> For the complete documentation index, see [llms.txt](https://docs.limecall.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.limecall.com/ai-receptionist/summaries-and-tagging.md).

# Summaries & tagging

How each call is written up afterwards.

Open **AI Receptionist → Summaries & Tagging**.

Every call the assistant handles is written up. This is what your team reads instead of listening to a recording.

## How call summaries are written

Set what the summary should contain and how it should read.

A useful summary answers, in order: who called, what they wanted, what was agreed, and what needs to happen next. Anything else is padding.

You can instruct it to always surface specific things — a quoted price, a promised callback time, a competitor mentioned.

Keep summaries short. A summary nobody reads because it is six paragraphs long has failed at its only job.

## AI write-up versus transcript

Three things are produced per call:

| Artefact       | What it is for                      |
| -------------- | ----------------------------------- |
| **Summary**    | The quick read. What happened.      |
| **Transcript** | The exact words, searchable.        |
| **Recording**  | The audio, if recording is enabled. |

Most people read summaries and only open the transcript when something is disputed or unclear.

## Call tags

Tags categorise calls so you can filter and report on them. Define the set that matches how you actually think about your calls — "new patient", "rebooking", "complaint", "supplier", "wrong number".

The assistant applies them automatically from the content of the conversation.

Keep the list short. Twenty tags means inconsistent tagging and useless reports; six well-chosen ones get used.

## Sentiment

**How you judge sentiment** sets what counts as a good or bad call for your business. This is genuinely business-specific — a raised voice in a complaints line is normal, while the same tone on a booking line is a problem.

Sentiment feeds the escalation rules. If **Escalate frustrated callers automatically** is on, this setting decides what "frustrated" means.

## Collected fields

Any field defined on a call type appears on the record as structured data, separate from the prose summary. These are what integrations and reports read, so they are worth defining properly rather than relying on the summary text.

## Where it all appears

On the call record under **Calls**, on the lead under **Leads**, and in the conversation in your **Inbox**.

If you have a CRM connected, the summary and fields sync there too. See [Integrations](/integrations.md).