For a reliable CSV-to-JSON conversion, preserve the values first and decide their types afterward. An identifier such as 00123 may look numeric, but converting it to a number changes it. A comma inside a customer note should remain part of the note, not become an extra column.

Check the input shape before converting

The companion CSV and JSON converter is for delimited text and a flat array of JSON records. It expects a header row with distinct, non-empty column names and consistent row widths. It does not read Excel workbooks, preserve spreadsheet formatting, or flatten arbitrarily nested JSON.

Choose the delimiter that the source actually uses. A semicolon-separated export is not repaired by pretending it uses commas. Keep a copy of the original file and inspect its first few rows in a text editor if the structure is unclear. Export the spreadsheet as text deliberately rather than renaming an .xlsx file to .csv.

A sample that catches common conversion mistakes

Input with an identifier, a comma, quotes, and an empty value
id,name,note
00123,נועה,"Window seat, please"
00456,Sam,"Said ""hello"""
00789,Lee,

Try the CSV example file and compare it with the expected JSON records. These files contain only synthetic data.

With string-preserving conversion, the first identifier should remain "00123". The first note is one field, the second contains ordinary quotation marks around “hello,” and the last note is an empty string. There are three records and three fields per record.

Expected first JSON record
{
  "id": "00123",
  "name": "נועה",
  "note": "Window seat, please"
}

RFC 4180 documents the familiar CSV convention: quote a field containing a delimiter, quote, or line break, and double a quote inside a quoted field. A line break inside quotes can belong to one record. Counting visible lines is therefore not always the same as counting records.

Convert, then verify meaning as well as syntax

  1. Choose CSV → JSON under Conversion and select the correct CSV delimiter. Paste or import the supported UTF-8 text.
  2. Choose Convert data. Resolve a reported header or row-width problem in a copy of the source; do not silently discard a troublesome column.
  3. Inspect the number of records and keys, then check an identifier with a leading zero, an empty field, a quoted comma, and any Hebrew text.
  4. Copy or download the output as a separate file. If needed, inspect the result with the JSON formatter and validator. The formatter preserves numeric spellings without rounding. Keep identifiers as strings when the receiving application requires exact identifiers, and compare exported data with the original.
  5. Import a small sample into the receiving system. Confirm that it keeps the types and values your workflow needs before importing the whole set.

Imports are limited to 1 MiB, with 10,000 records, 200 columns, and 200,000 cells including headers. The editor also caps input at 1,048,576 characters. Stay within every applicable limit.

Generated JSON has a separate limit of 8,000,000 UTF-16 code units, predicted before result objects or output strings are built. The count includes header names repeated in each record, escaped characters, separators and pretty-print indentation. This is not an 8 MB file-byte limit: even a small CSV can exceed it when long headers repeat many times. Use fewer rows or shorter headers and cell values.

JSON-to-CSV conversion rejects unsafe integers and numeric literals whose decimal value would change through JavaScript number conversion. Overflow, underflow to zero, and loss of a negative-zero sign also stop the operation. Store long identifiers and precise decimals as quoted strings before parsing. Equivalent spellings such as 1.0 and 1e3 may still become 1 and 1000; quote values when their original spelling matters. This is a rejection safeguard, not arbitrary-precision conversion, and cannot recover digits already rounded in a source system.

For exact line-ending preservation inside quoted CSV fields, import the UTF-8 file and convert before editing that input. The companion tool retains the imported source for this path. Typing, pasting, or editing uses the browser textarea’s LF line endings; copying text manually can normalize them too. Download the result when generated line endings matter.

Valid JSON can still hold the wrong information. A string, a number, an empty string, and null have different meanings. Do not rely on a round trip through CSV to preserve every type distinction. Nested objects, arrays, and missing keys need an explicit data model; use another workflow when a flat table is insufficient.

Review formula-like values before opening a spreadsheet

A spreadsheet may interpret imported text beginning with characters such as = as a formula. CSV quoting solves delimiter structure, not every spreadsheet interpretation risk. OWASP’s CSV injection guidance explains this boundary.

When exporting JSON to CSV, the formula-like cell and header apostrophe-prefix option is on by default. Review its warnings and the changes it makes, especially for untrusted text. A prefixed apostrophe can alter the literal value, and handling varies by receiving application. Keep a data-faithful copy separate from a spreadsheet-oriented copy. Test the actual import settings and do not describe an escaped file as safe for every future use.

What to check when conversion fails

  • Unexpected extra columns: check the delimiter and whether embedded commas were quoted.
  • A header or duplicate-property error: give each CSV column one distinct, non-empty name and each JSON object unique property names. The converter rejects duplicates rather than silently discarding a value.
  • Missing leading zeros: inspect the source first. A spreadsheet may already have removed them before export.
  • Hebrew appears corrupted: confirm the source encoding. Changing delimiters cannot repair text decoded with the wrong character set.
  • A large file is rejected: follow the displayed limit or split it using a data-aware workflow. Splitting at an arbitrary newline can break quoted multiline fields.

The companion converter works locally in the browser and makes no processing-service request with your input. Browser memory still limits the amount of data it can handle. The receiving application, downloaded files, and the host site’s analytics are separate parts of the privacy picture.