> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rootly.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Rootly AI in Retrospectives
> Rootly AI auto-generates retrospective sections — summary, impact, root cause — from incident data, Slack channels, and bridge-call transcripts.
## Draft Your Retrospective Instead of Starting from Scratch
Writing a retrospective is one of the most time-consuming parts of incident management. The information almost always already exists — in the incident's data, the Slack channel, and the bridge call — but someone still has to read it all and turn it into prose.
**AI blocks** solve the blank-page problem. Inside the [Rootly retrospective editor](/collaborative-retrospectives/using-the-editor), an AI block is a section — like Summary, Impact, or Root Cause — that Rootly AI drafts for you from the incident's own context. You decide which blocks belong in a retrospective by adding them to a [template](/retrospectives/configuring-templates), and every retrospective created from that template generates those sections automatically.
The goal is simple: get retro owners to a solid first draft in minutes instead of hours, turning a multi-hour writing exercise into a short review — so more retrospectives actually get completed.
## How AI Blocks Work
AI blocks live in the [Rootly retrospective editor](/collaborative-retrospectives/using-the-editor), right alongside your text, data blocks, and Liquid variables.
* **Each block has a purpose-built prompt.** A Root Cause block knows how to write a root cause; an Impact block knows how to describe impact. You don't write the prompt — you just add the block.
* **They pull context automatically.** Every block draws on three sources with no setup required: your **incident data** (metadata, severity, services, timeline), the **incident Slack channel**, and **bridge-call transcripts** (when captured via the Rootly AI Meeting Scribe).
* **They stay live in the document.** A generated block isn't frozen text — you can regenerate it, edit it inline, convert it to plain text, comment on it, and rate it. It behaves like any other part of the document.
* **You stay in control.** Steer how a block writes with custom instructions on the template, see the sources and the prompt behind any block, and edit anything by hand at any time.
## Key capabilities at a glance
| Capability | Description | Where |
| :------------------------------ | :---------------------------------------------------------------------- | :------------------------ |
| Preset AI blocks | Six ready-made section types, each with an expert prompt | Template builder + editor |
| Custom AI block | Write your own section with a title and a plain-language prompt | Template builder + editor |
| Rootly AI-powered templates | Add AI blocks to a template so every retro generates them automatically | Template builder |
| Custom instructions | Steer how each block (or the whole template) is written | Template builder |
| Preview against a past incident | See exactly what a block produces before rolling out the template | Template builder |
| Insert on the fly | Add an AI block to any retro document with the slash menu | Editor |
| Regenerate & edit | Regenerate any block, edit it inline, or convert it to plain text | Editor |
| Sources & prompt | Inspect what a block drew on and the prompt behind it | Editor |
## The AI block types
Rootly ships six preset blocks, plus a custom block for anything else:
| Block | What it drafts |
| :------------------- | :---------------------------------------------------------------------------------------------------------------------- |
| **Summary** | A concise overview of what happened, the impact, and how it was resolved — written for a broad audience. |
| **Impact** | Who and what was affected — customers, services, scope, and duration. |
| **Root Cause** | The underlying cause and contributing factors. |
| **Mitigation** | The immediate steps taken to reduce or stop the impact. |
| **Resolution** | How the incident was fully resolved. |
| **Curated Timeline** | A readable, narrative timeline of the key moments — distinct from the raw Timeline data block, which lists every event. |
| **Custom AI block** | Any section you define yourself with a title and your own prompt. |
Learn how to add and configure these in [Building AI Templates](/ai/ai-in-retrospectives/building-ai-templates), and how to work with them in a document in [Using AI Blocks](/ai/ai-in-retrospectives/using-ai-blocks).
## How permissions work
AI blocks respect your existing Rootly retrospective permissions — there's no separate AI permission system.
* Anyone who can edit a retrospective can insert, generate, regenerate, edit, and convert AI blocks.
* **Observers (read-only users) cannot modify AI blocks**. They can't regenerate, edit, convert, or delete them, just as they can't edit the rest of the document.
## Where to go next
Add AI blocks to a template, steer them with instructions, and preview against a past incident.
Insert, generate, edit, regenerate, and give feedback inside a retrospective document.
The editor AI blocks live in — formatting, data blocks, Liquid, and collaboration.
How retrospective templates work overall.
***
## FAQs
Rootly AI in Retrospectives is available in Configuration → Rootly AI. Ask your Rootly Admin to enable it for your workspace. Once it's on, AI blocks will appear in the template builder palette and the editor's slash menu.
From your incident data, the incident's Slack channel, and the bridge-call transcript (when captured via the Rootly AI Meeting Scribe). No other external sources are used.
Yes. Generated content is fully editable — type over it, restructure it, or convert the block to plain text to lock it in. You can also regenerate it at any time.
Rootly AI warns you before regenerating a block you've edited by hand, so you don't lose your changes by accident.
Yes. When you publish or export a retrospective, AI block content renders as static content in the exported document, just like the rest of the doc.
No. Incident context is used to generate your draft and is not used to fine-tune base models.