T
Teach AI Tools
🏠 Home → 📰 Blog & Guides → 🎓 Courses → 👤 About Sourabh → ✉️ Contact → ⭐ Sponsor → 🗺️ Sitemap →
Our AI Platforms
🎮 AI Game Dev Tools → ⚡ LLM Pulse → 📈 FinTech AI Terminal → 🛡️ Cyber AI Terminal → ⚖️ AI Governance Dashboard → 🖥️ MLOps & AI Infrastructure →
AI Tools · Meeting assistant comparison

Fieldy, Otter, Fireflies and Fathom: AI Meeting Assistants Compared in 2026

Compare Fieldy, Otter, Fireflies and Fathom by price, capture method and privacy. Includes wearable versus desktop choices and a worked accuracy-check example.

The most useful AI meeting assistant is the one that produces a follow-up you can approve quickly: the correct decision, the correct owner and a deadline that was actually agreed. A readable summary is only the starting point. If “Friday, if finance approves” becomes “launch on Friday,” the assistant has changed the plan.

Fieldy AI is an AI wearable that captures conversations, generates summaries, and turns important action items into reminders. It can be worn as a watch or pendant and also offers a desktop app for online meetings. Shortlist it for authorised in-person capture when its hardware and transcript-centred workflow fit your needs. For free desktop capture, consider Fathom; for transcript review, Otter; for a shared meeting library, Fireflies. Check existing Zoom or Teams entitlements before adding another subscription.

These are recommendations based on documented features and access requirements, not measured accuracy rankings. The original comparison's prices and sources were checked on October 14, 2026; Fieldy was added and researched on October 15, 2026. No Otter, Fireflies, Fathom or Fieldy accuracy test was performed; the worked transcript below is an original fictional test fixture, not output from any vendor.

Commercial disclosure: the previous version displayed promotional Descript and ElevenLabs placements. Those ad blocks have been removed from this article. Neither placement provides evidence about the meeting tools compared here. The source links below are ordinary research references.

Choose by the work you need after the call

Start with the destination of the notes. A freelancer who sends one approved email needs a different workflow from a sales manager searching hundreds of calls. A tool that is convenient in one setting can add unnecessary administration in the other.

Option Suggested first use What should decide the trial?
Fieldy Wearable capture for in-person meetings and interviews; desktop capture also offered Device cost, included minutes, phone/sync requirements and transcript-only review
Fathom Individual meeting capture and basic summaries Capture availability on your device; whether free summaries are sufficient
Otter Reviewing and correcting a searchable transcript Monthly minutes, meeting length and file-import allowance
Fireflies A shared library across meeting platforms Storage, language needs and access to the paid workflow features
Zoom Workplace built-in meeting AI Teams already using an eligible paid Zoom plan Included features and your administrator's settings
Teams intelligent recap Organisations working primarily inside Teams Premium or Copilot entitlement, transcription and meeting policies

My first shortlist option for wearable, in-person capture is Fieldy. For a solo worker who needs free desktop capture, consider Fathom Free; for an organisation already paying for a native assistant, start with that existing entitlement. Move to a separate paid product when the pilot reveals a specific missing requirement, such as imports, exports or a shared searchable history. This avoids buying a second assistant simply because its demo summary looks polished.

Keep the meeting host's role separate from the note recipient's role. Work out who must capture the call, who reviews the summary and who needs ongoing access to the library. Do not automatically purchase one paid seat for every attendee. Equally, a single shared login is a poor substitute for named access and an accountable reviewer.

Verified prices and limits that affect the bill

The table uses public dollar prices. Monthly means a monthly subscription; annual means the advertised monthly equivalent with annual billing, not a cancellable monthly price. Taxes, currency conversion, negotiated contracts and extra usage are outside these figures. Sources: Otter pricing, Fireflies pricing, Fathom pricing and Fieldy memberships. Fieldy's hardware cost is separate from its membership.

Plan Monthly billing Annual billing equivalent Main allowance or restriction
Fieldy Free membership $0 after device purchase $0 after device purchase 150 transcription minutes/month; hardware is not free
Fieldy Unlimited membership $19.99/month $15.99/month Device sold separately; unlimited monthly transcription; broader chat, search and export access
Otter Basic $0 $0 300 transcription minutes monthly; 30 minutes per conversation; 3 lifetime file imports
Otter Pro $16.99/user/month $8.33/user/month 1,200 in-app recording minutes monthly; 90-minute conversations; 10 monthly file imports
Fireflies Free $0 $0 Limited AI summaries; 400 storage minutes/team; transcription offer has qualifications and rate limits
Fireflies Pro $18/seat/month $10/seat/month 8,000 storage minutes/seat; unlimited AI summaries; 20 included AI credits; 2-hour recording limit
Fireflies Business $29/seat/month $19/seat/month Unlimited storage; 30 included AI credits; 3-hour recording limit
Fathom Free $0 $0 Unlimited recordings and transcripts; basic instant summaries
Fathom Premium $20/user/month $16/user/month Adds advanced summaries, AI action items and a conversational assistant
Fathom Team $19/user/month $15/user/month Minimum 2 users; team collaboration and shared search
Fathom Business $34/user/month $25/user/month Minimum 2 users; CRM field sync and coaching features
Team reviewing a client video call with notes and follow-up tasks
AI-generated editorial illustration of team reviewing a client video call with notes and follow-up tasks.

Fieldy hardware: the device order page displayed Fieldy 3 at $129 on sale, with $179 crossed out, when checked October 15. The displayed subtotal excludes tax. That is an observed promotional price, not a guaranteed future price. At $129 plus twelve months at the advertised annual membership equivalent, the illustrative first-year subtotal is $129 + ($15.99 × 12) = $320.88, before tax and any extras. Confirm the actual annual charge and local checkout terms before paying.

Otter's regular rates above are explicitly listed in USD. Its page also shows a discount restricted to India-issued cards; that promotion is excluded from the comparison rather than presented as a worldwide price. A checkout offer may differ from the regular rate, so record the renewal price as well as the introductory amount.

The important distinction is recording allowance versus retained storage. Eight 75-minute calls total 600 minutes. That exceeds Otter Basic's monthly allowance, and each call also exceeds its 30-minute conversation cap. Pro's listed in-app allowance and 90-minute cap fit that particular recording pattern. Imported files have a separate allowance, so this example does not establish that every upload workflow fits Pro.

A storage allowance answers a different question: how much meeting history can remain available? Ten people each capturing four hours weekly generate 2,400 minutes a week if those are distinct recordings. Duplicate capture of the same call can increase the library unnecessarily. Count unique recordings and decide the retention period before estimating a team's storage needs.

For a two-seat comparison, Fireflies Pro costs $36 monthly, or $240 for a year at the advertised annual rate. Fathom Team costs $38 monthly, or $360 for a year at its annual rate. These are subscription arithmetic examples, not equivalent-feature packages. Fathom's minimum makes Team an unsuitable one-seat budget assumption even though its annual per-user rate is lower than Premium's.

Fieldy: a wearable option for face-to-face conversations

Fieldy deserves consideration when the conversation happens around a table or during an interview rather than entirely inside a video-meeting platform. Its official product page presents Fieldy 3 as a device worn as a pendant or on the wrist, with transcription, summaries, tasks and reminders. The useful distinction is the capture format: you are considering physical hardware alongside an AI service. This guide has not tested microphone pickup, battery life or transcription accuracy.

For an interviewer or consultant who wants less manual note-taking during a permitted conversation, that is a practical reason to shortlist Fieldy. It is less compelling if your existing desktop assistant already captures all the meetings you need. For online use, Fieldy Desktop lists macOS on Apple Silicon and Windows, with transcripts synced across devices. Check your computer's compatibility rather than assuming the Mac download supports every Mac.

What the included membership covers

The membership comparison lists five AI-chat requests daily and transcript export as TXT on Free; Unlimited adds unlimited chat and PDF/Markdown exports. It also marks Always-On capture as a paid feature. The included minutes can therefore support a limited pilot, but “Free” describes a membership allowance after buying hardware, not a free wearable or unrestricted all-day capture.

For two 45-minute interviews a month, the planned total is 90 minutes, inside the listed 150-minute allowance. Four such interviews total 180 minutes, above it. These are workload calculations; retries, other captured speech and your actual usage need to be counted. Plan around the conversations you are permitted to capture, rather than recording continuously just because an unlimited tier exists.

The documented capture-to-summary workflow

The Fieldy app guide documents these steps: tap Start Transcribing, end the conversation with End Transcribing, then open the saved conversation and tap Summarize. This is a documentation-based walkthrough, not a test performed for this guide. The same guide explains that sharing a conversation exposes its summary and transcript to anyone holding the share link; Make Private disables access. Review the content and recipients before creating that link.

Live wearable processing uses the phone connection described in Fieldy's FAQ. The vendor's offline-storage announcement also describes recording locally when disconnected, then syncing for transcription after reconnecting. Offline capture is therefore different from fully offline AI processing. Capacity claims differ across the vendor's pages; confirm the current limit for your exact device and firmware before relying on an extended disconnected session.

The review trade-off: transcripts rather than replayable audio

Fieldy's privacy policy says personal-user audio is deleted when no longer needed for processing and is not available to the user; transcripts remain. It names external speech/AI processors, so this is not an entirely local workflow. The policy also describes an exception for organisation agreements, under which transcripts and audio may be delivered to the organisation. Verify which arrangement applies to your account.

That distinction matters for this guide's accuracy protocol. If you need to replay the original wording of a disputed amount or quotation, the documented personal-user workflow does not supply that audio. For a controlled accuracy trial, use a separately authorised reference recording; for routine follow-up, explicitly confirm important owners, amounts and conditional decisions with the participants. Do not treat a transcript as proof of what was agreed.

My recommendation: shortlist Fieldy for authorised in-person capture when its wearable format solves a real problem and a transcript-centred record is sufficient. Prefer a workflow retaining accessible source audio when replay is essential. Fieldy's advertised capabilities justify a trial; they do not establish that it is more accurate than Otter, Fireflies or Fathom. The result for Fieldy remains not tested below.

AI features, languages and speaker names

Treat these as documented capabilities to evaluate, not evidence that one system recognises every voice better. Transcription language support, translated summaries and a conversation that switches languages are separate requirements.

Tool Documented feature to evaluate Language detail Check in your own pilot
Fieldy Conversation summaries, speaker identification, reminders and AI chat Vendor lists 100+ languages Recognition of your speakers and terminology; whether transcript-only review fits the job
Otter Speaker identification, editable speaker tags, AI meeting notes Pricing page lists English, Spanish, French, German, Japanese and Chinese Names attached to the right remarks, especially similar names
Fireflies AskFred, action items, searchable meeting history Advertises transcription in 100+ languages; Business lists Multi-language Mode Correct language handling and useful answers grounded in the call
Fathom Summaries, Ask Fathom and paid AI action items Older language help lists 38; October 2026 upgrade guidance states English plus 38 additional languages for the new experience Your actual version, rollout eligibility and summary language

The Otter and Fireflies rows are supported by their linked pricing pages; Fieldy's capabilities come from its membership comparison. Fathom's language help and Fathom 3.0 upgrade guidance describe different product stages. The newer guide says multilingual availability is rolling out in phases. Choose using the experience available to your account rather than assuming every language feature has reached every user.

Otter: choose it when the transcript is your working document

Otter is a sensible shortlist candidate when you routinely return to the wording of a discussion. In the trial, find one changed amount, correct one speaker label and locate a disputed decision without reading the entire transcript. Record how long these tasks take. Those review tasks are more relevant to your purchase than whether the default summary uses attractive headings.

Prepare a small vocabulary list before an important call: participant names, product names and abbreviations. Then check those terms in the output. An assistant recognising ordinary speech does not establish that it will recognise a client's unusual company name or attach it to the correct person.

Fireflies: choose it when meeting history needs to be shared

Fireflies is worth shortlisting when multiple colleagues need to find earlier discussions. Build a pilot around a real retrieval question: “What is the latest approved amount, and where was it corrected?” Compare the assistant's answer with the source passage before writing anything to a client record.

Do not confuse included AI credits with storage minutes. Give one person responsibility for checking which operations consume credits and how extra usage is billed. A plan with enough retained audio may still require a separate budget for the AI operations your workflow uses.

Fathom: choose it for a low-cost capture pilot

Fathom Free is a useful first step when the immediate requirement is capturing meetings and reviewing basic summaries. Use Premium or Team as a response to a demonstrated need for richer follow-up or collaboration. During the pilot, export a useful excerpt and ask the intended recipient whether it provides enough context to act.

Device limitation: Fathom's bot-free capture is a Mac beta in the checked pricing and upgrade documentation. The new 3.0 experience is not promised for Windows yet. Its upgrade guide distinguishes transcript-only, audio-only and bot-based audio/video capture. Pick the available capture mode deliberately; a transcript-only mode still processes what people say and still needs an appropriate privacy process.

Speaker separation also deserves its own check. Several voices sharing one conference-room microphone create a different input from several attendees joining on separate devices. Write down your audio setup beside the result. A participant list alone is not proof that every utterance belongs to the right person.

Check your existing Zoom or Teams licence first

For organisations already using one meeting platform, the first purchase question is whether another service adds enough value to justify its cost and another copy of the meeting data.

Zoom Workplace meeting AI

Zoom's current naming needs care. Its official explanation of ZoomMate and AI Companion says core features such as meeting summaries and transcription remain built into eligible paid Zoom Workplace plans at no additional cost. They now appear under functional names. ZoomMate is a newer add-on product; it is not necessary merely to retain those included features.

The buying recommendation is therefore to inspect your existing entitlement before purchasing either a standalone note taker or a ZoomMate add-on. Ask your administrator whether meeting summaries are enabled for the actual host account. Do a short permitted pilot, check how recipients receive the recap and verify that a guest can follow the approved action list without receiving unnecessary access to a recording.

Teams intelligent recap and Copilot

Microsoft's meeting guidance states that intelligent recap is available through Teams Premium or a Microsoft Copilot licence. This is a licensing route, not a claim that every Teams user receives it free. Copilot availability also depends on organisation policy and meeting settings; access to Copilot from the Recap tab depends on transcription.

Ask IT for the incremental cost for your current tenant, including any required base licence. A standalone headline price would be misleading for someone who already owns the entitlement through a bundle. Confirm supported language and recap features for the specific meeting configuration, then compare the same review tasks you would use for Otter, Fireflies or Fathom.

A native assistant may simplify access administration because the organisation already manages that platform. It does not automatically settle retention, external sharing or permission to transcribe a client's call. Make those decisions explicitly whichever route you choose.

Worked example: a corrected amount and a conditional launch

Here is an original fictional script you can use to design a permitted trial. It is intentionally short enough to check line by line. Maya and Mia are different people. The meeting is dated October 14, 2026; dates in the script are stated explicitly to avoid ambiguous “next Friday” interpretation.

Maya: The draft campaign budget is $12,000.

Mia: Correction: the budget we agreed is $10,500. The $12,000 figure is from the older draft.

Maya: Use $10,500 in the proposal. Leo, can you send the revised proposal on Tuesday, October 20?

Leo: I am away. Mia will own that task instead.

Mia: Agreed. I will send the revised proposal on Tuesday, October 20.

Maya: We could launch on Friday, October 23, if finance approves. Finance has not approved it yet.

Leo: Keep that launch date provisional. We have not agreed a fallback date.

The reference answer below was written from this authored script. It is an answer key for the example, not a reconstructed tool report.

Item Correct interpretation Error that changes the follow-up
Latest budget $10,500 for the proposal Keeping $12,000 as the current figure
Proposal owner Mia Assigning it to Maya or Leo
Proposal deadline Tuesday, October 20, 2026 Replacing it with the launch date
Launch status Friday, October 23 is conditional on finance approval Calling Friday a confirmed launch
Approval and fallback Finance has not approved; no fallback date agreed Inventing approval or a replacement date

Original example showing a confirmed proposal task beside a provisional launch with finance approval still pending; not a vendor screenshot.

An approved follow-up for this script could read: “Use $10,500 in the revised proposal. Mia will send it on Tuesday, October 20. Friday, October 23 remains a possible launch date, subject to finance approval. There is no agreed fallback date.” It preserves both the correction and the open dependency.

Notice that a summary can contain all the right numbers and still get the decision wrong. That is why checking keyword presence is inadequate. Review the relationship between the amount, the owner, the date and the condition. A missing condition is an important error even when most of the surrounding text is correct.

A repeatable accuracy trial, with results left unfilled

The publication status is explicit:

Product Transcript accuracy Owners, dates and conditions Observed review time
Fieldy Not tested Not tested Not measured
Otter Not tested Not tested Not measured
Fireflies Not tested Not tested Not measured
Fathom Not tested Not tested Not measured

To run a useful comparison, use the same authorised source and record the plan, capture mode, language, date and settings for each product. If one product processes an uploaded recording and another joins a live meeting, disclose that difference. Input routes can affect the comparison, and a result from one route should not be silently generalised to the other.

  1. Write the reference first. Use the five checks above, then add requirements relevant to your work. Include a corrected number, a reassigned owner and a decision that remains open.
  2. Capture with permission. Use a short fictional meeting with willing participants. Keep the audio source and speaking order as consistent as practicable. Avoid client material for an initial setup test.
  3. Save the first output. Preserve the untouched transcript and summary before fixing anything. Record whether you supplied names or vocabulary in advance.
  4. Grade each item separately. Mark it correct, incorrect or omitted. Keep an invented commitment as a separate serious-error flag instead of hiding it in an average.
  5. Time the review. Measure from opening the output to approving the follow-up. Include speaker corrections and re-listening to unclear audio.
  6. Repeat on representative calls. Add your usual microphone setup, accents and terminology. One clean scripted call can reveal a failure, but cannot establish a universal accuracy rate.

A simple results record should include the source passage, the tool's actual wording, the reference interpretation, the correction and the review time. Save timestamps when available so another reviewer can understand why an action was changed. Do not publish participants' real voices or names merely to make a comparison look more authentic.

If you publish your own findings, label the scope precisely: number of calls, total minutes, languages, recording environment and test date. “Three scripted calls in English” is useful evidence; “the most accurate assistant” would require a much stronger comparison. A vendor marketing percentage is also not a substitute for your own observed result.

Recording rules vary by jurisdiction and by the circumstances of the conversation. A visible bot or a software notification is not a universal substitute for the permission or legal basis your meeting requires. Cross-border calls need attention to participants' locations and the relevant rules; the software does not decide which law applies.

For a concrete US distinction, 18 U.S.C. §2511(2)(d) contains a federal exception for a participant or prior consent from one party, subject to the criminal or tortious-purpose exception. State requirements can differ. For example, the published California Penal Code §632 text addresses recording confidential communications without consent from all parties. Its definition and exceptions matter; this is not a rule that every public conversation always requires the same consent.

For UK organisational processing, the ICO's lawful-basis guidance requires an appropriate lawful basis and transparency. Consent is one possible basis, not the only one. A friendly verbal request does not replace the organisation's work to identify and document the applicable basis. Obtain advice for your particular jurisdiction and regulated or sensitive use; these examples are not a complete legal checklist.

As a practical meeting policy, give advance notice of the tool, purpose, recipients and retention period. Offer manual notes if capture is inappropriate or someone objects. Here is an illustrative notice, with a deliberately chosen retention period that must match your actual policy:

I would like to record and transcribe this call using [tool] for internal follow-up. Access will be limited to [named group], and we plan to delete the recording after 30 days. Is that acceptable? We can use manual notes instead.

Confirm more than recording permission before enabling automatic joining. Set which calendars are in scope, who can share links, whether external guests can access a recap and who can delete retained material. Also check what data is sent to integrations. An approved summary for a colleague does not imply permission to expose the full call to a wider workspace.

Treat vendor security badges as information to investigate alongside contracts and your intended use. Ask the responsible administrator about processing terms, data location, international transfers, deletion and any regulated-data requirements. Transcript-only capture still creates meeting content, and removing the audio does not erase the privacy implications of the transcript.

Turn the reviewed notes into a decision you can use

Finish every reviewed call with three sections: confirmed decisions, actions with owners and dates, and open questions or dependencies. Link an uncertain item to its source passage and ask the participants to clarify it before updating a contract, CRM commitment or delivery schedule.

For the worked example, the amount belongs under confirmed decisions; Mia's proposal belongs under actions; finance approval belongs under open dependencies. The possible launch date can be listed beside that dependency, clearly marked provisional. This format makes the uncertainty visible to someone who never attended the call.

Measure economic value after that review. If 20 calls require eight minutes of manual follow-up each, the baseline is 160 minutes. If the assistant also requires eight minutes per approved follow-up, it has not reduced that part of the workload. Track observed review time and useful retrieval separately before attributing a saving to the subscription.

Which should I try first?

Start with Fieldy on the shortlist for in-person wearable capture when hardware cost and a transcript-centred record fit your needs. Consider Fathom Free for suitable desktop capture, Otter for transcript review and defined minute allowances, or Fireflies for a shared library and language needs. If your organisation already has eligible Zoom or Teams features, pilot those first. These picks concern workflow fit; the unperformed accuracy contest has no winner.

Is a free plan enough for client calls?

It can be enough for a limited workflow, but match the call length, monthly usage, import route, required exports and retention to the actual plan. Permission, sharing policy and review remain necessary whichever plan you select. Start with a permitted internal test before inviting a client.

Should AI write directly to a CRM?

For the initial rollout, require approval of commitments and important fields. A corrected amount or conditional date can look like a simple field update while changing the commercial meaning. Expand automation only after checking representative outputs and assigning someone responsibility for reviewing exceptions.

The practical purchase decision is to choose the smallest plan that fits your capture and access needs, then judge it by the follow-up you can approve. Keep the source evidence, preserve unresolved conditions and compare the work that remains after the AI has produced its draft.

T
AI Tools Assistant