My tools

Text extractor

Paste any text — emails, links and numbers come out as a clean list.

How to use it

  1. Paste your text — emails, links and numbers are scanned at once, and the hit counts on the chips update as you type.
  2. Emails and links are on by default. Switch on “Numbers” only when you really want figures: dates, version numbers and order codes come along too.
  3. Copy the list or download it as .txt. Both follow the order on screen — emails, then links, then numbers, one item per line.

Example

Input

Berlin offsite — booking confirmations
Contact: sarah.k@acme.com (she/her), cc info@acme.com
Deck: https://acme.com/deck/berlin.pdf
Brief: https://acme.com/brief — 3 pages
Backup address: sarah.k@acme.com

Extracted (defaults)

sarah.k@acme.com
info@acme.com
https://acme.com/deck/berlin.pdf
https://acme.com/brief

Default options only (emails and links on, numbers off, duplicates removed). The repeated backup address is dropped and counted, and the “3” stays in the hidden Numbers group, so it does not appear in the list — the Numbers chip still shows it is there.

Rules, limits and known issues

  • Pattern matching, not verification: an address that looks like an email is never checked against a server, and a link may well be out of date. Extraction favours catching real matches, so skim the list for the odd false positive.
  • Links must start with http://, https:// or www. A bare domain such as example.com is not treated as a link, and neither are ftp: or mailto: — pasted prose is full of sentence-final “.com” that would otherwise be noise.
  • Trailing sentence punctuation is trimmed from links, while brackets that belong to the URL are kept: https://zh.wikipedia.org/wiki/Foo_(bar) keeps its closing bracket, a “)” ending a sentence does not.
  • An email must look like local@domain.tld. File-name lookalikes such as logo@2x.png are ignored, but .zip, .app and .dev are real top-level domains, so those addresses are kept.
  • Numbers are taken exactly as written: thousands separators and decimal marks stay as typed, and 1.234 is never guessed to be 1234 or 1.234 — that reads differently in different countries. There is no sum, average or currency conversion, and a minus sign is not part of the number (−5 gives 5).
  • Because nothing is interpreted, dates and versions split up: 2026-01-05 gives three numbers, v1.2.3 gives 1.2 and 3, and an order code like A12345 gives 12345. Leave “Numbers” off when you only want contacts.
  • Text inside a link or an email address is never extracted again: user123@mail.com gives one email and no number, and https://x.com/item/2026 gives one link and no number.
  • Duplicates are compared per type: emails ignore case, links are compared exactly (paths are case-sensitive), numbers by their written form — “1,000” and “1000” are two different entries.
  • Full-width (１２３) and Arabic-Indic (١٢٣) digits are recognised. Chinese and Arabic pages offer a “Half-width digits” switch that turns them into 0-9, which is what spreadsheets expect; it is off by default so your text is never rewritten silently.
  • Above 2,000 matches only the first 2,000 are listed, while copy and download still contain everything; above 200,000 characters the result updates about 0.6 s after you stop typing. No phone numbers, ID-card or tracking numbers (they need country-specific rules), no file upload and no history — everything runs in your browser.

What people usually need next, once the list is out.

Text cleaner

Find & replace