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

# Fix missing or inaccurate knowledge answers

> Check source readiness, assignment, and representative answers before changing model settings.

Choose one question the assistant answered incorrectly and identify the source that should answer it. This is more useful than changing several model settings at once.

## Check what is available to the assistant

1. Open **Knowledge** in the intended account and open the knowledge base.
2. Find the source with the expected answer. Check that processing completed; an upload acknowledgement does not make the source ready.
3. Inspect the processed content. A URL or file can import without containing the exact policy, table, or section you expected.
4. Open the assistant's **Knowledge Base** panel. Confirm the knowledge base is **Assigned**, or choose **Add to assistant**.
5. For voice, check the knowledge base's voice setting and account access, then **Publish** the assistant before a live call test.

Retrieval selects relevant passages from assigned, completed sources. It does not load every source into every answer. Repeating a missing fact in the prompt may hide an ingestion problem without fixing the knowledge base.

## Source status reference

| Status | Meaning and next action |
| - | - |
| `PENDING` | Source is waiting to process. It is not ready for retrieval. |
| `PROCESSING` | Extraction or processing is underway. Wait for a terminal result before testing its content. |
| `COMPLETE` | Processing finished. Inspect the extracted text and assistant assignment; completion alone does not prove the expected answer was extracted. |
| `ERROR` | Processing failed. Read the source error, correct the input, and retry. If the original input is unavailable, re-upload or re-enter it. |

A retry returns the source to `PENDING`. See [source types and upload options](/v3/guides/knowledge/add-sources) for Text, FAQ, File and URL inputs and their constraints.

## Improve the source

State policies directly, name the business or service clearly, and keep related conditions near the answer. Replace outdated material and resolve conflicting versions. For a short policy, a focused Text or FAQ source is often easier to inspect than a large mixed document.

If a page depends on a login, interactive controls, or content that was not extracted, provide the required text as an accessible source instead. Avoid uploading multiple conflicting copies of the same policy.

## Retest the answer

Ask the original question, a paraphrase, and a question whose answer is absent. Compare the answers with the source. For the missing answer, the assistant should acknowledge the gap or ask for clarification rather than invent a policy.

Repeat in the actual channel you use. A correct chat test does not by itself verify voice retrieval. Record the source, expected passage, question, answer, and call/conversation ID if the discrepancy persists.

Changing temperature does not repair missing or contradictory source content. Start with source readiness and assignment, then review the prompt's instruction to use the provided business information.

See [knowledge setup](/v3/guides/knowledge/add-sources), [retrieval](/v3/guides/knowledge/retrieval), and [billing](/v3/billing-guide).


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