# Predictive Text on Phones

> How your phone keyboard guesses the next word you will type. Why that small guess is the same idea a large language model is built on.

**Course:** [AI Assistants and Chatbots](https://learn.benibix.com/learn/ai-use-case)  
**Time:** 8 min

**After this lesson you can:**

- Say how predictive text picks the words it shows
- Explain next-token prediction in one sentence
- Tell a keyboard model from a large language model

## How does your phone guess the next word?

Predictive text is the part of a phone keyboard that guesses your next word. As you type, it shows a few words above the keys, often three. Tap one to add it, or keep typing. A small language model makes the guesses, from common phrases and from the way you write.

A language model is a model that has learned which words tend to follow which. After 'see you', words like 'soon', 'later' and 'tomorrow' are likely. The keyboard ranks the most likely few and shows them. It updates the list with every letter you type.

It also learns your habits. If you always write 'see you at the gym', the word 'gym' starts to show up sooner for you than for others. Some keyboards learn on the phone itself, so the words you type do not have to leave it to improve the guesses.

This is next-token prediction, the same idea a large language model is built on. A big model also guesses one piece of text at a time, adds it, and guesses again. The difference is size. The keyboard model is small enough to run fast on your phone, and a large model runs in a data centre.

A small language model sits between the two. It is far smaller than the biggest models, but it can still write whole sentences. Phones and laptops now run some of these on the device itself. That keeps your words on your device, and it works with no internet.

Predictive text can also lead you. If you always tap the first word it offers, your messages start to sound like everyone else's. It can also keep offering a word you typed by mistake once. Most phones let you remove a word it learned, or turn the guesses off.

![A loop. The text so far goes into the model, the model gives a chance for each next token, one token is picked and added to the text, and the loop starts again.](https://learn.benibix.com/diagrams/llm-next-token.svg)

### Example: Jack's quick message

Jack types 'running 10 minutes late, see you'. The keyboard offers soon, there and later.

He taps 'soon', and the message is done. Over a day of messages, he taps a guess for about 1 word in 5 instead of typing it.

**Words in this lesson:** [Predictive Text](https://learn.benibix.com/term/ai/predictive-text), [Language Model](https://learn.benibix.com/term/ai/language-model), [Next-Token Prediction](https://learn.benibix.com/term/ai/next-token-prediction), [Small Language Model (SLM)](https://learn.benibix.com/term/ai/small-language-model)

**Question:** How is predictive text like a large language model?

**Answer:** Both guess the next piece of text from the text so far. Both look at the text so far and guess what comes next. That is next-token prediction. A keyboard does it once for one word. A large model does it again and again for a whole reply.

---

Source: https://learn.benibix.com/learn/ai-use-case/predictive-text-on-phones
Author: Behnia Zarei (a developer in Vienna, Austria, who builds and teaches data, AI and business intelligence at BENIBIX)
Publisher: BENIBIX Learn (https://learn.benibix.com/)
Last updated: 2026-11-18
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote freely with attribution.
