Date Format Normalizer
Paste text whose dates came out as 03/04/2024, 5 Jan 2024, 2024-01-05 and 5.1.24 all at once, and get every one of them rewritten into the single format you choose. The day-first vs month-first question is settled from the text itself. Nothing leaves your browser.
About this tool
Real text almost never uses one date format. A document that passed through a spreadsheet, an
export, an email thread and someone's handwriting ends up with 03/04/2024, 15 April 2024,
2024-04-22 and 5.1.24 sitting in the same paragraph. This tool finds every one of them and
rewrites them into the single format you pick, leaving the rest of the text byte-for-byte
untouched.
The interesting part is not the rendering — it is deciding what 03/04/2024 means. Instead of
guessing per date, the tool reads the whole text first and lets the dates that can only be one
thing settle it for the rest: a single 15/04/2024 in the paste proves the source writes
day-first, so 03/04/2024 is 3 April, not 4 March. When nothing in the text settles it, the
month-first reading is used and the report mode tells you exactly which dates were guessed.
Worked example
Input:
Order 4471 was placed 03/04/2024 and shipped 15/04/2024.
The invoice is dated 22 April 2024 and falls due 2024-05-06.
Support ticket opened Friday, 3 May 2024 at 2:30 PM.
Output with the default ISO 8601 format:
Order 4471 was placed 2024-04-03 and shipped 2024-04-15.
The invoice is dated 2024-04-22 and falls due 2024-05-06.
Support ticket opened 2024-05-03T14:30.
Three source forms, one target form. 15/04/2024 settled the day-first reading for
03/04/2024; the weekday name was absorbed into the date it belonged to; the clock time came
along with its date; and the sentence's full stop survived.
Switch the format to Day first with a dot separator and the same text renders
03.04.2024, 15.04.2024, 22.04.2024, 06.05.2024. Switch to Year first with the
separator set to none and scan 5 Jan 2024.pdf becomes scan 20240105.pdf — the stamp that
makes a folder of files sort itself.
What it recognizes
| Form | Examples |
|---|---|
| ISO 8601 | 2024-01-05, 2024-1-5, 2024-01-05T14:30:00Z, 2024-01-05 14:30 |
| Numeric | 01/05/2024, 5.1.2024, 5-1-24, 2024/01/05 |
| Month name | January 5, 2024, Jan. 5th, 2024, 5 Jan 2024, 5th of January 2024 |
| With a weekday | Friday, 5 January 2024, Fri, 05 Jan 2024 14:30:00 +0100 |
| Unix epoch | 1704465000, 1704465000123 — only when you switch timestamp detection on |
A clock time written next to a date is picked up with it, including the at 2:30 PM phrasing and
a trailing Z, UTC or +01:00.
Output formats
ISO 8601 · year-first, day-first and month-first numeric (with your choice of dash, slash, dot,
space or no separator) · full or abbreviated month names in either order · RFC 2822 email dates ·
Unix seconds or milliseconds · or any chrono/strftime pattern you want, such as %B %-d, %Y or
%Y%m%d.
Three ways to get the result back: the text with the dates rewritten in place, a list of just the normalized values one per line for pasting into a spreadsheet column, or a report giving each date's line and column, what was found, what it became, and whether the day/month order had to be guessed.
Limits and edge cases
- Up to 1,000,000 bytes of text per run. Split anything larger and run the parts separately.
- Strings that look like dates but are not —
2024-02-30,13/13/2024, a version number like1.2.2024where no valid reading exists — are left exactly as written, never guessed at. - Bare 10- and 13-digit numbers are only read as timestamps when you ask for it, and then only when they land between 1973 and 2100. Order numbers and phone numbers stay untouched.
- Two-digit years need a century: the pivot decides it, defaulting to the POSIX rule where 68 and below mean 20xx.
- The timezone setting moves only the dates that carry an explicit offset. A date written with no zone has nothing to convert from, so it stays where it is.
- Everything runs in your browser. No upload, no account, no network round-trip, and the same input always produces the same output.
FAQ
How does it decide whether 03/04/2024 is 3 April or 4 March?
It reads the whole text before rewriting anything. Any numeric date with a field above 12 can
only be read one way — 15/04/2024 must be day-first, 04/15/2024 must be month-first — and
those dates set the reading for every ambiguous date in the same text. If the text contains no
such date, or contradicts itself, the month-first reading is used and the report output mode
flags every date that was guessed, so you can check them instead of trusting them. You can also
skip the inference entirely by setting the reading to Day first or Month first yourself.
Does it change anything other than the dates?
No. In text mode the output is the input with each detected date span swapped for its
rewritten form; every other character, including line breaks and punctuation, is passed through
untouched. Two things are deliberately absorbed into the date they belong to: a weekday name
written in front of it (Friday, 5 January 2024 becomes one date) and a clock time written after
it. A sentence-ending full stop after 2:30 pm. is kept — only the pm is treated as part of
the time.
What happens to times and timezones?
A time found next to a date travels with it and is rendered on the 24-hour or 12-hour clock, with
seconds shown only when the source had them. Turn Keep the clock time off to reduce
everything to bare dates. The timezone field moves the dates that carry an explicit offset — an
ISO stamp ending in Z or +01:00, an RFC 2822 date, a detected epoch value — into UTC, an IANA
zone such as Europe/Berlin with daylight saving applied per date, or a fixed offset. That can
change the calendar day, which is the point. Dates written without any zone are left exactly
where they are.
Can I get a format that is not in the dropdown?
Yes — choose the custom format and write a strftime pattern. %d.%m.%Y gives 05.01.2024,
%B %-d, %Y gives January 5, 2024, %Y%m%d gives 20240105, %A %d %b %y gives
Friday 05 Jan 24. The common fields are %Y/%y for the year, %m/%d for month and day
(%-m/%-d without the leading zero), %B/%b for month names, %A for the weekday,
%H:%M:%S for a 24-hour clock, %I:%M %p for a 12-hour one and %z for the offset. An invalid
pattern is reported rather than half-rendered.
Why are my order numbers not being converted?
Because timestamp detection is off by default. Most long numbers in real text are order ids, account references or phone numbers, and turning every 10-digit number into a date would ruin more documents than it fixes. Switch Also read bare 10- and 13-digit numbers as unix timestamps on when you are working with log exports or API payloads where epoch values really are dates — and note that even then only values between 1973 and 2100 are accepted.
How do I check what it actually did?
Set the output to report. You get a # header with how many dates were found, the mix of
source forms detected, and which day/month order was chosen and why, followed by one
tab-separated line per date: its line and column in the source, the original string, the
rewritten value, and an ambiguous marker when the day/month order had to be inferred. It pastes
straight into a spreadsheet if you want to review a large document column by column.
Developer & Automation Access
Run it from the terminal
Same engine as this page, headless — via the gizza CLI:
gizza tool date-format-normalizer "Order 4471 was placed 03/04/2024 and shipped 15/04/2024.
The invoice is dated 22 April 2024 and falls due 2024-05-06.
Support ticket opened Friday, 3 May 2024 at 2:30 PM."New to the CLI? Get gizza →
Open it by URL
Pre-fill and auto-run this tool with query parameters — the names match the API/CLI:
https://gizza.ai/tools/date-format-normalizer/?text=Order%204471%20was%20placed%2003%2F04%2F2024%20and%20shipped%2015%2F04%2F2024.%0AThe%20invoice%20is%20dated%2022%20April%202024%20and%20falls%20due%202024-05-06.%0ASupport%20ticket%20opened%20Friday%2C%203%20May%202024%20at%202%3A30%20PM.&output_format=iso&custom_format=%25d.%25m.%25Y&separator=dash&month_style=full&year_style=four&leading_zeros=true&input_order=auto&two_digit_year_pivot=68&keep_time=true&time_style=24h&output_timezone=source&detect_timestamps=true&output_mode=textMachine-readable descriptor: tool.json — title + parameters JSON Schema for agents.