How to check an Anki UTF-8 import when characters look wrong

Keep the earliest available version that still displays the intended characters correctly. That might be the source document, spreadsheet, or a previous export. Save it under a distinct name and avoid overwriting it while testing. Record the app used to create each candidate file and the export choice you made. A filename ending in CSV does not explain how its text was encoded, and changing the filename does not rewrite the underlying characters.

In the fictional example, the spreadsheet still shows the correct accent while the exported file does not. That narrows the first investigation to export and decoding choices. If the spreadsheet itself already contains damaged text, importing into Anki cannot restore the original from context reliably. Return to the source or a correct earlier copy. Treat the diagnosis as a chain: source, working table, exported text, import preview, stored field, and rendered card. The earliest wrong stage tells you more than the final screenshot alone.

Choose actual characters whose meaning matters in the file. A language sheet may include accented vowels, a non-Latin word, and punctuation used in examples. A science sheet may include a minus sign, degree symbol, or Greek letter. Include a plain Latin word as a control. The set should reflect your content rather than an impressive list of scripts that never occur in the collection. Record the intended source and meaning for each selected item.

Then place the test items in short, identifiable rows with ordinary text beside them. For example, distinguish a symbol used as a variable from the same shape used decoratively in a heading. Inspect whether a visually similar replacement would change the answer. A hyphen and a minus sign may look similar but serve different notation roles. Do not normalize every unusual character automatically to make the file appear cleaner. The purpose is faithful content transfer, not forcing the subject into a smaller alphabet that loses distinctions required by the course.

Compare the same stored text in more than one appropriate viewer while preserving the original bytes. If the correct character is present but one viewer shows an empty box, investigate display support. If the exported text contains a sequence of wrong characters where the source has one intended symbol, investigate the encoding path. These are diagnostic clues rather than universal rules, so keep the source comparison as the reference. A screenshot alone cannot establish which layer failed.

The fictional learner copies a suspect word from the stored field into a plain-text editor and compares it with the untouched source. They also inspect the import preview and actual card separately. If the field is correct but the card is not, a template or rendering issue deserves attention. If the preview is already wrong, fix the file or decoding choice before importing. Changing a display font will not repair text that has already been transformed into the wrong characters. Conversely, repeatedly converting a sound file can introduce damage when the original problem was only display support.

Use a harmless short label beside each test item so you can track it even when its subject characters are unreadable. For example, item A can contain the accented vocabulary term and item B the scientific sign. The label identifies the row without becoming part of the accepted answer. That small separation helps when comparing screenshots, exported text, and stored fields across several applications during the diagnosis.

The Anki text-import manual requires UTF-8. Use an explicit UTF-8 export option in the tool preparing the file and inspect a small result before continuing. Keep separator and quoting choices stable while testing character preservation; changing every file-format setting at once makes the cause harder to identify. The manual is the reference for Anki’s requirements, while the export tool’s own interface determines how you produce that file. Do not assume a generic save command selected the intended encoding.

Open the candidate in a viewer that lets you identify the selected encoding, then compare the test rows with the source. Import only that sample and inspect the actual fields. Record the result for each character rather than writing a vague note that the file looks better. If one item remains wrong, preserve it as an unresolved example. A partly successful conversion is not a reason to replace the entire collection, especially when the remaining failure changes the accepted answer. Use the smallest confirmed repair path and keep the good original available.

A repeated broken sequence may tempt you to replace it everywhere with a guessed character. That can be reasonable only after the mapping is established and the scope is verified. The same visible sequence might occur legitimately in another field, or different original characters may have collapsed into the same replacement. An irreversible replacement cannot recover distinctions that were already lost. Prefer regenerating from a correct source whenever possible rather than inventing content based on what seems likely.

For the fictional language sheet, the learner checks a few affected words against the class source before proposing a replacement. They then inspect exceptions and preserve a before-and-after sample. For a scientific formula, they verify the variable’s role and notation rather than guessing from visual similarity. This is content checking, not merely file cleanup. If the source is unavailable and meaning is uncertain, mark the answer for review instead of presenting a repaired-looking guess as authoritative study material. A visible uncertainty is safer for learning than a polished but unsupported symbol.

After the sample passes, repeat the same tested procedure on the full file and inspect representative items from several regions. Include the exact rows that failed earlier. Check field mapping and generated questions too; correcting characters does not prove that columns or answer criteria are right. Keep a short transfer record with source filename, export choice, sample observations, and the recovery copy. That record makes a later course update easier to diagnose without repeating the original investigation.

If you evaluate Elly with CSV, TSV, or TXT, repeat the character check at its receiving stage rather than assuming that a file tested in Anki has no further interpretation choices. Text import does not preserve Anki media, templates, scheduling, or review history. Keep the native collection and compare a permitted source-based task on a small sample. The success criterion is that the receiving question retains the intended wording and meaning. A file that uploads successfully but changes a meaningful accent or sign has not completed that task correctly.