To convert PDF tables to Excel with AI, extract one table at a time, keep the original text beside the converted values, and reconcile the rows and amounts with the PDF. A spreadsheet that opens successfully can still contain a lost minus sign or a total counted twice.
This guide uses an original fictional purchase statement and a deliberately flawed extraction to show what to check. It is an editorial exercise, not a live test or accuracy comparison of AI products. The source below represents a PDF table; no downloadable PDF or workbook is required for the exercise.
1. Choose an Extraction Route for Your PDF
Try selecting a sentence and a number in the PDF. If the page is an image, use a tool that can read scanned pages or an OCR step first, then inspect the transcription. Selectable text helps, but it does not prove that rows, merged cells, and notes will transfer correctly.
For a readable table, your Excel installation may offer Data → Get Data → From File → From PDF. Choose the table in Navigator and review it with Transform Data before loading it. Availability depends on the Excel version and platform; check Microsoft’s connector availability list if the command is missing. The PDF connector documentation also notes problems that can occur with multi-line rows.
For the AI route, use an assistant that supports the file or page images you provide. Ask it to identify the table and its source page before extracting rows. If it cannot read the attachment, supply a clearer crop or readable source text. Share only documents you are permitted to upload, using an approved tool for confidential material.
2. Read the Source Before Converting It
Fictional source: Purchase statement, page 2. Amounts are USD, a period is the decimal separator, and a comma groups thousands. Parentheses mean a credit. Tax is excluded. Codes are identifiers, not quantities.
| Code | Description | Amount shown |
|---|---|---|
| 0017 | Equipment | 1,250.50 |
| 0018 | Delivery | 240.00 |
| 0019 | Return credit | (35.50) |
| 0020 | Setup fee waived | 0.00 |
| — | Statement total | 1,455.00 |
There are four detail rows and one summary row. Preserve that distinction. The zero setup fee is a supplied value, not a missing amount. The credit is negative because the source note says so; do not apply that interpretation to every document without checking its notation.
3. Ask AI for Source Text and Converted Values
Start with a single page or table. Keep a raw extraction separate from a working copy. This prompt requests transcription and explicit conversions, rather than permission to invent missing values:
Extract only the Purchase statement table on page 2. Return four detail rows with these columns in order: Code, Source_page, Description, Amount_as_printed, Amount_USD, Review_note. Preserve codes as text, including leading zeros. Keep the printed amount separately from the numeric value. Under this source’s rules, parentheses indicate a credit, commas group thousands, and the decimal separator is a period. Keep zero as zero. Do not guess unreadable or missing values; leave the numeric field blank and explain the issue. Put the statement total and source notes in a separate summary, not among the detail rows. Report the number of detail rows and any uncertain cells. Do not add formulas to extracted source cells.
For a scanned page, compare ambiguous characters with the image yourself. If a digit could be 3 or 8, a total that happens to balance is not permission to choose one. Request a better source or keep that cell unresolved.
4. Bring the Reviewed Result Into Excel
If the assistant can generate a workbook, request a source sheet, a detail sheet, and a checks sheet. Open the actual file and inspect it. If it returns only a table, copy the reviewed rows into a new sheet with the six headers above. Format the Code and Amount_as_printed columns as Text before pasting; keep Amount_USD numeric.
For a text or CSV file, use Excel’s import preview rather than assuming a double-click will preserve the fields. Set identifier columns to Text and check the delimiter and decimal convention before loading. Microsoft’s leading-zero guidance explains text handling during import. Applying a display format after 0017 has become 17 does not establish what the source originally contained.
For this six-column layout, Code is column A, Source_page B, Description C, Amount_as_printed D, Amount_USD E, and Review_note F. Put the four detail records in rows 2–5. In E2:E5 the checked numeric values are 1250.50, 240.00, -35.50, and 0.00. Keep the printed total, 1455.00, on the checks sheet.
5. Check Rows, Signs, and Totals Separately
A flawed extraction might turn (35.50) into +35.50 and append the statement total as another transaction. These two errors produce different clues:
- Correct detail sum: 1250.50 + 240.00 − 35.50 + 0.00 = 1455.00.
- Credit sign lost: the detail sum becomes 1526.00, which is 71.00 too high. Changing −35.50 to +35.50 changes the sum by twice 35.50.
- Printed total also counted: 1526.00 + 1455.00 = 2981.00. Five numeric records now appear where there should be four detail rows.
On a separate checks sheet, use the following formulas if the four records are on a sheet named Detail. These formulas are specific to this sample layout; adapt the ranges for your own table.
| Check | Formula | Expected |
|---|---|---|
| Nonblank codes | =COUNTA(Detail!A2:A5) | 4 |
| Numeric amount cells | =COUNT(Detail!E2:E5) | 4 |
| Detail sum | =SUM(Detail!E2:E5) | 1455.00 |
| Difference from printed total | =ROUND(SUM(Detail!E2:E5)-1455,2) | 0.00 |
Inspect the entire detail sheet for extra rows beyond row 5; a fixed formula range cannot detect an appended total outside that range. If COUNT returns 3, investigate the missing numeric value even if the sum looks right. Losing this example’s zero row would leave the total unchanged. Two offsetting transcription errors could also produce a zero difference.
Check all four codes, descriptions, printed amounts, numeric amounts, and page references against the source. Confirm that the tax exclusion remains visible. For larger documents, reconcile each table or page group and inspect uncertain cells before combining results. Do not use a matching grand total as your only accuracy test.
6. Save a Result Someone Else Can Check
Keep the original PDF, its filename and version, the raw extraction, the reviewed detail rows, and the separate checks together. Record corrections such as “restored credit sign from page 2” and “kept statement total outside detail rows.” Label unresolved cells so another reader can see what still needs review.
Once extraction is verified, use our spreadsheet cleanup workflow for genuine duplicate or missing-data decisions, and our Excel formula guide for further calculations. To turn the checked figures into prose while preserving notes and conditions, see summarizing documents with AI. Browse the AI data-analysis tool guide when choosing an assistant.
If you need to bring several checked exports together, follow our guide to combining Excel files with AI. Keep a source reference for each row and reconcile file counts before reviewing overlaps.
Written September 14, 2026. Microsoft documentation checked on that date. Examples are fictional and independently calculated for this guide.