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

# Oracle

> Scan AI responses for inaccurate claims, review the evidence, and keep your brand facts current

Oracle checks what AI models say about your brand against your verified Knowledge Base facts. It groups confirmed inaccuracies, possible Knowledge Base issues, and disagreements between models so you can decide what needs attention.

Open **Oracle** in the sidebar. Your website needs Oracle access, at least **25 verified facts** in its Knowledge Base, and a completed prompt run before you can scan. Starting a scan requires website admin access.

## Run an Oracle scan

1. Check that your facts are current in [Knowledge Base](/knowledge-base).
2. Open [Oracle](https://app.athenahq.ai/oracle).
3. Click **Run Oracle**. Athena scans responses from your latest completed prompt run against your current facts.
4. Follow the progress on the page. You can leave and return while the scan runs, or click **Cancel** to stop it.
5. Review the completed scan. Use **Run history** in the header to open an earlier run.

You do not select a date range in the current scan flow. If you have corrected your Knowledge Base, you can run Oracle again against the latest completed prompt run. A website can have only one Oracle scan running at a time.

## Read the results

The dashboard includes:

* **Inaccuracy**: the share of scanned responses with a confirmed finding, alongside the number of responses flagged.
* **Responses scanned**: the number of responses included in this run. Read any scope or truncation message alongside the total.
* **Priority findings**: major pending findings to review first. Open a finding or use **See all** to review the full list.
* **Knowledge Base issues**: facts that may need correction or clarification.
* **Model disagreements**: conflicting statements from different models.
* **Model accuracy**, **Topics AI gets wrong**, and **Prompt topics**: breakdowns that help you locate recurring problems.

Earlier history can be labeled **Imported**. Those runs preserve previous findings but do not have the same scan statistics as current runs.

## Review an inaccurate claim

1. Open a finding from the dashboard or the full list.
2. Compare the **AI claim** with the **Known fact** and read **Why it was flagged**.
3. Click the AI claim to inspect the full response. Click the known fact to inspect its Knowledge Base record and use **Edit fact** if it needs correction.
4. Review any **Possible Sources** associated with the claim. These are evidence to investigate, not proof that an external page caused the model's answer.
5. Choose **Acknowledge** when you have reviewed the finding, or **Ignore** when it is not actionable. Use **Reopen** to return it to Pending.

Acknowledging or ignoring a finding changes its review status. It does not edit the underlying fact, publish content, or change what an AI model says. Edit incorrect facts explicitly in Knowledge Base, then run another scan when you want a fresh comparison.

## Verify facts against external sources

Fact verification lives in **Knowledge Base**, separately from Oracle's response scans:

1. Open **Knowledge Base > Verify facts**.
2. Select the pillars to check and start the analysis.
3. Open **View results** when it completes to inspect external contradictions and internal inconsistencies.

See [verifying Knowledge Base facts](/knowledge-base#verifying-facts-against-the-outside-world) for the complete flow.

## Frequently asked questions

### Why can't I start a scan?

Check Oracle access, your website admin role, and whether a scan is already running. You also need at least 25 verified facts and a completed prompt run. If a run fails because its Knowledge Base or prompt scope exceeds a supported limit, follow the explanation shown on the page.

### Does acknowledging a finding fix the AI response?

No. It records that you reviewed the finding. Use the evidence to correct your facts or content and monitor future responses for changes.

### Where did the old Check responses and Verify facts menu go?

Use **Run Oracle** to scan the latest completed prompt run. **Verify facts** and its results are now in Knowledge Base. Older fact-verification links redirect there.

### Where are my earlier remediation drafts?

Existing drafts are available in **Content > Content Hub** under in-progress content. Review and edit them there. Finding review in the current Oracle page does not automatically create or publish a draft.

For the detailed page reference, see [Oracle controls and states](/guides/oracle).