Short answer: Choose based on where your meetings actually happen, not on feature lists. Bot-based tools only capture video calls; upload services add a manual step; human services are accurate but slow and expensive. If your meetings happen in person or by phone, meaty records from the phone in your pocket and returns a speaker-labeled transcript, summary, and action items — free to start.
The phrase "online transcription tool" covers at least four very different kinds of product, and most buying guides blur them together. This guide gives you a decision framework instead: seven criteria, in the order they should actually eliminate options. By the end you should know which category fits you — even if the answer isn't us.
What counts as an online transcription tool for meetings?
Before comparing features, sort the market into its real categories. Almost every tool you'll evaluate is one of these four:
- Bot notetakers (Otter, Fireflies, Fathom): an AI assistant joins your scheduled video calls via calendar integration and transcribes them, often live.
- Upload services (Sonix, Trint, Descript): you record audio yourself with something else, then upload the file to a website for transcription.
- Human transcription services (Rev, GoTranscript): professionals type your audio for a per-minute fee, usually within hours or days.
- Phone-first recorders (meaty): you record the conversation directly from your phone — in the room or on a call played out loud — and the tool transcribes and summarizes it afterward.
Each category makes a different bet about where your conversations happen and how much work you'll tolerate. The criteria below expose those bets one by one.
Start with where your meetings actually happen
This is the criterion that eliminates the most options fastest, and it's the one most comparison pages skip.
Bot notetakers assume every meeting worth capturing has a calendar invite and a video link. That's true for some teams. But if your important conversations are client lunches, site walkthroughs, one-on-ones across a desk, sales visits, or interviews in a café, a bot has nothing to join. There is no Zoom link to a conference room.
So audit a week of your real meetings before you evaluate anything. If 80% of them are scheduled video calls, a bot notetaker is a rational choice — our deep dives on Otter for in-person work and what Fireflies does and doesn't capture compare the leading options honestly. If a meaningful share happens in the room or on ad-hoc phone calls, you need a tool that captures through a microphone you physically carry, which means an upload service plus a separate recorder, or a phone-first tool that does both jobs.
The criteria checklist
Here's the full framework at a glance. Score any tool you're considering against each row.
| Criterion | Question to ask | Why it matters |
|---|---|---|
| Capture | Does it record in-person meetings, or only video calls? | Eliminates whole categories immediately |
| Speaker labels | Does it separate who said what (diarization)? | A wall of unattributed text is nearly useless for meetings |
| Timestamps | Can you jump from any line back to the audio? | Verifying quotes and disputed points |
| Summaries & action items | Does it produce notes, or just a transcript? | The transcript is rarely what you actually send to people |
| Setup friction | Calendar integrations? App install? Bot permissions? | Friction is why tools get abandoned after week two |
| Privacy & consent | Where does audio go? Can you record openly and simply? | Legal and trust obligations don't disappear because a tool is convenient |
| Pricing model | Free tier, subscription, or per-minute? | Different meeting volumes make different models cheap |
The rest of this guide works through the rows that need more than one sentence.
Speaker labels, timestamps, and search: the transcript features that matter
Raw accuracy gets all the marketing attention, but three structural features determine whether a meeting transcript is actually usable.
Speaker labels (diarization). A meeting is a multi-voice document. "We'll ship by Friday" means something completely different depending on whether the client said it or you did. Any tool you shortlist should attribute each line to a speaker — this is table stakes now, and most AI tools in every category offer it, but verify it on your own audio, because diarization quality varies with crosstalk and room acoustics.
Timestamps. When someone disputes what was agreed, you want to tap a line and hear the original audio, not scroll through an hour of recording. Timestamped transcripts turn "I think she said X" into a ten-second check.
Search. Three months from now you won't remember which meeting contained the budget number. Searchable transcripts across your whole archive are what make transcription compound in value instead of piling up as files. meaty produces all three — a timestamped, speaker-labeled, searchable transcript — automatically after each recording.
Do you need a transcript, or do you need meeting notes?
Be honest about what you'll do with the output, because it splits the market again.
Journalists, researchers, and lawyers often need the verbatim transcript itself — the exact words are the product. If that's you, transcript quality and export formats should dominate your decision, and human services deserve consideration for anything high-stakes.
Most people in meetings need something else: a summary they can skim, decisions on the record, and action items with owners. Nobody rereads a 6,000-word transcript of a status meeting; they read the five bullets that came out of it. If that's your reality, weight the tools that generate structured notes automatically — see our guide to how AI meeting minutes actually improve follow-ups for what good output looks like.
meaty sits in the second camp by design: after you stop recording, it generates a title, a summary, and an action-item list on top of the full transcript, so the shareable notes exist the moment the meeting ends.
How much setup friction and privacy exposure are you taking on?
Setup friction predicts abandonment better than any feature. Count the steps between "I decided to capture meetings" and "this meeting is captured": account creation, app-store download, calendar OAuth, bot permissions, per-meeting toggles. Bot notetakers front-load integration work; upload services add a recurring manual step (record, export, upload, wait) to every single meeting. meaty's bet is minimal friction: it's a browser PWA — no app-store install — you add it to your home screen, tap record, and that's the whole workflow.
Privacy and consent deserve equal weight. Recording laws vary by jurisdiction, and some places require every participant's consent, so the practical rule is simple regardless of tool: tell people you're recording, every time. Then evaluate the tool's posture. A bot appearing as a participant makes recording visible, which is honest — but it also means a third-party service sits inside calls with clients who never agreed to your vendor choices, and some organizations block unknown bots outright. With a phone-first tool, consent is a sentence you say out loud ("I'm going to record this so I can send us both notes — okay?"), which most people find more natural in a room than a silent bot in a grid of faces. Whatever you pick, read where audio is stored and whether it trains models before you put client conversations into it.
Pricing models: free tiers, per-minute rates, and where the money goes
Transcription pricing comes in three shapes, and your meeting volume decides which is cheapest.
Metered free tiers. Most AI notetakers offer a free plan capped by minutes. Otter's free tier, for example, is around 300 transcription minutes per month with a per-conversation cap (as of July 2026, per Otter's published pricing) — roughly five hours, which a meeting-heavy week can burn through fast. Prices and limits change; check their site for current numbers.
Per-minute human transcription. Human services charge per audio minute: Rev's standard human transcription is $1.99 per minute as of July 2026, per Rev's pricing page — about $120 for a single one-hour meeting. Again, prices change; verify before buying. That rate is unjustifiable for routine standups, but for a deposition, a broadcast interview, or anything where a misheard word has real consequences, paying for human ears is rational.
Subscriptions. Paid AI tiers typically run per-user monthly fees for higher limits. The trap here is paying for seats or integrations you don't use because the free tier's cap forced an upgrade.
meaty is free to start, so you can run real meetings through it and judge the output on your own audio before any money changes hands.
Honest trade-offs: when another type of tool is the better choice
No tool wins every row of the checklist, including ours. Here's where each category genuinely earns its place — and where meaty doesn't.
Choose a bot notetaker if virtually all your meetings are scheduled video calls and you want zero-touch, calendar-driven capture — including meetings you personally skip. A bot attends without you; a phone-first tool can't, because the phone has to be where the audio is.
Choose a human service if you need near-perfect verbatim accuracy for legal, medical, or publication use, and the audio is difficult — heavy accents, poor recordings, technical jargon. AI has closed much of the gap, but for transcripts with legal weight, humans still justify their per-minute price.
Choose an upload service if you already have a pile of existing recordings from other devices that just need transcribing.
And know meaty's own limits. Transcription happens after the recording ends, not in real time — so if you need live captions on screen during the meeting, meaty is the wrong tool. It captures what your phone can hear, so a muffled speakerphone across a huge room will produce a worse transcript than a phone on the table. And it won't attend meetings for you. What it does is capture the meetings every bot misses — the ones happening in front of you — with the least friction of any approach we know.
Frequently asked questions
What is the best online transcription tool for in-person meetings?
For in-person meetings, a phone-first tool like meaty fits best, because bot notetakers can only join video calls. meaty records the room from your phone, then generates a speaker-labeled, timestamped transcript plus a summary and action items automatically. It runs as a browser PWA with no app-store install and is free to start.
Do online transcription tools identify who is speaking?
Most modern AI transcription tools offer speaker diarization — separating the transcript by voice and labeling each speaker's lines. Quality varies with audio conditions like crosstalk and distance from the microphone, so test any tool on a recording of your real meetings before committing. meaty includes speaker labels on every transcript by default.
How much does human transcription cost compared to AI?
Human transcription is priced per audio minute — Rev charges $1.99 per minute as of July 2026, per its pricing page, so a one-hour meeting costs roughly $120. AI tools typically offer free tiers or flat subscriptions instead. Prices change, so check vendors' sites. Humans are worth it for legal or publication-grade accuracy; AI wins for everyday meetings.
Is it legal to record a meeting for transcription?
Recording laws vary by country and, in the US, by state — some jurisdictions require consent from every participant. The safe, universal practice is to announce the recording and get agreement before you start, regardless of what your local law minimally allows. It also builds trust: people speak more openly when recording is transparent rather than discovered.
Do I need real-time transcription for meetings?
Usually not. Real-time captions matter for accessibility and for live multilingual meetings, but for notes, decisions, and action items, a transcript delivered minutes after the meeting is just as useful — and post-processing often produces cleaner results. meaty transcribes after recording ends, then adds a summary and action items so the notes are ready to send.
Try meaty free — record your next real meeting from your phone and judge the transcript on your own audio, not a demo.