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

# Keep knowledge accurate

> How automatic training works, what the status badges mean, and how quota is measured.

Whenever you add, edit, or remove a source on the **Content** tab — [links](/help/ai-employees/content/add-your-website), [documents](/help/ai-employees/content/upload-documents), [text](/help/ai-employees/content/add-plain-text-knowledge), or [Q\&A](/help/ai-employees/content/build-qa-knowledge-base) — the employee automatically retrains. There is no manual **Retrain** button or trigger.

<Tip>
  Training runs every **5–15 minutes**, all day. After a change, wait a few minutes for it to be picked up in the next cycle rather than looking for a way to force it.
</Tip>

## Training status badges

Every content source shows a badge so you always know where it stands.

| Badge                 | Meaning                                                                   |
| --------------------- | ------------------------------------------------------------------------- |
| **Untrained**         | Brand new — has not been picked up by a training cycle yet.               |
| **Training required** | You edited or added something and it needs to be retrained.               |
| **Trained**           | Ready — the employee can use this source in conversations.                |
| **Failed**            | Something went wrong. Check the source and try again, or contact support. |

<Frame>
  <img src="https://mintcdn.com/heyy-c8bd5c9e/1coCVJUJuVs6SA9x/assets/ai-employees-content-training-status.png?fit=max&auto=format&n=1coCVJUJuVs6SA9x&q=85&s=4e538c91b6a46c042160f461fbcfbb75" alt="Content training status badges" width="1347" height="627" data-path="assets/ai-employees-content-training-status.png" />
</Frame>

If a source shows anything other than **Trained**, the employee cannot answer from it yet — a question that depends on it comes up empty, not because the content is wrong, only because processing has not finished. If a source sits on **Failed** for a while, re-check the file or link and re-add it rather than leaving it as-is; the employee is treating that content as unavailable the entire time.

## What gets extracted

| Source        | What training extracts                                                                                                                                        |
| ------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Documents** | Smart-scanned: tables get reconstructed, multi-column layouts get read in proper order — see [Upload documents](/help/ai-employees/content/upload-documents). |
| **Links**     | Crawled for text only; images and scripts are stripped out.                                                                                                   |

## Content limits

The Content tab shows how much of your knowledge base quota you have used, as a running total against your plan's limit — see [plan limits](/help/ai-employees/getting-started/hire-an-ai-employee).

<Frame>
  <img src="https://mintcdn.com/heyy-c8bd5c9e/1coCVJUJuVs6SA9x/assets/ai-employees-content-limits.png?fit=max&auto=format&n=1coCVJUJuVs6SA9x&q=85&s=66f91d3e107958c162756bceda86566b" alt="Content quota usage" width="1366" height="627" data-path="assets/ai-employees-content-limits.png" />
</Frame>

<Tip>
  The quota counts extracted text, not original file or page size. A 20 MB PDF with 1 MB of actual text and tables only counts as 1 MB. The same applies to crawled pages. In practice, even a smaller plan's quota goes further than the megabyte number suggests.
</Tip>

## Reviewing content over time

Review the Content tab whenever offers, policies, or products change:

* **Update or remove** outdated links, documents, text, and Q\&A entries.
* **Rename** sources so your team knows what each one covers at a glance.
* **Delete** sources that are no longer valid — stale content is what causes wrong answers, not a gap in coverage.

After a batch of edits, [test in the Playground](/help/ai-employees/going-live/test-in-playground) with the questions customers actually ask, not just a general check. If your content includes product data that changes on its own, see [why you shouldn't crawl store data](/help/ai-employees/content/add-your-website#dont-crawl-store-data-shopify-woocommerce) for the more reliable alternative.