How to transcribe a conversation: the complete 2026 guide

The right way to transcribe a conversation is decided before you speak, not after. The single choice that determines whether you get clean, speaker-labeled text for free or fight a noisy recording later is where the conversation happens: a video call, a phone call, an in-person talk, or a file you already have. Most people get this backwards. They record first on whatever device is handy, then go hunting for a transcription tool, which is exactly why they end up with garbled "Speaker 1 / Speaker 2" text, missing context, and a legal exposure they never thought about.

So here is the short version. For the most common case (online meetings), the genuinely best approach is to capture text live at the source rather than record audio and upload it afterward. Live capture gets per-speaker attribution and structured notes automatically, and you never re-record your own screen. For everything else, pick the capture method that matches where the conversation lives, then worry about the tool. This guide walks each path with exact steps, real costs traced to the source, and the consent rules that actually apply.

What "transcribe a conversation" really means

Three things get lumped together, and keeping them straight saves you a lot of grief:

• Transcription is the full spoken text, ideally with speaker labels and timestamps. This is the searchable record.
• Captions are the live on-screen text during a call. They scroll away and are not saved unless you also transcribe or record.
• Summary / notes are the distilled version: decisions, action items, and key moments pulled out of the transcript.

You usually want all three, but they come from different features and different tiers. The transcript is the foundation; captions and summaries are downstream of it.

The decision that changes everything is not which app to buy. It is matching your capture method to the setting before the conversation starts.

Before you record: is it legal?

Skipping this step is the most expensive mistake on the list. Here is the general framework for the United States (this is information, not legal advice; rules differ by country and by state).

The federal baseline is one-party consent. Under the Wiretap Act (18 U.S.C. section 2511), a person who is not acting under color of law may record a communication when they are a party to it, or when one party has given prior consent, as long as it is not done to commit a crime or a tort. (Source: Cornell Legal Information Institute, 18 U.S.C. section 2511, as of June 2026.) That is where the phrase "one-party consent" comes from: if you are in the conversation, federal law generally lets you record it.

The catch is the states. Roughly 11 states require all-party consent, meaning everyone on the call has to agree. These include California, Florida, Illinois, Pennsylvania, and Washington. On a call that crosses state lines, the safest assumption is that the strictest applicable rule controls.

Method 1: online meetings (the most common case)

This is where most conversations that need transcribing actually happen, and it is where people overpay or get stuck. Built-in transcription exists on the big three platforms, but it is gated behind specific paid tiers.

• Google Meet built-in transcription works only on certain paid Workspace editions (Business Standard and Plus, Enterprise tiers, Teaching and Learning Upgrade, Education Plus, and Workspace Individual). The transcript saves to the meeting organizer's Google Drive, and only spoken words are captured (chat is excluded). Free personal Google accounts get neither recording nor transcription. (Source: Google Meet Help, as of June 2026.)
• Zoom cloud-recording transcription requires a Pro, Business, Education, or Enterprise account, applies only to cloud recordings (not local ones), and saves the transcript as a VTT file. You have to enable both cloud recording and audio transcription in account settings first. (Source: Zoom Support, as of June 2026.)
• Microsoft Teams transcribes via More actions, then Record and transcribe, then Start transcription. Transcription also starts automatically when you record. Live translated transcription needs the organizer to have a Teams Premium license. (Source: Microsoft Support, as of June 2026.)

So the built-in path works, but only if you already pay for the right tier, and the transcript lands wherever the platform decides (the organizer's Drive, the Zoom cloud, the Teams chat). That matters for privacy: whoever controls that location can see the record.

For Google Meet specifically, there is a cleaner option that does not require a paid Workspace edition: a browser-based notetaker that captures the conversation live. Scribbl runs as a Chrome extension, records, transcribes, and summarizes Meet calls in the browser with no bot joining the meeting, and is free for individuals for up to 15 Google Meet calls a month with no credit card. (Source: scribbl.co, as of June 2026.) Because it captures text live at the source, you get per-speaker labels and automatic notes and action items, and there is nothing to upload afterward.

If you want the exact platform walkthroughs, see how to record a Google Meet, the Google Meet recording transcript guide, getting a transcript from a recorded Zoom meeting, and recording a meeting in Teams.

Method 2: phone calls and in-person conversations

When the conversation is not on a video platform, you are capturing audio with a microphone, and speaker separation becomes the weak point. Two practical paths:

On an iPhone, Apple Voice Memos transcribes recorded speech on iPhone 12 and later, and you can watch the transcription live while recording or read it afterward. It supports English, Spanish, Portuguese, Italian, French, German, Japanese, Korean, and Simplified and Traditional Chinese, though it is not available in every region. (Source: Apple Support, as of June 2026.)

On Android or any other device, record with the built-in voice recorder or a call-recording app, then run the file through an AI transcription tool (Method 3 below).

The thing that decides quality here is not the model, it is the capture. A single mixed audio track from a phone in the middle of a table is exactly why generic transcribers struggle to tell speakers apart. Put the mic close to the people talking, kill background noise, and ask people not to talk over each other. For phone calls specifically, the consent reminder is sharpest: announce that you are recording before the conversation starts.

Method 3: transcribing a recording you already have

If you already have an audio or video file (an interview, a webinar, a voicemail, a meeting someone else recorded), you upload it to an AI transcription tool and let it run. Before you pick one, check four things:

1. Speaker diarization. Does it label who said what, or hand you a wall of undivided text? For meetings, interviews, and sales calls, this is the difference between a usable transcript and a chore.
2. Supported languages and accents. Confirm your language is supported, not just English.
3. File format and length limits. Match your file type and duration to the tool's caps.
4. Accuracy on your audio. Test on a real clip, not the vendor's demo. (More on why accuracy claims mislead below.)

Cost varies more than people expect, from free open-source engines you run yourself to per-minute paid services. For a breakdown of what transcription actually costs, see the cost of transcription services, and for free options sorted honestly, the best free transcription software guide.

How to get a clean, accurate transcript

The model matters less than the audio you feed it. A few habits raise quality far more than switching tools:

• Reduce crosstalk. One speaker at a time. Overlapping voices are the number-one cause of garbled output and bad speaker labels.
• Use a real mic. A headset or dedicated mic on clean audio beats a better model on bad audio every time. Mic placement and a quiet room win.
• Add custom vocabulary where supported. Names, products, and jargon get mangled by default. Feeding the tool a vocabulary list fixes most of it.
• Proofread the load-bearing parts. Always check names, numbers, dates, and technical terms before you share. Those are exactly the words AI gets wrong, and they are exactly the words that matter.

Turning a transcript into something useful

A raw transcript is a means, not an end. The value is what you pull out of it: the decisions, the action items, the searchable key moments. Tools that capture live (like a notetaker on a Meet call) generate the summary and action items as a byproduct, so you skip the second pass entirely.

Once you have the structured output, the work is sharing it safely (mind who can see the storage location) and tracking what you committed to. For the next steps, see action item tracking to turn decisions into follow-through, and for choosing a transcription tool for recurring meetings, the best meeting transcription software guide.

If your conversations are mostly recurring online meetings with a team, that is where capturing live at the source pays off most: see Scribbl for teams, which adds Zoom and Microsoft Teams alongside Google Meet, and Scribbl for sales if these are customer calls. You can compare plans on the pricing page.

Frequently asked questions

Can I transcribe a conversation for free?

Yes, depending on the setting. For Google Meet, a browser notetaker like Scribbl is free for individuals up to 15 calls a month with no credit card (source: scribbl.co, as of June 2026). On an iPhone 12 or later, Apple Voice Memos transcribes recordings for free. For files you already have, open-source engines run locally at no cost. What you do not get for free is built-in transcription on free personal Google Meet accounts or local Zoom recordings; those are gated behind paid tiers.

What is the most accurate way to transcribe a conversation?

Capture text live at the source rather than recording audio and transcribing it later. Live capture on a video platform preserves who said what (per-speaker attribution) instead of handing a generic transcriber one mixed audio track to untangle. After that, the biggest accuracy lever is the audio itself: a good mic, a quiet room, and one speaker at a time beat any model upgrade.

Is it legal to transcribe a call without telling people?

In the United States, federal law is one-party consent: if you are part of the conversation, you can generally record it (18 U.S.C. section 2511). But about 11 states require all-party consent, including California, Florida, Illinois, Pennsylvania, and Washington, and the strictest applicable rule tends to control on multi-state calls. The simple, safe practice is to announce that you are recording at the start. This is general information, not legal advice.

Can I transcribe an in-person meeting?

Yes. Record it with a good microphone in a quiet room (on iPhone 12 and later, Voice Memos transcribes as you record), then clean up the text. The weak spot is speaker separation from a single mixed track, so mic placement and reducing crosstalk matter more than the tool you choose.

How accurate is AI transcription really?

Better than ever on clean audio, and meaningfully worse on messy audio. Vendor accuracy numbers assume ideal conditions: clear speech, no overlap, no background noise. Real conversations with accents and crosstalk land lower, which is why you should always proofread names, numbers, and technical terms before sharing a transcript.