Source: https://atomic-works-test.docs-staging.pageloop.ai/guides/setting-up-knowledge-for-atom

# Setting up knowledge for Atom

Atom, Atomicwork's Universal AI Coworker, answers employee questions and assists your agents from day one. Adding or connecting the right knowledge to Atom is the first step to making those answers specific to your organization — your policies, your tools, and the way your teams actually work.

## How Atom uses knowledge

Everything you add as knowledge flows to Atom, and Atom puts it to work in two places:

- **Employee answers.** When someone asks Atom a question in Slack, Teams, the portal, the browser, or over email, Atom answers from your knowledge and deflects the request before it becomes a ticket. For example, when someone asks "How do I connect to the VPN?", Atom pulls the answer from your knowledge and replies right there in the channel they're already in.
- **Agent Assist.** When a request does reach a human agent, Atom surfaces the same knowledge inside the request, so the agent can resolve it faster without hunting through docs.

Because the same knowledge base powers both, every improvement you make pays off on the employee side and the agent side at once.

### Knowledge levels

Knowledge lives at two levels, and it's worth knowing which is which before you start:

- **Global (tenant-wide).** Applies across every workspace. Content settings, Acronyms, and Trusted Knowledge are configured here — under **Settings > Configuration (under Atom AI) > Behavior** — and are always global. You can also add documents, URLs, and connected-source content globally to make it available across your whole organization.
- **Workspace (scoped).** The same documents, URLs, and connected sources, added inside a specific workspace so they serve that workspace's audience. Verified answers and Slack channel learning are managed within a workspace.

Most content can go in at either level — you choose based on who should see it. Only Content settings, Acronyms, and Trusted Knowledge are global-only. A good rule of thumb: if it should apply to everyone in the company, add it globally; if it belongs to one team's domain, add it in that team's workspace.

## Ways to add knowledge

You can add knowledge to Atom in several ways:

| Input type                 | Best for                                                                                                                 | Set it up                                                                                                                                                                                                                                |
| -------------------------- | ------------------------------------------------------------------------------------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Uploaded documents**     | Policies, handbooks, and guides you own as files (PDF, DOCX, XLSX, PPTX, CSV, TXT, and more)                             | [Add and organize knowledge](/product/ai-knowledge/knowledge-management/add-and-organize-knowledge)                                                                                                                                      |
| **URLs**                   | Pages that change often — Atom relearns them automatically so you don't re-upload                                        | [Add and organize knowledge](/product/ai-knowledge/knowledge-management/add-and-organize-knowledge)                                                                                                                                      |
| **Connected sources**      | Living documentation in Notion, SharePoint, or Google Drive — synced and relearned on a cadence                          | [Notion](/product/ai-knowledge/learning-sources/learning-from-notion-pages) · [SharePoint](/product/ai-knowledge/learning-sources/learning-from-share-point-documents) · [Google Drive](/integrations/knowledge-management/google-drive) |
| **Slack channel learning** | Turning the answers your team already gives in Slack into reusable knowledge                                             | [Learn from Slack channels](/product/ai-knowledge/learning-sources/learning-from-slack-channels)                                                                                                                                         |
| **Verified answers**       | High-stakes FAQs where the answer must be exact and consistent — these override everything else                          | [Verified answers](/product/ai-knowledge/knowledge-management/verified-answers)                                                                                                                                                          |
| **Trusted Knowledge**      | Vendor questions (Microsoft, Google, Okta, Zoom…) answered from curated public documentation, with no authoring required | [For employee questions](/product/ai-knowledge/knowledge-management/trusted-knowledge-for-employee-questions) · [In Agent Assist](/product/ai-knowledge/knowledge-management/trusted-knowledge-in-agent-assist)                          |
| **Content settings**       | Choosing which categories Atom can answer beyond IT support                                                              | **Settings > Configuration (under Atom AI) > Behavior**                                                                                                                                                                                  |
| **Acronyms**               | Company shorthand, so Atom reads your terms the same way in questions and content                                        | **Settings > Configuration (under Atom AI) > Behavior**                                                                                                                                                                                  |

## Configuring your knowledge

### 1. Connect your channels

Decide how employees reach Atom, and connect those channels first — Slack, Microsoft Teams, the portal, the browser, and email. This is where answers get delivered, and connecting Slack also unlocks **Slack channel learning**, so Atom can learn from the answers your team already gives in real conversations.

See [Using Atom in Slack](/product/ai-knowledge/atom/using-atom-in-slack) and [Using Atom in Microsoft Teams](/product/ai-knowledge/atom/using-atom-in-microsoft-teams).

### 2. Set Atom's global knowledge behavior

These apply across your whole organization, so set them once up front, under **Settings > Configuration (under Atom AI) > Behavior**:

- **Content settings** — choose which categories Atom can answer beyond IT support.
- **Acronyms** — define your company's shorthand so Atom reads your terms correctly in both employee questions and your content. If your device policy says "MBP," make sure Atom knows that means MacBook Pro.
- **Trusted Knowledge** — let Atom answer common vendor questions from public documentation without you writing any of it. If you've connected identity sources like Okta or Azure AD, help centers for provisioned apps are detected automatically.

See [Trusted Knowledge for employee questions](/product/ai-knowledge/knowledge-management/trusted-knowledge-for-employee-questions) and [Trusted Knowledge in Agent Assist](/product/ai-knowledge/knowledge-management/trusted-knowledge-in-agent-assist).

### 3. Organize each workspace into folders

Within each workspace, group related knowledge into folders so Atom retrieves the right content in context. We'd recommend keeping categories broad — one "Microsoft 365" folder beats separate folders for Word, Excel, and PowerPoint. It's easier to maintain and works as well.

### 4. Add your workspace knowledge

Now bring in the content: upload documents, add URLs, connect Notion, SharePoint, or Google Drive, and turn on Slack channel learning for the channels where your team already answers questions. We'd recommend using a URL or connected source whenever the content changes, so Atom stays current without you re-uploading anything.

Full walkthrough: [Add and organize knowledge](/product/ai-knowledge/knowledge-management/add-and-organize-knowledge). For Slack, see [Learn from Slack channels](/product/ai-knowledge/learning-sources/learning-from-slack-channels).

### 5. Add verified answers for your top questions

For questions that come up constantly and have one correct answer — how to connect to the VPN, how to request a specific license — create **verified answers**. Atom surfaces these word-for-word, and they take priority over anything it finds in a document or connected source. Have a subject-matter expert review each one.

See [Verified answers](/product/ai-knowledge/knowledge-management/verified-answers).

### 6. Scope knowledge with audience segments

Not everyone should see everything. Use **segments** to control which employees get answers from which content, so a Finance-only policy never surfaces for someone in Sales. We'd recommend setting the segment as you add each document, rather than auditing permissions across a large knowledge base later.

A few things worth knowing:

- You can set a segment on uploaded documents, URLs, and Notion content.
- **SharePoint is the exception** — those documents keep the permissions set in SharePoint or Azure AD, so you can't override the segment inside Atomicwork.
- You can restrict employees from downloading a source document while still letting Atom answer from it.

For example, if an employee asks about a confidential policy in a public Slack channel, Atom answers them in a DM instead of posting it in the channel — the right people get the answer without exposing it to everyone.

See [Managing audiences with segments](/product/identity-access/people-roles-and-teams/managing-audiences-with-segments).

### 7. Validate before you roll out

Before employees ever see it, pressure-test Atom's answers. Use the Test Suite to evaluate Atom's responses against a golden dataset of expected answers, so you catch gaps and regressions before they reach anyone.

See [Evaluating Atom's responses](/guides/evaluating-atoms-responses).

## Keeping knowledge fresh

Knowledge management isn't a one-time exercise — the value comes from the loop. Once you're live:

- **Let Recommendations tell you what to fix.** [Recommendations](/product/ai-knowledge/atom/improve-deflection-with-recommendations) in the [Insights report](/product/analytics/executive-views/insights-report) (**Settings > Analytics > Insights**) analyzes real conversations and surfaces specific, prioritized actions — which knowledge articles to add or edit — with the projected deflection impact of each. It focuses on subthemes where the deflection rate is below 80% with enough volume to matter, so you spend effort where it moves the needle. Work the list, then feed the fixes back in as new documents or verified answers.
- **Let living sources stay living.** URLs and connected sources relearn on a cadence, so keep the source current and Atom follows. This is why a URL usually beats a one-time upload.
- **Revisit verified answers as policies change.** They're authoritative, which means a stale one is worse than none. Review them on a regular cadence.
- **Re-validate after big changes.** Any major content update is a reason to evaluate Atom's responses again.

Global behavior is largely set-and-forget; your workspace knowledge is where the ongoing work lives.

## Best practices

- **Start small, then let Recommendations lead.** You don't need to document everything before launch. Turn on Trusted Knowledge, add your core policies and most-asked verified answers, and go live — then use Recommendations to fill the gaps employees actually hit. You'll cover real demand faster than you would by guessing at it.
- **Keep one source of truth per topic.** When the same policy lives in two places — say an uploaded PDF and a connected Notion page — the two drift apart, and Atom can end up surfacing the outdated one. Give each topic a single home, and point any other references to it rather than copying the content across.
- **Make your content easy for Atom to read.** Atom learns from text, not from images or code. Spell out each acronym at least once, and keep the important details in the body of a document rather than locked inside a screenshot or diagram. For the full treatment, see [Best practices for creating AI-ready knowledge](/guides/best-practices-for-creating-ai-ready-knowledge).
- **Retire outdated content as you add new content.** A stale or contradictory source confuses Atom as much as a missing one. When a policy changes, update or remove the old version instead of layering a new one on top.