When a CSV shows garbled names or addresses in Excel, close it without saving and keep the original untouched. Re-import a working copy through Excel's text/CSV preview so encoding, delimiter, and column types can be diagnosed separately before any value is overwritten.
Protect the source before diagnosing the symptom
The safest first action is to preserve the exported file exactly as received. Make a working copy with a clear suffix such as customers-import-test.csv, record where it came from, and avoid pressing Save in the workbook that already displays corrupted text. Saving from that state can replace the only useful evidence with Excel's current interpretation.
Choose a few known rows that exercise different risks. A useful fixture includes a name with accents such as José Núñez, a CJK name, an address containing a comma, a note containing quotation marks or a line break, an emoji, and an identifier such as 001284. The fixture should be synthetic or authorized for testing; customer names and addresses should not be uploaded or copied into unrelated services merely to diagnose an import.
Record the expected row count, column count, and exact values for those rows. This turns “it looks wrong” into separate questions: Are the characters wrong? Did a comma split one field into two columns? Did a leading zero disappear? Did a date change meaning? Each failure has a different correction.
Distinguish encoding damage from delimiter and type problems
Garbled sequences such as José or æŽä½³ usually mean bytes were decoded with a character set that does not match the file. Literal question marks already present in every viewer can mean the exporting system replaced characters before the CSV reached Excel. Selecting UTF-8 during import cannot reconstruct information that is no longer in the source bytes.
Shifted columns point elsewhere. An address such as 10 Market Street, Suite 4 needs CSV quoting to remain in one field. A note containing a line break also needs valid quoting. If the exporter omitted those rules or Excel detected the wrong delimiter, later values can move under the wrong headers even though every character is readable.
Identifiers and dates are a third category. Excel may interpret 001284 as a number and remove the leading zeros, display a long identifier in scientific notation, or interpret 03/04/2026 according to a regional date convention. Changing the file encoding will not fix any of those type decisions.
Do not rename .csv to .xlsx as a repair. Renaming changes the filename, not the bytes, delimiter, encoding, or workbook structure. A real XLSX file is a packaged workbook format, while CSV is delimited text.
Use a browser preview as a diagnostic comparison
Open a non-sensitive working copy in the CSV / Excel Online Preview. The local csv-preview component accepts CSV, TSV, XLSX, and XLS, reads the selected file in the browser after loading its spreadsheet parser, lets you select a sheet, and supports a text search across rows.
If the expected names and notes appear correctly in the browser preview but Excel's direct-open result is garbled, that is useful evidence that Excel's opening path deserves attention. It is not conclusive proof of the source encoding, because the two parsers can infer text differently. If both views show literal question marks or the same broken characters, return to the export system or inspect the bytes with an encoding-aware editor.
The preview tool has important limits. It has no encoding selector, renders at most 500 matching data rows, and does not edit the original file. Its Export CSV action creates a new CSV from the currently parsed sheet; it should not be treated as an encoding-repair command. A parser that displays the sample correctly also cannot prove that every row, business rule, or relationship in a large file is intact.
Use the preview for narrow comparisons: search the known fixture rows, count visible columns, and confirm whether a quoted comma remains in one cell. Keep the authoritative diagnosis tied to the source system, file specification, and Excel import preview.
Import UTF-8 through Excel's explicit preview
Microsoft says a UTF-8 CSV saved with a byte order mark can normally be opened directly in Excel. When the file has no BOM, its current guidance recommends Power Query through Data > Get Data > From File > From Text/CSV, or the Get Data From Text route that opens the Text Import Wizard.
For the supported desktop instructions, start a new workbook and choose Data > Get & Transform Data > From Text/CSV. Select the working copy, then inspect the preview before choosing Load. Confirm that the file origin or encoding is UTF-8, select the actual delimiter, and check the known names, addresses, notes, and header boundaries.
Choose Transform Data when identifiers, dates, or decimal values need explicit types. Set account codes, postal codes, SKUs, and other non-arithmetic identifiers to text before loading. Compare dates with an unambiguous source value, and verify decimal separators and currency meaning rather than relying on how one sample happens to display.
Microsoft's current text/CSV page applies to Excel for Microsoft 365, Excel 2024, Excel 2021, Excel 2019, and Excel 2016. Ribbon groups can differ by update channel. Excel for Mac and Excel for the web can also expose different Power Query commands, so use the product-specific Microsoft instructions when the exact desktop path is absent.
Directly opening a CSV is less controlled because Excel uses its current default data-format settings to interpret columns. The explicit import path provides a preview and a decision point; it does not modify or repair the original CSV by itself.
Verify characters, boundaries, and types after loading
Load the controlled result into a new worksheet, then save it as a new .xlsx workbook. Do not overwrite the CSV. Close and reopen the new workbook so the final check covers both import and save behavior.
Reconcile more than the first visible row:
- Compare the total row and column counts with the recorded source expectations.
- Check the first, a middle, and the last known row.
- Search the accented, CJK, and emoji fixture values exactly.
- Confirm that commas, quotation marks, and embedded line breaks remain inside their intended fields.
- Verify that leading-zero identifiers remain text and retain their original length.
- Compare several dates and decimal values against unambiguous source records.
- Confirm that no address or note has shifted into a neighboring customer's row.
If a value fails, stop before importing the full file. Revisit the relevant layer: request a correct UTF-8 export when source characters are already lost, fix quoting or delimiter settings when columns shift, or adjust the Power Query type step when identifiers and dates change. Manual cell-by-cell repair makes the result hard to reproduce and can hide additional failures.
For a small authorized text sample, the CSV to JSON converter can make header-to-value mapping easier to inspect. It is not an encoding detector, does not open the source file, and cannot decide whether a field belongs to the correct person. Use it only after the text and delimiter are already understood.
Common questions
Why does From Text/CSV work when double-clicking the CSV does not?
Direct opening uses Excel's current default import settings. The explicit path provides a preview where encoding, delimiter, and transformation choices can be reviewed before data is loaded.
Does selecting UTF-8 fix every CSV problem?
No. Encoding controls how bytes become characters; it does not repair invalid quoting, choose the correct delimiter, preserve leading zeros, or restore characters that the exporter already replaced.
Should I add a UTF-8 BOM to every file?
Microsoft notes that Excel can normally open UTF-8 CSV files with a BOM, but altering a delivered file should follow the receiving system's specification. Preserve the original and test a copy rather than silently changing an exchange format.
Can the online preview repair a garbled CSV?
No. It can provide a comparison view and export the parsed sheet as a new CSV, but it has no encoding control and does not edit the source. Treat it as diagnostic evidence, not a repair or completeness guarantee.
Is the import safe after five sample rows look correct?
Five rows reduce uncertainty but do not prove the entire file. Reconcile counts, include difficult characters and field shapes, test in a controlled destination, and follow the data owner's approval process before a full import.