> ## Documentation Index
> Fetch the complete documentation index at: https://docs.vendexlabs.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Templates & playbooks

> How templates, playbooks, evidence review, and AI drafts fit together in an assessment.

VendexLabs' AI features follow one rule: **cited facts, not guesses.** Every answer and every drafted sentence traces back to a source you can open. Two building blocks, both under **Assessments** in the sidebar, decide what the AI checks and how it writes.

## The flow

**1. A template sets the questions**

A [template](https://docs.vendexlabs.com/guides/templates.md) is your question set: the controls you expect a vendor's evidence to cover.

**2. Evidence review answers them**

[Evidence review](https://docs.vendexlabs.com/guides/evidence-review.md) reads the vendor's SOC 2 report and answers each template item with a verdict, a confidence level, and a quote checked against the PDF.

**3. A playbook sets the voice**

A [playbook](https://docs.vendexlabs.com/guides/playbooks.md) tells drafts how your team writes: what to lead with, how to phrase conditions, how long to be. It's picked automatically by the vendor's tier and tags.

**4. AI drafts the comments**

[Draft with AI](https://docs.vendexlabs.com/guides/draft-comments-with-ai.md) combines the reviewed evidence, your research and business context, and the playbook into security, compliance, and privacy comments, each fact with a source chip.

**5. You decide**

You edit, save, and set the approval status. The [audit log](https://docs.vendexlabs.com/guides/audit-log.md) records it.

## Evidence, strongest first

When a draft needs a source, VendexLabs prefers the strongest evidence it has:

1. **Verified** quotes from the vendor's own report, with a page number.
2. **Cited research** from enrichment, linked to the public page it came from.
3. **Your assessment fields**, such as business context and data collected.
4. An explicit **gap** when nothing covers the question. Gaps are stated, never filled in.

## What the AI won't do

* Invent a citation. Sources it wasn't given are dropped.
* Decide for you. Only the compliance comment suggests an outcome, and setting the approval status is always a person's action, including from the [Claude connector](https://docs.vendexlabs.com/integrations/claude.md).
* Use the [risk score](https://docs.vendexlabs.com/concepts/risk-scoring.md) as evidence. Drafts cite facts, not the score.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.