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

# Review AI-extracted documents

> How Lift AI reads an uploaded PDF for you — splitting, classifying, and extracting each document — and how to review, approve, or correct what it finds.

Instead of tagging a PDF page by page, you can hand it to Lift AI. It splits a multi-document file into its parts, works out what each one is, and pulls the key fields — type, dates, amounts, vendor — off each document. The results wait for you under **AI-extracted Docs**, where you review each one and approve it to create the record. Use this article to upload for AI processing and work through what comes back.

<Note>
  AI processing is limited to the **consultant** role. The **AI-extracted Docs** entry and the **AI pipeline** upload option don't appear for other roles.
</Note>

## Upload PDFs for AI processing

You can start from two places:

* **AI-extracted Docs** in the sidebar — open the page and drag PDFs onto it, or select **Upload**.
* **Upload Document** in the top bar — in the **Add to Property** dialog, choose the **AI pipeline** tab (marked *Recommended*).

<Steps>
  <Step title="Choose a portfolio">
    Pick the **portfolio** the documents belong to. It scopes how Lift AI matches each document to a property and lift.
  </Step>

  <Step title="Add one or more PDFs">
    Drop in a whole batch — each file is processed on its own (up to 75 MB per file, PDF only). For a file you know is a single document that shouldn't be split, tick **Keep as one document**; Lift AI still detects its type.
  </Step>

  <Step title="Start extraction">
    Select **Start extraction**. Each file becomes its own upload and begins processing.
  </Step>
</Steps>

<Tip>
  Under **Advanced options** you can turn on **Auto-file confident invoices & tickets** — see [Auto-file confident documents](#auto-file) below.
</Tip>

## Track processing on AI-extracted Docs

Your uploads appear in the **My Uploads** table with their filename, status, page and document counts, how many documents are still pending review, and when they were uploaded. The status updates on its own as processing moves along:

| Status               | What it means                                                                                       |
| -------------------- | --------------------------------------------------------------------------------------------------- |
| **Pending**          | Queued, not started yet.                                                                            |
| **Splitting Pages**  | Finding where each document starts and ends.                                                        |
| **Classifying**      | Working out each document's type.                                                                   |
| **Extracting Data**  | Reading the fields off each document.                                                               |
| **Review needed**    | The proposed split needs your confirmation — see [Confirm where documents split](#boundary-review). |
| **Ready for Review** | Extraction is done; documents are waiting for you to review.                                        |
| **Reviewed**         | Every document in the upload has been handled.                                                      |
| **Failed**           | Processing couldn't complete.                                                                       |

<Note>
  If an upload stalls or fails, select the **Reprocess** control on its row to restart splitting, classification, and extraction from the beginning.
</Note>

<h2 id="boundary-review">
  Confirm where documents split
</h2>

When a batch PDF holds several documents, Lift AI proposes where each one begins and ends and asks you to confirm before it extracts anything. An upload waiting on this shows **Review needed**; open it to reach the **Boundary Review** screen, adjust any boundaries that look wrong, and approve. Processing then resumes and moves on to classifying and extracting each document. A file you marked **Keep as one document** skips this step.

<h2 id="review-and-approve">
  Review and approve each document
</h2>

Open a **Ready for Review** upload to see its documents. For each one, Lift AI shows the source pages next to the fields it extracted and any records it suggests matching to. You can:

* **Approve** — create the record (an invoice, a ticket, and so on) from the extracted data. Fix any field first if it needs it.
* **Reject** — discard a document that shouldn't become a record.
* **Reclassify** — change the document's type if Lift AI read it wrong.
* **Flag for review** — set it aside for manual handling, for example when it was split incorrectly.
* **Retry extraction** — have Lift AI read the document again.

Approved records land where that type lives — invoices and proposals in [Spending](/spending/review-spending), tickets in the [Tickets table](/tickets/browse-tickets), certificates and violations in [Compliance](/compliance/compliance-dashboard).

<h2 id="auto-file">
  Auto-file confident documents
</h2>

Turn on **Auto-file confident invoices & tickets** (under **Advanced options** at upload) to skip the last approval step for the clear-cut ones: high-confidence invoices and tickets are created for you automatically once extraction finishes. Everything else — lower-confidence documents, and any other type — still waits for your review. Auto-filing needs a portfolio selected.

## Find these in your Work Queue

Items that need you appear in your [Work Queue](/work-queue/triage-your-work-queue) alongside your other work: a **Split review** item when an upload is waiting on boundary confirmation, and a **Document review** item for each extracted document ready to approve. Selecting one opens its review in place, so you can clear it without leaving the queue.

## Related

* [Upload a document](/documents/upload-a-document)
* [Add a document manually](/documents/add-a-document-manually)
* [Review an invoice](/invoices/review-an-invoice)
* [Triage your Work Queue](/work-queue/triage-your-work-queue)