LiveLoop makes WhatsApp AI transcripts a separate choice from recording

LiveLoop’s [October 29 changelog](https://liveloop.one/docs/changelog) adds recorded WhatsApp calls and requested AI transcripts. Its documentation distinguishes the recording provider from the service that turns the recording into text. The [calling documentation](https://liveloop.one/docs/channels/whatsapp‑calling) says calling is beta, off by default and requires account and inbox enablement. Full WhatsApp Cloud API numbers are required. [LiveLoop recording](https://liveloop.one/docs/channels/whatsapp‑calling#recording‑and‑transcripts) enables a notice by default; Meta can optionally auto‑transcribe. LiveLoop transcripts are requested separately after account setup. An operator looking for a transcript therefore needs to identify more than whether a call appears in history. Which service recorded it, does a recording exist, and has the chosen transcription route been enabled? A call record and a completed text transcript are different artifacts. The default notice describes a configured product behavior. It is not evidence that a particular recording meets every applicable consent requirement. OpenTools did not place a call, activate an inbox or assess a customer’s recording practices.

A LiveLoop transcript uses two AI credits per started minute of recording. The cost appears before the request; credits are taken after creation, with none taken on failure.

The word started matters when estimating a request. A recording that runs beyond a whole‑minute boundary moves into another charged minute, rather than being rounded down to its completed minutes. The request should be evaluated using the actual recording length and the displayed cost, not the time someone expects to spend reading the text.

This is an announced product charge, not a measured bill from an OpenTools test. A missing transcript can reflect a request that has not been made, account setup or a failed creation; it does not by itself establish that the call contained no speech. Check which stage produced the available artifact before drawing that conclusion.

The [provider choice](https://liveloop.one/docs/channels/whatsapp‑calling#recording‑and‑transcripts) also affects labels: LiveLoop recordings name the agent’s turns, while a Meta recording shares one track and gives LiveLoop no speaker labels. That is a useful distinction for anyone extracting follow‑up work from a conversation. Text without speaker labels needs a different review from a transcript that identifies the agent. Neither format, by itself, establishes that every word or intended commitment has been captured correctly. Before using the text to prepare a follow‑up, an operator can compare an important passage with the recording and confirm who said it. That is a proposed review approach, not an accuracy test or a claim that LiveLoop resolves every ambiguous conversation. The release describes access and workflow changes; OpenTools did not test transcription quality, language detection, billing or a customer deployment. Archival photograph of a call‑center workplace in São Paulo, Brazil, December 23, 2012. It shows people gathered at a workstation, not a LiveLoop or WhatsApp deployment, transcript or observed test. Photo: [Carlos Ebert / Wikimedia Commons](), [CC BY 2.0](https://creativecommons.org/licenses/by/2.0/). Cropped for display.