# DocAssist

DocAssist is an AI Agent that helps you maintain accurate, up-to-date documentation by analyzing your support conversations and identifying documentation gaps. Instead of manually reviewing documentation to find missing or outdated content, DocAssist automatically detects patterns in your support interactions and generates specific, actionable suggestions for improving your knowledge base.

## What DocAssist Does

DocAssist reviews resolved support conversations and compares them against your documentation to identify:

* Missing information frequently asked by customers
* Outdated instructions that no longer reflect product behavior
* Unclear explanations that lead to repeated questions
* Gaps in coverage for specific use cases or scenarios

When a meaningful gap is found, DocAssist provides:

* The exact location to update
* Proposed content
* Clear reasoning behind the recommendation

## How DocAssist Works

* DocAssist treats your configured **Knowledge Sources** (for example, Zendesk Help Center, Confluence, or public documentation sites) as the **official documentation** it evaluates and helps improve.
* **Support conversations** (from Slack request or triage channels, email, Zendesk tickets, etc.) are used only as **input context** for analysis.\
  They are **not treated as documentation** and are never quoted or edited by DocAssist.
* If you also index Slack channels as a Knowledge Source, use **tags** and the AI Agent’s **Pick Specific** option to control what DocAssist evaluates:
  * Include only your documentation repositories (for example, `tag=zendesk-kb`)
  * Exclude Slack Knowledge Sources so suggestions don’t quote Slack threads as official documentation
  * Slack channels can still be part of the **Collections** where DocAssist runs (so conversations are analyzed), while the **targets for improvement** remain only the selected documentation Knowledge Sources.

### Conversation Analysis

When triggered, DocAssist examines the full support conversation and filters out cases that don’t require documentation updates, such as:

* Bug reports
* Engineering issues
* Scheduling or administrative messages
* Incomplete conversations

If no update is needed, DocAssist explains why, allowing teams to move on without manual review.

### Documentation Review

For relevant requests, DocAssist searches your connected knowledge sources to understand the current state of your documentation. It:

* Searches for related articles across your knowledge base
* Compares existing content with the customer's question
* Evaluates whether the guidance is complete, accurate, and clear

When your documentation already addresses the issue adequately, DocAssist returns the relevant excerpts and links, giving your team immediate clarity without manual searching.

### Improvement Suggestions

When gaps are found, DocAssist generates a structured proposal including:

* **Location** – URL and section to update
* **Proposed content** – Ready-to-use text
* **Reasoning** – Why the change is needed

<figure><img src="https://3455705434-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FE2O2wTaNovd6fXpEuLKz%2Fuploads%2FNXvV9JePySjsCgcl4xRE%2Funknown.png?alt=media&#x26;token=aed5b615-61ff-45bb-8976-4f7a01259c05" alt="" width="563"><figcaption></figcaption></figure>

## Creating DocAssist

DocAssist can be created from Agent Studio. See details [here](https://docs.clearfeed.ai/clearfeed-help-center/clearfeed-ai/ai-agents/build-ai-agents)

## Configuration

DocAssist runs entirely through **Automations**, giving you full control over when and how it runs.

### Setup Options

You can configure the below:

* **Knowledge Sources**
* **Agent Instructions**

  * **Helpdesk Context**
  * **Documentation Update Rules**
  * **Response Style**

  <figure><img src="https://3455705434-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FE2O2wTaNovd6fXpEuLKz%2Fuploads%2F4LT8XVMlA1o8NDumKmea%2Funknown.png?alt=media&#x26;token=b5aac43e-f843-4b95-a116-7c72430fc086" alt="" width="563"><figcaption><p>DocAssist deployed with public documentation website as a target.</p></figcaption></figure>

<figure><img src="https://3455705434-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FE2O2wTaNovd6fXpEuLKz%2Fuploads%2FUhiqbEOc2sVAXaIRh3mU%2Funknown.png?alt=media&#x26;token=bf419842-7add-47f1-93fd-38f8d182da24" alt="" width="563"><figcaption><p>Prompt used internally within ClearFeed to configure the Documentation Agent</p></figcaption></figure>

For more details, you can check [here](https://docs.clearfeed.ai/clearfeed-help-center/ai-agents/build-ai-agents#documentation-agent-instructions)

Example:

{% code overflow="wrap" %}

```
## Helpdesk context
This is a documentation improvement agent for Clearfeed's external support helpdesk agent. ClearFeed is a conversational helpdesk built for teams that support customers and employees directly in Slack.  The tickets in this helpdesk consist of questions from Clearfeed's customers about various aspects of the product like feature requests, bug reports, pricing related inquiries, etc.

## Response style
Keep responses concise and actionable. Use bullet points for multiple suggestions. Focus on specific improvements.

## Documentation update rules
  1. Users often report bugs in the product and tickets get resolved after we have fixed the bug. THESE BUGS don't have to go into the documentation!
  2. When users come in with a feature request (or point out a limitation of the system) - tickets may be solved after a commitment to fix or when agents mention they will prioritise the request. These should also NOT make it to the documentation.
  3. Do not suggest the same change across different documents, as it would be redundant - if you are providing multiple proposals for the same ticket they should cover different topics.
```

{% endcode %}

## Testing DocAssist

Before enabling DocAssist for live tickets, test it on historical conversations to ensure suggestions are relevant and actionable.

## Recommended Automation Flows

### Primary Flow: Trigger on Ticket Resolution

The most effective approach is to invoke DocAssist immediately after a ticket is marked as solved, when the complete context and resolution are available.

* Trigger: Request is Updated → Status is changed
* Conditions:
  * Status = Solved
  * (Optional) Add filters for request type, channel, tags, or customer segments
* Delay: 0 minutes (run immediately)
* Action: Invoke AI Agent → Documentation Improvement Agent
  * Include clear instructions such as: Review the conversation and identify outdated or missing sections in our documentation. Provide suggestions with specific sections or titles where changes are needed.
* Filter Results:
  * Add condition: {{Action.should\_update\_documentation}} is True
  * This ensures suggestions are posted only when meaningful gaps are detected
* Post Suggestions:
  * Send to a designated Slack channel using {{Action.agent\_response}}
  * Include metadata like request link, triage link etc
* Scope: Select the Collections where this automation should apply

### Fallback Flow: Time-Based Trigger

Some tickets remain open for extended periods due to external dependencies or cross-functional involvement. This fallback ensures you don't miss documentation opportunities.

* Trigger: Request is Created
* Conditions:
  * Status ≠ Solved (only run if ticket is still open)
  * (Optional) Add additional filters
* Delay: Set based on your typical ticket lifecycle (e.g., 4 hours, 8 hours, or 24 hours)
* Action: Invoke Documentation Improvement Agent (same configuration as primary flow)
* Filter and Post: Use the same logic as the primary flow to check for updates and route suggestions

## Best Practices

* **Run DocAssist After Resolution:** The best results come from analyzing complete conversations with full context and confirmed resolutions.
* **Create a Dedicated Documentation Channel:** Route all DocAssist suggestions to a specific Slack channel where your documentation team can review and action them.
* **Set Clear Rules:** Define which ticket types should and shouldn't trigger DocAssist to avoid noise and focus on actionable insights.
* **Review Regularly:** Treat DocAssist suggestions as a workflow item. Set aside regular time to review and implement recommendations.
* **Iterate on Configuration:** Adjust your Documentation Rules and Response Style based on the quality and relevance of suggestions you receive.