How to Convert CSV to JSON in Python (csv.DictReader and pandas)

CSV and JSON represent data very differently — CSV is flat rows of strings, JSON supports nesting and real types — so “converting” between them always involves a decision about how to handle that mismatch. Here’s how to do it in Python, and where it quietly goes wrong.

Standard library: csv.DictReader

No dependencies needed for a straightforward conversion:

import csv
import json

with open('data.csv') as f:
    rows = list(csv.DictReader(f))

print(json.dumps(rows, indent=2))

DictReader uses the first row as keys automatically. The catch: every value comes out as a string, even ones that look numeric or boolean. "42" stays "42", not 42; "true" stays the string "true", not a JSON boolean. If the consumer of your JSON expects real numbers or booleans, you need to coerce fields explicitly:

for row in rows:
    row['age'] = int(row['age'])
    row['active'] = row['active'].lower() == 'true'

There’s no way to infer this automatically and reliably — a column of "007" values could be a numeric ID that should stay a zero-padded string, not become the integer 7. This is a modeling decision, not something the conversion can guess correctly on its own.

pandas: shorter, with the same type caveat

import pandas as pd

df = pd.read_csv('data.csv')
print(df.to_json(orient='records', indent=2))

orient='records' produces the same shape as DictReader — a JSON array of objects, one per row. pandas does attempt automatic type inference (numeric-looking columns often come out as actual numbers), which solves the string problem for the common case — but it can also silently misinfer types on edge cases (a column that’s almost all numbers with one text value becomes object dtype, and a column of IDs might get coerced to float64 if any value is missing, turning "1042" into 1042.0). Always spot-check the output rather than assuming pandas guessed correctly.

Nesting: CSV has none, JSON might need it

If your target JSON schema needs nested objects (e.g. flattening address_street, address_city columns back into a nested address: {street, city} object), neither DictReader nor plain pandas do this automatically — you’d write an explicit reshape step, or reach for pandas.json_normalize’s inverse manually.

Skipping the script for a one-off file

If you just have a CSV file and need JSON output right now — no repeatable pipeline, no type-coercion logic to maintain — CSV to JSON does the conversion directly in the browser, auto-detecting headers, with the result immediately visible so you can confirm types look right before using it. Going the other direction, JSON to CSV flattens a JSON array of objects back into a CSV, which is the more common need when exporting API data to a spreadsheet.

The quick decision

• One-off file, need JSON now, no script to maintain → CSV to JSON
• Repeatable pipeline, need explicit control over type coercion → csv.DictReader + manual field coercion
• Already doing data analysis in pandas anyway → df.to_json(orient='records'), but verify the inferred types with JSON Formatter before trusting them downstream