An AI call summary is a short written account of a phone call that AI generates from the call transcript, recording why the person called, what was said or agreed, and what happens next.
It replaces the notes a person would otherwise type after the call. Summaries land on the call record or in the CRM, so the next agent, a supervisor or a follow-up system can see what happened without listening to the recording.
A summary can only be as accurate as the transcript under it. If the speech recognition misheard an order number or a name, the summary repeats the mistake, so names and numbers that matter are worth checking against the transcript or capturing in their own field.
A useful AI call summary answers four questions in two to five sentences: why the caller called, what was done, what was agreed, and what is still open.
Example from the Retell docs: "The agent called the user to ask about their purchase inquiry. The user agreed to book an appointment next Monday at 10am." The sample payload pairs it with a sentiment of Positive and a success flag of true.
A transcript is the full word-for-word record of the call. Call notes are what a person chose to write down afterward. An AI call summary sits between them: generated from the full transcript like a record, but short and selective like notes, and written the same way on every call.
Every call on Retell AI gets a call_summary field through post-call analysis. When the call ends, an LLM reads the transcript and writes the summary onto the call record, together with the caller's sentiment, whether the call was successful, and whether it reached voicemail, per the Post Call Extraction docs.
call_analyzed webhook (not call_ended, which fires before analysis runs), in the Get Call API, and in mapped CRM fields for Salesforce, HubSpot, Dynamics 365, GoHighLevel and Zoho.Yes. AI transcribes the call with speech-to-text, then a large language model writes a short summary of the reason for the call, what happened and the next steps.
They are as accurate as the transcript and the prompt behind them. Misheard names or numbers carry into the summary, so capture critical details in their own fields and spot-check summaries against transcripts.
No. A transcript is the full word-for-word record. A summary is a few sentences generated from it, covering the reason, outcome and next steps.
In Call History in the Retell dashboard, in the call_analyzed webhook payload, and through the Get Call API. If you connect a CRM, it can be mapped to a contact field.
Retell AI is an AI voice agent platform for contact centers. Each call its agents handle comes with a summary, sentiment and the custom fields you define. See post-call analysis.
How Retell handles audits, data retention and access control.
Table of contents