Sources
The sources array carries the reference material Grok cites for an answer, each entry expanded with preview snippets, author and site metadata, and thumbnail imagery.
Overview
Sources are delivered inline as part of a Grok extraction response, so you never issue a separate request to obtain them. Every element starts from the shared citation shape used across cloro engines — position, url, label, and description — and Grok layers on a set of engine-specific fields that describe the underlying page in more detail. Send a prompt to the Grok endpoint and the returned payload includes a populated sources list.
{
"prompt": "latest developments in quantum computing",
"country": "US"
}
Grok-specific source fields
| Field | Type | Description |
|---|---|---|
| preview | string | Short excerpt pulled from the body of the cited page. |
| searchEngineText | string | Display line as the source appears in search results. |
| siteName | string | Human-readable name of the publishing website. |
| metadataTitle | string | Title declared in the source page's own metadata. |
| creator | string | Author or content creator credited for the material. |
| image | string | URL of a preview or cover image associated with the source. |
| favicon | string | URL of the website's favicon. |
Response example
{
"success": true,
"result": {
"text": "Quantum computing saw notable milestones through 2026...",
"sources": [
{
"position": 1,
"url": "https://example.com/quantum-2026",
"label": "Quantum Computing in 2026",
"description": "A survey of recent progress in quantum hardware and algorithm design.",
"preview": "A newly demonstrated 1,000-qubit processor pushed error-corrected workloads further than...",
"searchEngineText": "Quantum Computing in 2026 - Recent Breakthroughs",
"siteName": "Example Tech",
"metadataTitle": "Quantum Computing in 2026 | Example Tech",
"creator": "Jane Doe",
"image": "https://example.com/quantum-cover.jpg",
"favicon": "https://example.com/favicon.ico"
}
]
}
}
Notes
Grok-specific fields are best-effort: any of them may be absent when the underlying page does not expose that metadata, so treat each as optional when parsing. The base citation fields (position, url, label, description) remain the reliable anchor for every entry. Questions about the response shape can go to [email protected].