Bank CSV opens in one column or shows garbled text: fix the import

A bank CSV that opens in one column needs a delimiter check; readable columns with garbled accents need an encoding check. Diagnose those separately on an unchanged source file. A third problem—broken quoted descriptions—can shift fields even when most rows look right. This guide gets the file into the right text columns before you interpret dates or money.

Already have a transaction CSV? You do not need a PDF converter to fix its import. LocalExtract is an option when your source is a PDF; the steps below work on an existing CSV.

Find which reading stage failed

Open a copy in a local plain-text editor first. Do not save over the bank download. Record the export’s stated encoding and separator if its instructions provide them; a filename ending in CSV does not establish either setting.

Start again from the source if an earlier open-and-save has already changed text. Changing a display setting cannot restore characters or leading zeros that were overwritten. If the raw source is already corrupted, obtain a new export or verify the affected entries against the statement instead of guessing names.

Use a four-column control instead of trial-and-error replacements

This synthetic example uses a semicolon delimiter, UTF-8 text and a decimal comma. It represents no real bank or customer. Its reference field is extra teaching data, not a field supplied by LocalExtract.

date;description;amount;reference
2026-10-01;"Café, Nord";-12,50;00017
2026-10-02;"Desk ""A""";25,00;00018

The expected result is three records including the header, with four fields in each record:

Choosing a tab for this example leaves one field per record. Choosing a comma misreads both the description punctuation and decimal amounts. Choosing the semicolon and honoring quoted fields produces four. The doubled quotation marks inside the second description represent one literal quotation mark on each side of A; they are not stray characters to delete.

We checked these expectations using Python’s CSV reader with explicit settings. Download the synthetic reading cases for the exact text, UTF-8 hex bytes and expected rows. They are diagnostic data, not an accounting import template or evidence of a successful Excel import.

Import a copy through Excel’s preview

For Microsoft 365 desktop Excel on Windows, Microsoft documents Data → From Text/CSV. Select the copy, set the delimiter in the preview, and choose Transform Data when you need to control column types before loading. Directly opening a CSV uses default format settings and may reinterpret its values; see Microsoft’s text/CSV import instructions. Menu wording and available controls differ across versions; this is not a Mac or Excel for the web menu recipe.

For the control above, select a semicolon. Inspect both transaction rows, not only the header: the description must stay in a single field, amount must remain separate, and reference must contain five characters. If Power Query added a type-conversion step automatically, remove that step while inspecting raw values and set reference to Text before loading. Converting an already shortened numeric 17 to text will not recover 00017.

Do not change the Windows system list separator just to repair one import. A query-specific setting keeps this diagnosis local to the file; changing a system setting can affect other applications. Do not globally replace commas with semicolons: the commas in this example have two different meanings, neither of which is the field delimiter.

Fix decoding without changing the transaction text

The word Café has UTF-8 bytes 43 61 66 C3 A9. Reading those bytes as Windows-1252 produces Café. That is a controlled decoding example, not a rule that every garbled file is UTF-8. Choose UTF-8 only when the source specification or a trustworthy comparison supports it; test more than one accented description.

Microsoft explains that Excel can normally open a UTF-8 CSV directly when it has a byte order mark (BOM). Without that mark, use Power Query or the text-import route described in Microsoft’s UTF-8 guidance. The UTF-8 BOM is EF BB BF at the beginning of a file. Its presence says nothing about the delimiter or the correctness of amounts.

Use the import’s file-origin or encoding control where available, returning to the unchanged bytes for each attempt. Re-saving garbled text as UTF-8 preserves the wrong text; it does not reverse the earlier decoding. A replacement character such as � is a reason to stop and compare with the source, not proof that choosing a different font will repair the file. Do not add a BOM to every accounting upload: follow the actual recipient’s accepted encoding.

Hand off only after the text checks pass

Keep a small before-and-after record: source filename, encoding, delimiter, expected field count and one description containing punctuation or accents. Check the first, last and exceptional records, plus quoted multiline descriptions if your source contains them; a physical line count alone does not count CSV records reliably.

Proceed only when descriptions retain their characters, each record has the expected fields, and identifiers remain exact text. Then use the separate date and decimal diagnosis. For target-software mapping and completeness checks, return to the accounting CSV guide. Correctly reading a file does not prove that its transactions are complete or ready for upload.

If your only source is PDF, LocalExtract processes one PDF at a time locally, including OCR, and exports date,description,amount CSV. It has no desktop batch queue. Account services and periodic Pro verification need a network. Export CSV, then save an XLSX workbook in Excel if required. Download LocalExtract for a PDF-based workflow.

Official Microsoft sources checked October 15, 2026. The worked data is synthetic; the reproduced results concern byte decoding and CSV parsing, not a bank-specific format guarantee.