First letter of every word
Each whitespace-separated word contributes the first letter it contains - punctuation and digits before that letter are skipped, so O’Brien gives O and & Partners gives P. A word with nothing letter-shaped in it contributes nothing, so a phrase like "Q1 financial reports" ignores the digit and still finds the F in financial.
The letters are capitalized and kept in order, duplicates included, so the acronym reads exactly as the phrase builds it. Nothing is dropped to make a prettier initialism, which is why "Do Not Disturb" comes back as DND rather than DN - the tool does not silently edit your phrase into a nicer shape.
The extraction follows a fixed rule applied to the text, run on the built-in example and in your browser for everything you type, so the same rule produces the same letters everywhere.
Letter by letter, the mapping stays visible
Below the acronym, a table pairs each initial with the word it came from, in the order they appear. That mapping is the reusable part when you need to explain the acronym to a reader - or to check which word a letter actually belongs to when a phrase repeats an initial or carries a connecting word.
The same pairs that build the acronym drive the hero, the letter-by-letter table and the copy, so what you see cannot disagree with what you paste: the hero shows the initials, the table shows where each one came from, and the download carries both.
Deterministic: no randomness, no seed
The same phrase always gives the same acronym, on any device, any year. There is no button to press and no re-roll - the answer is a fixed rule applied to what you typed, which is what makes it trustworthy enough to paste into copy.
There is no randomness and no seed anywhere in the path, so the result never depends on when it was asked or which machine asked it. That determinism also means you can reproduce exactly what a colleague typed and get the same initials back, which is the property a naming tool needs.
If no word in the phrase has a letter in it - only punctuation or digits - the page refuses rather than inventing initials, because an acronym with nothing to extract is a guess, not a result.
Private and trademark-honest
The phrase is the only input and it is noUrl, so it never enters the address bar, the share link, the copied summary or the print stamp, and the page makes no network requests at all. The same boundary keeps the mapping, the copy and the download inside the tab.
The page also prints that it checks no trademark, naming or registration: some acronyms are trademarks (UNESCO, NASDAQ, OECD) and some are words in their own right, and deciding which one you have is not something initials can answer. It extracts what you typed and stops there, which is the honest limit of a first-letter tool.
Initialism, acronym, and the word you did not mean
The initials come out as letters, and what you do with them is a separate decision. Two words describe the results: an initialism is read letter by letter (FBI, HTML), an acronym is read as a word (NASA, UNESCO). Nothing in the extraction knows which one you have, because pronounceability is a property of the letters you got, not of the rule that produced them - and the same phrase can give one or the other depending on which words you choose to keep.
That is also where the accidents live. Four initials can spell a word nobody wants on a slide, and a string with no vowels reads as a stutter when it is said out loud. This page fixes neither: it never drops a letter to make the string pronounceable, because dropping one would break the mapping between the acronym and the phrase, and that mapping is the whole point of the table under the result. The fix is editorial - choose different words, or accept that this string is read letter by letter.
Two checks belong to you before it ships: say the letters out loud, and search the string to see what it already means. Some acronyms are trademarks and some are ordinary words (this page says plainly that it checks neither), and a three-letter string that is free in one industry is usually taken in another. The generator stops at an honest extraction; the naming decision is the part that needs a human. If the string will end up on a legal document or a product label, get it in front of whoever owns the naming before it is printed, because a rename after launch is not measured in letters.
Danfio