How is a call transcript made?
Automatic speech recognition (ASR) turns speech into text, either live or after the call. Each turn is then labelled with who spoke, from separate audio channels for each side or by speaker diarisation, which tells voices apart. Some systems also add timestamps.
A transcript isn’t a recording. It keeps the words but not the tone, pauses or background sounds, and ASR can mishear names, numbers and product terms, especially through a poor microphone or in a noisy room. The W3C’s accessibility guidance describes a basic transcript as a text version of the speech and non-speech audio information needed to understand the content.
Live captions are related but different: they show the words during the conversation, while the transcript is the record kept afterwards.
For example, a support transcript might read “Customer: How do I invite a teammate?” then “Agent: Open Settings, choose Team, then Invite.”
Why do transcripts matter for a website voice agent?
A transcript is the main evidence of what an AI voice agent actually said. Reading transcripts shows wrong answers, questions your docs don’t cover and places where the agent misheard the caller, which separates a knowledge problem from a speech recognition problem.
Transcripts also hold personal data, because callers mention names, email addresses and account details. Decide who can read them, how long they’re kept and how callers are told about them before you launch.
How Kawl Now saves transcripts
Kawl Now saves a two-sided text transcript of each conversation, shown in the dashboard as Customer and Your agent lines. The caller’s side is the voice model’s transcription of their speech, so check it for mishearings before you blame a source. No call audio is recorded.
During a call, the caller can turn on live captions that show the last eight turns. Afterwards, the Conversations tab lists each conversation with its topic, status and a summary line, which is the first 700 characters of the caller’s own words rather than an AI write-up. Very long transcripts keep only their last 100 turns.
Open a conversation to read the full transcript, or fetch it with GET /api/manage/conversations/{id}. The list_conversations MCP tool and the JSON export of your latest 100 conversations leave transcripts out, and the dashboard and API don’t show which sources an answer drew on.
Automatic insights, shown in the dashboard and available over the API and MCP, group recurring questions and issues from recent calls and quote callers’ words, so treat them with the same care as transcripts.
There are no retention settings yet, so transcripts are kept until you ask for them to be deleted. Before each call, the widget tells callers that a transcript will be saved.
Common questions
Is a call transcript the same as a call recording?
No. A recording keeps the audio, and a transcript keeps the words as text. Kawl Now keeps only the text transcript and doesn’t record call audio.
What affects the accuracy of automatic call transcripts?
The audio, the speaker and the vocabulary. Clear speech through a good microphone usually transcribes well, while names, numbers, accents and specialist terms are common sources of error, so check important details against what the caller meant.