# Review answers and sources (https://www.clavy.ai/help/review-answers-and-sources)

See exactly where each answer came from before you rely on it.

Every auto-filled cell keeps a record of how it was made. Check it before you rely on an answer, especially for decisions or client work.

## View a cell's details [#view-a-cells-details]

Hover over a filled cell and select  **View details**:

- **Web sources** link to the pages the agent used.
- **File sources** name the document and page. Select one to open the document at the cited passage, highlighted.
- **Reasoning** shows the agent's explanation, if reasoning is on.
- The model that filled the cell.

## Read a full row or cell [#read-a-full-row-or-cell]

Long answers are cut off in the grid. Select  **Expand** on a cell to read it in full, or  **Expand row** to see every column for one row at once.

## Fix an answer [#fix-an-answer]

If an answer is wrong, you can:

- **Rerun** the cell. Hover over it and select  **Re-run agent**.
- **Edit** the cell directly. Select it and start typing.
- **Fix the cause.** If several answers are off in the same way, the prompt usually needs to be more specific. See [Improve results](/help/improve-results).

## How reliable answers are [#how-reliable-answers-are]

> AI answers can be wrong, even with sources. Spot-check a few rows before relying on a whole column.

Answers are kept grounded: every answer keeps its sources, citations link to the exact passage in a document, select columns can only answer with the options you define, and each cell is answered on its own from that row's data. Still, keep in mind:

- **Long documents are read in parts.** The agent reads the passages most relevant to the prompt, so a broad prompt can miss information elsewhere in the document. See [Write column prompts](/help/write-column-prompts#prompts-for-documents).
- **The web changes.** Web answers reflect what the agent found when the cell ran. Rerun cells to refresh them.
- **Answers can vary between runs.** Asking the same thing twice can give differently worded, or occasionally different, answers. Specific prompts and select columns make results more consistent.
- **Scans are harder to read.** Text in scans, handwriting, and dense tables is read automatically, but less reliably than clean digital text.