# The call log Recordings, transcripts and filters for every call, and how a good answer from one call becomes a fixed phrase for the next. Source: https://docs.omazy.ai/how-to/voice/call-log/ import Figure from '../../../../components/Figure.astro' import { Aside } from '@astrojs/starlight/components' Every call is recorded and transcribed. You read them under **Voice**. The call log is a phone log, not a message inbox. Chats live in [Engage](/how-to/engage/inbox/); this screen is only calls, so the columns are the things you actually want on a call: who rang, how long, what was said, and the audio. ## Reading a call Selecting a call gives you the recording and the transcript side by side. The recording gets the width by default, because listening is usually the point. The detail panel starts collapsed and opens when you want the caller and routing information. Play the audio while reading the transcript. A transcript tells you what was said and hides everything about how it landed: the pause before an answer, the caller getting impatient, two people talking over each other. Those are the problems worth fixing and none of them are visible in text. The log updates as calls end. You do not need to reload after a test call. ## What the log works out on its own A few minutes after a call ends, it gets read and labelled: - **An outcome.** One per call, and the specific one the conversation actually supports: Booked, Ordered, Message, Answered, Transferred, Missed, Dropped, Failed, or Needs review. A booking that was discussed but never agreed is not Booked. - **The caller's sentiment**, and Unknown when they barely spoke. - **A one-line summary**, written to be read in a list rather than opened. - **Next steps**, when the call left the business something to do. Empty is a normal answer; nothing is invented to fill the field. **Unclassified** means the pass has not reached that call yet, or there was nothing to read: a call with no transcript and no recording cannot be labelled, and guessing would be worse than saying so. You can override any outcome yourself, and an override sticks. Nothing reclassifies a call a person has already decided about. ## Filtering Filter by date, direction, duration, outcome and language. The combinations that earn their keep: - **Very short calls.** Callers who hung up in the first few seconds. Usually the greeting. - **Very long calls.** The agent going in circles, or a caller who needed a human several minutes earlier. - **By language.** Whether the agent is answering in the language callers open in. ## Turning a good answer into a fixed phrase The most useful thing in the call log is a well-phrased answer the agent already gave. **Save as response** on any agent turn turns it into a saved response.

Two things to do before publishing one: **Edit it.** Speech that read well in context is often too long standing alone, and frequently refers to something the caller said earlier. Cut it to the answer itself. **Decide if it should be fixed at all.** Only make a phrase fixed if the exact wording matters. If it does not, leaving the agent to answer freshly handles the variations of the question you have not thought of. ## Knowing which turns were fixed Turns spoken from a saved response are marked in the transcript. That mark answers a question that is otherwise very hard to answer while reading: whether the agent worked this out, or read it. It matters when something is wrong. A bad fixed answer is fixed by editing the phrase. A bad model-written answer is fixed in the brief or the knowledge. Without the mark you can spend a while editing the wrong thing. ## Recordings and consent Calls are recorded by default. Recording rules differ by country and by state, and several require you to tell the caller. If yours does, the greeting is where that belongs. See [Data and privacy](/workspace/data-and-privacy/).