Extract numbers from text means identifying every numeric value inside a block of prose and returning them as a clean list or as summary statistics, with explicit, switchable rules defining what qualifies as a number. The numbers live wherever they appear: inside invoices, reports, log files, survey exports and free-form notes, and the extractor returns them one per line, comma-separated, or summarized as count, sum, minimum, maximum and average. By default the rules recognize integers and decimals, including negative values, a leading plus, and bare decimals written without a leading zero. Thousands separators and scientific notation sit behind independent toggles, and a separate dedupe toggle collapses repeats while keeping the first-seen order. The output reports how many numbers were found so every extraction is auditable against your expectation before you copy it onward. The whole pass runs in a single linear scan capped at one million characters, so even large pastes return instantly.
What Counts as a Number: The Switchable Rules
Most "extract the digits" tools make a single guess about what a number is and live with the consequences. The Extract Numbers from Text tool does the opposite: every rule that decides whether a character sequence becomes a number is a switch on the page, and the page states what each switch does before you flip it.
By default the tool recognizes integers and decimals, including:
- Negative values such as -42 and -3.14.
- A leading plus such as +7 when you keep that toggle on.
- Bare decimals written without a leading zero, like .5 in "latency was .5 seconds".
The sign rule is deliberately careful. A minus is kept only when it is genuinely attached to the number, at the start of a line, after whitespace, or after an opening bracket or a comparison operator. The expression 5-3 therefore yields five and three, not five and minus three, so arithmetic in pasted text is not misread as a negative value.
Thousands separators get their own toggle because they are the single most common source of false matches. When grouping is well formed, 1,000,000 is one number rather than three fragments; when grouping is malformed, the extractor splits at the malformation. Switching the toggle off lets you treat commas as ordinary punctuation and pull literal digit runs instead, which is what you want for data that uses commas in non-numeric positions.
Scientific notation sits behind its own toggle, off by default, because 1e5 in ordinary prose is more often a typo or an identifier than a hundred thousand. When your data is genuinely scientific, switch it on and the tool will recognize mantissa-exponent pairs and report them as a single number.
Pulling Numbers from Text in Three Steps
- Paste the text containing the numbers you want to extract into the input field. The tool accepts up to one million characters and processes the whole paste in a single linear pass.
- Adjust the rules if needed. Turn the thousands-separator, scientific-notation, or dedupe toggles on or off, and adjust the sign rule depending on your data.
- Choose list or statistics output. List mode returns one number per line, and statistics mode returns the count, sum, minimum, maximum and average. The found count is always shown so you can audit the result against your expectation before you copy it.
Because the tool runs entirely in your browser, nothing is uploaded, stored, or attached to an account, so figures from confidential documents can be pasted without a privacy question hanging over the workflow.
Reading the Statistics Block
Switching to statistics output turns the list into a five-line summary that can be pasted straight into a report or a ticket:
- Count, the exact number of numeric values the tool found in your paste. This figure is exact, not an estimate.
- Sum, the total of every extracted value, computed with standard JavaScript double-precision arithmetic.
- Minimum and maximum, the smallest and largest extracted values, useful for sanity-checking the range of measurements or amounts.
- Average, sum divided by count, computed in the same double-precision domain.
The page is honest about what that double-precision arithmetic means. Sums of clean decimals can carry the familiar floating-point residue that turns 0.1 + 0.2 into 0.30000000000000004, which matters at the fifteenth significant digit and not before. For ordinary invoicing, reporting, or quick aggregation, the residue is invisible; for exact-decimal accounting, a spreadsheet with arbitrary-precision arithmetic already exists for a reason.
Where These Numbers Come From: A Quick Comparison
The same extractor handles a wide range of pasted inputs. The differences are mostly in what each source looks like and what kind of output you tend to want from it.
| Source of pasted text | Typical content | Useful output mode |
|---|---|---|
| Invoice or receipt | Line totals, tax, dates, mixed with words | Statistics for the total, list for the line items |
| Application log file | Latencies, status codes, byte counts | Statistics for min, max and average latency |
| Survey or CSV export | Numeric answers mixed with free text | List for the raw values |
| Report or prose document | Embedded figures like "Q3 revenue was $2.4M" | List to feed into a spreadsheet |
| Raw HTML or JSON snippet | Numbers embedded in markup | List, with dedupe on for repeated IDs |
Switching between list and statistics is free, so the same paste can be inspected in either mode. The tool reports how many numbers it found in each run, which lets you cross-check the count against an expectation like "I pasted twenty log lines and expect about fifteen measurements".
Where the Rules End: Scope and Limits
An honest extractor tells you what it does not recognize. The tool's stated scope is ASCII digits only: full-width digits and numerals from other scripts pass through unextracted, and the page says so rather than half-supporting them. If you paste accented characters alongside Arabic-Indic numerals, the Arabic-Indic numerals will not be picked up unless you first convert them with a separate normalization step.
What the tool deliberately does not do is interpret. There is no currency conversion, no unit awareness, no date parsing, and no attempt to guess what a number means inside its sentence. "Q3 revenue was $2.4M" yields 2.4, not a parsed currency value; "by 2024-03-15" yields 2024, 03 and 15 as three independent integers, not a date. That restraint is the point: an extractor you can trust is one whose rules you can read, and the rules are pinned by more than twenty documented test cases covering sign boundaries, malformed grouping, currency, percentages, embedded digits and line-ending variants.
Two further behaviors worth knowing:
- An input with no recognizable numbers is reported as exactly that, an empty list and a count of zero, rather than treated as an error or a malformed paste.
- Embedded numbers are extracted wherever they occur: item123 yields 123, $19.99 yields 19.99, and 50% yields 50, because the job of a number extractor is to find every number, not to judge its context.
For phone numbers in the same kind of source text, the guide to extracting phone numbers from any text walks through a parallel workflow for that extractor.
Why the Extraction Runs Locally in Your Browser
The number extractor is part of this site's broader extractor family, alongside the email, URL and phone extractors, and it shares the same architecture: the entire scan happens client-side, in the browser tab, with nothing transmitted to a server. The grammar itself is a single-pass scan over your text, applying the documented attachment rule for signs and computing statistics in double precision with the limits stated on the page.
For one-off pastes the local execution is invisible. For invoice totals, log files, and any paste that contains figures you would not want sent over the network, it removes the privacy question that would otherwise hang over the workflow. The implementation methodology is the same one any reviewer would write: define the grammar, list the toggles, state the precision limits, and pin the behavior with test cases. Twenty-plus test cases pinned the rules before the implementation was merged, which is what makes "the rules on the page are the rules the tool applies" a statement with evidence behind it rather than a marketing line.
The practical upshot is that the workflow reduces to a paste, a couple of toggles, and a copy. Paste the document, switch the rules that match your data, pick list or statistics, and check the count against your expectation before you send the numbers anywhere else.