# Review & Verification: Check Extracted Values Against the Source Image

> Open a file's extracted table in the Review dialog to check every value against the original document, locate each value on the image in both directions, edit cells in place, and restore the AI's original value for a field you changed by mistake.

The Review dialog opens a file's extracted table next to the original image, so you can check the result against the source document without re-reading it. Inside it you can locate where each value came from on the image, correct a value in place, and — if you change one by mistake — restore the AI's original value.

## Opening a file for review

Any completed file with an extracted table can be reviewed. The Review dialog shows the extracted table above and the original image below, and both are connected: the table drives the image, and the image drives the table.

1. In the **Files** panel, select a processed file. Its result appears in the results panel.
2. Click **Review** in the results header. Hovering the button shows *Review Extracted Data*. Clicking the image preview opens the same dialog.
3. The Review dialog opens full screen. The **Extracted Data** table is at the top and the original image is below it.

Cells can be edited directly in the dialog: double-click a cell to edit it, then press Enter to save (Shift+Enter inserts a line break) or Esc to cancel. Changes are saved as you work — there is no save button.

Each file is tagged **Reviewed** or **Not Reviewed** at the top of the dialog. Click **Confirm** to mark it reviewed, which also changes its tag in the Files list from **Completed** to **Reviewed**. Use **Prev Unreviewed** and **Next Unreviewed** to move through the batch, or **Prev Page** and **Next Page** to step one file at a time. Editing a file that was already reviewed resets it to **Not Reviewed** and warns you to confirm it again.

Only table results can be reviewed. A file processed in To Word mode returns a document rather than columns, so cell-by-cell review and the locating controls are not available for it; the dialog says so and still shows the image.

## Locating a value on the image

To confirm that a value was read from the right place, have the app locate every extracted value on the original image.

1. In the Review dialog, click **Annotate on Image** in the **Extracted Data** header. Its tooltip reads: *“Locate each extracted value on the image, so hovering a cell highlights where it came from. Uses 1 credit.”*
2. While it runs, the header shows **Locating fields on image…**. When it finishes it shows **Annotated**, or **Annotated (partial)** when some values could not be found on the page.

Once values are located, the table and the image point at each other in both directions:

- **From the table:** hover any cell and a green dashed box marks where that value came from on the image.
- **Pin a value:** click a cell, or its target icon, to pin that box so it stays on the image while you edit; the cell scrolls into view. Click the same cell again to unpin.
- **From the image:** every located value is also a clickable area on the image. Click a region on the image and the matching cell is selected and pinned — so you can start from either side.
- **Zoom and pan:** scroll to zoom, drag to pan, and use **Reset** to return the image to its original position.
- **When a value cannot be located:** it is marked *Not visually located on the image*. A value that was calculated or inferred rather than printed on the page has no position to highlight, so it stays unlocated.

Locating values is a separate, chargeable action from processing the file. The tooltip on the button shows *Uses 1 credit*, which is the base cost; what is actually charged follows the [Model Tier](/docs/reference/glossary#model-tier) the file was processed with — 1 [credit](/docs/reference/glossary#credits) on Standard, 3 on Advanced, and 6 on Premium. [Limits & Specs](/docs/reference/limits-and-specs) publishes the same figure as *“Base 1 credit, scaled by the processing tier”*, and [Model Tier](/docs/concepts/model-tier) explains how the tiers differ.

## Locating values automatically after processing

Instead of clicking the button for each file, you can have the app locate values for you. With the **Auto-annotate** setting switched on, every table-mode file is located right after it finishes processing, so the highlights are already in place when you open a result.

There are two places to switch it on, and both control the same account-wide setting:

| Where | What you see | How it takes effect |
| --- | --- | --- |
| Review dialog, Extracted Data header | A checkbox labelled Auto-annotate , with the tooltip “Automatically run this for every extract-mode file, right after it finishes processing.” | Applies as soon as you tick it — it saves immediately, with no Save button. |
| Your profile page, under Account Settings | A checkbox labelled Auto-annotate on image after processing , with the tooltip “Automatically run bbox review-highlighting (1 credit) right after each extract-mode file finishes processing.” | Tick it and click Save Settings . |

The difference from the manual button is scope and timing. **Annotate on Image** locates only the file currently open, at the moment you click it. **Auto-annotate** locates every table-mode file as it finishes, whether or not you have opened that file yet — useful when you want to verify a whole batch. Both are charged the same way, per image at the file's Model Tier, and both apply to table-mode results only; To Word files are never located.

The setting belongs to the account, so it is not limited to the file you happen to have open. You can switch it off from either place at any time.

## Restoring the AI's original value

If you edit a value and then want the AI's original reading back, the Review dialog keeps it for you.

- An edited cell is tinted, and a small amber marker appears beside it.
- Hover the marker to see a tooltip showing *AI original: <value>* and a **Click to revert** link.
- Click **Click to revert** and the cell returns to the AI's original value. The amber marker clears and the app confirms *Reverted "<Column>" to AI original*.
- Typing a cell back to exactly the original value by hand also clears the marker, so the file no longer counts that cell as edited.

Edits and reverts are saved automatically, so what you see in the Review dialog is what is stored for the file.

## Common questions

### Which files can I review and locate values on?

Files processed in To Table mode, because they produce a table of columns and rows. A file processed in To Word mode returns a document instead, so there are no cells to review or values to locate on the image.

### If I locate values and then locate them again, am I charged twice?

Each locating run is charged per image at that file's Model Tier, whether you start it with **Annotate on Image** or it runs automatically. If the values are already located, they stay located — you do not need to run it again to keep using the highlights.

### Do corrections I make in the Review dialog appear in the export?

Yes. Your edits are saved to the file's result, and an export builds its file from the saved results, so corrections are included. See [Export](/docs/guides/export).

### Why is a value marked as not located?

The app could not place that value on the page, most often because the value is not printed there — a calculated or inferred value has no source region to highlight. The markers tell you which values can be checked against the image directly and which cannot.

### Do I have to review files one at a time?

You move through them one at a time, but **Next Unreviewed** skips you straight to the next file that still needs a check, and switching on **Auto-annotate** means the highlights are ready by the time you open each one. See [Batches](/docs/concepts/batches) for how files are grouped.

---

Source: https://imagetotable.ai/docs/guides/review-and-verification