Getting the words out of a sales call used to mean a transcription service and a two-day wait. Now the problem is the opposite: there are too many ways to do it, most of them are free or nearly free, and the right one depends on where the conversation happened rather than on which tool has the best reviews. This is a guide to picking the route, recording legally, and then knowing what to look for once the text is in front of you.
Where the conversation lives decides the route
Before comparing tools, answer one question: where did the call happen? Each answer has a cheapest route and a catch.
| Where the conversation happened | Cheapest route to a transcript | The catch |
|---|---|---|
| A video meeting on Zoom, Google Meet, or Teams | Turn on the platform's own transcription before the call | Speaker labels are rough and the file lives inside the meeting platform |
| A phone or dialer call | Your dialer's recording plus its transcription, or upload the recording to a transcription tool | Dialer transcripts are usually a team-plan feature |
| In person | Record on your phone's voice memo app, then transcribe the file | You are relying on one microphone in a room, so crosstalk suffers |
| Already text: chat, email, a messaging thread | Nothing. Export it | None, except that it is not a conversation in the sense that matters here |
Most salespeople live in the first two rows, so the four routes below are mostly about those.
The four ways to get a transcript
1. The meeting platform's own transcription
Zoom, Google Meet, and Teams all transcribe meetings now, and for most people on a paid plan it is already included.
In Zoom it is a cloud recording feature, so it has to be switched on in your recording settings before the call, and it does not apply to recordings saved to your laptop. Zoom's own article on enabling audio transcription for cloud recordings walks through where the setting lives on Pro, Business, Education, and Enterprise plans. In Google Meet, transcripts are a button inside the meeting on Business Standard and up, and the finished transcript lands in the organiser's Drive; Google's transcripts in Google Meet page lists which editions have it and who can start it.
What it costs you: nothing beyond the plan you have. The catch: speaker labels are the weak spot, especially when two people overlap, and the transcript lives wherever the platform keeps it, which is usually not where you want it.
2. A meeting recorder
Fireflies, Otter, Fathom, and tl;dv all do the same basic thing. By default a bot joins your call as a participant, records it, transcribes it with better speaker separation than the platform's own feature, and produces a summary with action items. All four now also offer a bot-free mode through a desktop app that captures your computer's audio instead, so nothing extra appears in the participant list, though the record it produces tends to be thinner, with weaker speaker labels and in some cases no stored audio. Most have a free tier that covers a handful of calls a month.
What it costs you: in bot mode, a participant with the tool's name sits in the call, which some prospects notice and a few object to. In either mode the audio and the transcript are processed and stored on the vendor's servers, which matters if your prospect works somewhere with a strict data policy. And the summaries are confident regardless of whether they are right.
3. Your dialer or revenue platform
If your team runs calls through Aircall, Gong, Chorus, or a similar platform, transcripts arrive with every call automatically and roll up across the team. This is the best route if you have it, and the one most people searching this question do not have.
What it costs you: these are priced for teams, not individuals, and they are the manager's tool. The transcript exists, but it exists inside a system built for pipeline review and call scoring at the team level, and you may not be able to get the raw text out.
4. Do it yourself with Whisper
OpenAI's Whisper is a free, open-source speech recognition model you can run on your own laptop, either directly or through a wrapper app built on it. Drop in any audio file and it returns a transcript. The accuracy is competitive with paid tools and there is no per-call fee. Run locally, nothing leaves your machine. Note the distinction, because OpenAI also sells Whisper as a hosted API, and that version sends your audio to their servers like any other cloud tool.
What it costs you: ten minutes of setup the first time, a laptop that can run it in reasonable time, and the speaker labels, which Whisper does not produce on its own. You will be labelling the turns by hand or adding a second tool for diarisation. For a rep who wants transcripts of a few important calls a month and does not want them on anyone's servers, it is the right answer. For a rep who wants every call done automatically, it is not.
Record it legally first
None of the four routes matter if the recording itself was not allowed, and the rules are more specific than most people assume.
In the United States, federal law requires the consent of at least one party to the conversation, which means you can record a call you are on. Eleven states go further and require every party's consent: California, Delaware, Florida, Illinois, Maryland, Massachusetts, Michigan, Montana, New Hampshire, Pennsylvania, and Washington, with four more requiring it for some conversation types. These are often called two-party consent laws, which the Reporters Committee for Freedom of the Press notes is technically wrong, since they require all parties' consent. Its recording guide has a chapter per state and is the reference to check, because the law that applies is usually the stricter of the states on either end of the call. In the EU and the UK, recording a call is personal data processing, which means telling the other person and having a reason.
The practical rule is simpler than the legal one. Say it at the start, every time: "I record my calls so I can take fewer notes, is that okay with you?" Almost everyone says yes. The ones who say no have just told you something useful about how the rest of the conversation will go. If your meeting recorder announces itself when it joins, that counts, but say it yourself anyway.
What a useful transcript has, and the five numbers inside it
A transcript is only as useful as its structure. Two things make the difference. Speaker labels, so you know who said what, and timestamps, so you know when. Without both, the transcript is a wall of text you will read once and never again. With both, it contains five numbers that tell you more about the call than the summary does.
Talk to listen ratio. Your words as a share of the total. This is the one number worth watching if you only watch one. Gong's 2025 analysis of 326,000 sales calls, vendor research but the largest dataset on the question, found that reps talked 57% of the time on deals that closed and 62% on deals that were lost, and that conversion fell once a rep passed roughly 65% of the call. Its older rule of thumb, 43% talking to 57% listening, still gets quoted, but the newer finding is less about a golden ratio and more about a ceiling.
Question count, and time to the first one. How many open questions you asked, and how far into the call the first one came. A discovery call with three questions in it was a pitch.
Longest monologue. The longest stretch you spoke without the other person getting a word in. This is where the pitch hides.
Filler words per minute. Um, like, you know, sort of. A few are normal. A cluster of them inside one long turn means you were improvising.
Pace. Words per minute. Too fast reads as nervous, too slow reads as scripted.
The right ranges depend on the call. A cold call runs hotter, with the rep talking 45 to 65 percent of the time, one real question, and no turn longer than about 30 seconds. A discovery call runs cooler, 35 to 55 percent rep talk, at least five questions, and no monologue past 45 seconds. Those are the scenario-specific targets Closemind scores an uploaded call against: you drop in the recording, it is transcribed, the two speakers are separated with a step where you confirm which one is you, and the five numbers come back alongside a rubric score with your own lines quoted as evidence.
The confirm-the-speaker step matters more than it sounds. If the labels are swapped, every one of those five numbers is inverted, and the report would be praising the prospect for asking good questions.
The mistakes that waste a transcript
- Recording without saying so. Even where it is legal, a prospect who finds out later feels tricked, and the recording was not worth that.
- Transcribing everything and reading nothing. A folder of a hundred transcripts is not a habit. Pick the call that mattered this week and read that one.
- Trusting unconfirmed speaker labels. Every automatic labeller gets some calls wrong, and a swapped label makes every number lie.
- Counting keywords instead of reading the turns. The summary tells you the call mentioned pricing. The transcript tells you that you brought it up, at minute two, before the problem was on the table.
If you wanted the transcript to get better
Most people want transcripts for notes, the CRM, or a manager who asked, and the four routes above cover all of that. If the reason was to review your own calls and get better, skip the reading. Upload the recording to Closemind, confirm which speaker is you, and the five numbers come back with a rubric score and coaching that quotes your own lines back to you. Then take the same call into a live roleplay against a counterpart who objects the way yours did, and run it again. The first minutes are free, no card required.
