’s audio while you jot down notes. After the call, it combines your notes with the transcript, so the final summary focuses on what you thought was important.
Fathom’s new bot-free mode closes the gap, but Granola has been bot-free from the start. It runs on macOS, Windows, iOS, and Android, while Fathom’s bot-free capture was still a Mac-only beta when I tested it.
The biggest difference is what you can save and access later. Fathom keeps full video recordings and clips you can share, while Granola’s free plan limits you to notes from the last 30 days.
Choose Granola for private, bot-free notes you shape yourself. Otherwise, choose Fathom for recordings and clips you can save, share, and access later.
Read my Granola review or visit Granola!
Fathom AI Review: The Right Tool For You?
Overall, Fathom is still the best place to start if you want an AI notetaker without paying. Unlimited recordings, plus advanced summaries, action items, and Ask Fathom on five calls a month, is more than most competitors give away.
What I liked most was the summaries. In both tests, they were well organized and got every number right, including a budget total Fathom worked out itself. Ask Fathom surprised me, too. It refused to invent a budget that was never discussed and flagged a date problem I hadn’t noticed.
Where Fathom really struggled was everything involving people. Names came out as “Ockhofer,” “R. Kaffer,” and “Alina Cowag 16,” a key sentence disappeared, and tasks landed on me. The follow-up emails carried those mistakes forward, including a client email that opened with “Hi there!” Setting up an unscheduled call also took more effort than I expected.
I’d recommend Fathom to freelancers, sales reps, and small teams who mostly meet on Zoom, Meet, or Teams and want searchable notes for free. Just make sure you review names before sharing anything. But if transcript accuracy is more important or you need to upload recordings, Otter or MeetGeek is a better fit. And if you don’t want a bot joining your meetings and prefer more control over your notes, try Granola.
Thanks for reading my Fathom review! I hope you found it helpful. If you want to try Fathom for yourself, you can sign up for free and see what it can build for you.
Frequently Asked Questions
Is Fathom AI free?
Yes, Fathom AI is free. The free plan includes unlimited recordings, transcription, storage, and clips. Advanced summaries, AI action items, and Ask Fathom are free on your first five calls each month, and Ask Fathom only searches your last two weeks on Free.
How accurate is Fathom AI?
In my tests, Fathom transcribed roughly 96% of words correctly in a team meeting and 92% in a noisy client call. Numbers and dates were always right, but unusual names were often misspelled, and a few sentences were dropped.
Is Fathom AI better than Otter?
Whether Fathom AI is better than Otter depends on what you need. Fathom is better if you want unlimited free recording on Zoom, Meet, or Teams. Meanwhile, I found that Otter was more accurate on the same test scripts and accepts uploaded files.
Can Fathom AI record meetings without a bot?
Yes, bot-free capture records from the Fathom desktop app. No notetaker joins the call.
What languages does Fathom AI support?
According to Fathom’s help center, it transcribes 38 languages.
來源:Unite.AI · unite.ai