T9 to Text Converter — phone keypad, both directions

The T9 to text converter above turns a message into the digit sequence a numeric phone keypad would produce, and turns a digit sequence back into text. It handles both of the input methods those keypads supported: multi-tap, where each key is pressed until the wanted letter appears, and predictive entry, where each letter takes one press and a dictionary resolves the ambiguity afterwards.

Arb Digital publishes it because the keypad layout outlived the phones. It is still the reason a phone number can be spelled as a word, it still governs vanity numbers, and it appears constantly in puzzles, escape rooms and capture-the-flag challenges where a string of digits is really a string of letters. The mapping itself is a published international standard, not a convention someone invented, which is why the same layout appears on hardware from every manufacturer.

What This T9 to Text Converter Does

In multi-tap mode, encoding walks each character to its key and repeats the digit as many times as that character sits along the key's sequence. The letter S is the fourth character on key 7, so it becomes 7777. Decoding reverses that exactly, splitting the input into runs of identical digits and cycling through each key's characters. Because the mapping is one-to-one in this mode, a message encoded and decoded returns identical to what went in.

In predictive mode, encoding puts one digit per letter: the word HELLO becomes 43556. Decoding cannot simply reverse that, because 43556 could in principle be any of thousands of letter combinations. The tool therefore matches each digit group against a built-in list of common English words and reports every word whose digit signature fits, in place of pretending there is a single answer. Where nothing matches, it says so rather than guessing.

Alongside the conversion, the grid counts key presses both ways and shows the saving predictive entry delivered. That saving is the entire commercial reason predictive text existed, and it is usually larger than people expect. If you are working with a different letter-to-number scheme, our letters to numbers converter handles the A=1 alphabet-position mapping, which is a different code and not interchangeable with this one.

How to Use It

1. Pick a direction — text into digits, or digits into text.
2. Choose multi-tap or predictive. Multi-tap is exact; predictive is one press per letter and needs the dictionary to decode.
3. Type or paste your message. In digits-to-text mode, use 0 for a space between words.
4. Set the separator if you are producing multi-tap output. A space between groups keeps the result unambiguous.
5. Read the press counts in the grid to see how much predictive entry actually saved on this specific message.

The Keypad Map and How It Is Calculated

The layout is fixed by ITU-T Recommendation E.161, Arrangement of digits, letters and symbols on telephones and other devices that can be used for gaining access to a telephone network. Key 2 carries ABC, 3 carries DEF, 4 GHI, 5 JKL, 6 MNO, 7 PQRS, 8 TUV and 9 WXYZ. Key 1 carries punctuation and key 0 carries the space. Keys 7 and 9 hold four letters each because twenty-six letters do not divide evenly into eight keys; every other letter key holds three.

Multi-tap press count follows directly from position. A letter in position i on its key costs i presses, so A costs one and C costs three. Averaged over normal English text the figure lands near two presses per character, because the common letters are not conveniently placed near the front of their keys: E is second on key 3, T is first on key 8, and S is fourth on key 7. The tool sums the real cost for your specific message rather than applying an average.

Predictive press count is simply one per character. The saving shown in the grid is the difference expressed as a percentage of the multi-tap total. For ordinary prose it usually falls between forty and fifty-five percent, which on a device with no other input method was a substantial difference. The standards work behind this layout sits with ITU-T, the Telecommunication Standardization Sector, which is why the arrangement is identical across manufacturers and countries.

Why Predictive Decoding Has No Single Answer

Eight keys have to carry twenty-six letters, so each digit stands for three or four possibilities. A five-letter code therefore covers between 243 and 1,024 letter combinations. Almost all of them are not words, which is what makes the scheme workable at all, but a meaningful number of them are, and those are called textonyms.

The classic example is 4663, which is GOOD, HOME, HOOD, HONE, GONE and INNE among others. Handsets resolved this by ordering candidates by frequency and letting the user cycle with a "next" key, and by learning which alternative a particular user picked. The consequence was a distinctive class of typing error where an entirely valid but wrong word was sent, because the sender never looked at the screen.

This tool does not pick a winner. It lists every match its dictionary contains, in the order they appear, so the ambiguity is visible rather than hidden behind a guess. If your word is missing, that is a limitation of a built-in list, not evidence the code is invalid — the same is true of every predictive implementation, which is why handsets let users add words.

Multi-Tap, Pauses and the Ambiguity of Repeated Keys

Multi-tap has one structural problem: two consecutive letters on the same key. The word HELLO needs L twice, and L is 555. Typed without a break that becomes 555555, which the handset reads as six presses on one key, cycling past L, then past 5, and landing somewhere else entirely. Real handsets solved it with a timeout — wait a moment and the cursor advanced — or with a dedicated pause key.

The separator control on this page is that pause made explicit. With a space between groups, HELLO is 44 33 555 555 666 and decodes exactly. With no separator it becomes 4433555555666, and the tool decodes it the way a handset would, by cycling, which produces something different. Switching the separator off and comparing the two outputs is the quickest way to see why the pause existed at all.

Cycling itself is worth understanding: pressing a key more times than it has characters wraps around to the start. Six presses on key 5, which holds J, K, L and the digit 5, lands on the second character, K. That behaviour is what lets a single key produce digits as well as letters, and it is faithfully reproduced here.

Where This Encoding Still Appears

Vanity phone numbers are the most visible survival. A number advertised as 1-800-FLOWERS is dialled as 1-800-3569377, and the conversion is exactly the predictive encoding on this page. Businesses choose numbers whose digits spell something memorable, and the mapping is the standard one, which is why it works on any handset.

The second common context is puzzles. Digit strings that decode to text through the keypad are a staple of escape rooms, geocaching and beginner cryptography challenges, usually alongside other simple substitutions. If you are working through a puzzle and the keypad reading gives nothing, the same string is worth testing against a shift cipher with our Caesar cipher translator, against ROT13, or against a symbol alphabet with the Morse code translator.

A third context is accessibility and legacy interfaces. Interactive voice response systems still spell surnames and reference codes through the keypad, and any system that asks a caller to "enter the first four letters of your surname using the number keys" is using this exact map. Our character counter and case converter are useful for preparing text before encoding it.

Common Mistakes to Avoid

• Confusing predictive digits with multi-tap digits — HELLO is 43556 in one and 44 33 555 555 666 in the other. Feeding one into the other produces nonsense.
• Omitting the pause between repeated keys — two letters on the same key merge into a single longer press and decode to something else.
• Expecting a unique predictive answer — digit codes map to many words, and only a dictionary and context can choose between them.
• Using the alphabet-position code by mistake — A=1, B=2 is an entirely different mapping and shares no digits with the keypad layout.
• Assuming case survives — a numeric keypad carries no case information, so any capitalisation in the original is lost in the digits.

Related Free Tools From Arb Digital

Map letters to their alphabet positions with the letters to numbers converter, shift an alphabet with the Caesar cipher translator, apply the fixed thirteen-place shift with ROT13, translate dots and dashes with the Morse code translator, or count what you have typed with the character counter and word counter. The full free online tools hub lists every developer tool we publish.

Frequently Asked Questions

This page explains a published keypad standard for educational purposes. The built-in word list is a limited sample of common English and is not a complete dictionary.