Have you ever pasted a long document into an AI chatbot and watched it stop halfway? Or seen an AI plan advertise a “128K token” limit and wondered what that actually means? Tokens sit behind almost everything AI chatbots do, and once you understand them, a lot of confusing AI behaviour suddenly makes sense.

In this guide, you’ll learn what a token in AI is, how your words get chopped into tokens, and why token limits explain the message caps, forgetful chats, and prices you run into every day. Plain English, no math needed. If you’re completely new to this topic, our simple explanation of what AI is is a good place to start.

What is a token in AI?

A token is a small piece of text that an AI model reads and writes. It’s the basic unit every AI language model works with. A token is not the same as a word. It can be a whole short word, part of a longer word, a punctuation mark, or even a space attached to a word.

OpenAI’s help guide gives some handy rules of thumb for English text:

• 1 token is roughly 4 characters
• 1 token is about three quarters of a word
• 100 tokens come to roughly 75 words

So a 1,000 word blog post like this one is somewhere around 1,300 to 1,400 tokens. Numbers, code, and unusual words push the count higher.

How AI turns your words into tokens

Before an AI model ever sees your message, a piece of software called a tokenizer splits it into tokens. Each token then gets an ID number, because the model works entirely with numbers, not letters. Common short words usually stay whole. Longer or rarer words get broken into smaller chunks.

GPT models use a subword method called Byte-Pair Encoding, as Microsoft’s guide to tokens explains. To give you a feel for the scale, OpenAI notes that the famous quote about missing all the shots you don’t take is 11 tokens long.

Language matters too. OpenAI points out that the Spanish phrase “Cómo estás” takes 5 tokens for just 10 characters. Many non-English languages break into more tokens per word, which makes the same request slower and more expensive than it would be in English.

Why do AI models count tokens instead of words?

Because tokens are literally how these models think. A large language model generates text by predicting the next token, one token at a time, then feeding its own output back in and predicting the next one again. Words are for us. Tokens are for the model.

Subword tokens also give models flexibility. When a model meets a brand new word, a typo, or an unusual name, it can still handle it by piecing together smaller chunks it already knows.

What is a context window?

Every AI model has a limit on how many tokens it can handle at once. That limit is called the context window, and it covers your input and the model’s output together. When a conversation grows past it, the oldest parts effectively fall out of view.

This is why a long chat starts to “forget” things you said earlier. The AI isn’t being lazy. Those early messages simply no longer fit inside the window it can see. The practical fix is to start a fresh chat and paste in only the details that still matter, or keep your prompts focused from the start. Our guide on writing better AI prompts helps a lot here.

Why tokens decide what AI costs

Most paid AI services charge by the token, and input tokens are often priced differently from output tokens. Free plans work the same way underneath, they just cap how much you can use. That’s why generative AI tools talk about tokens so much: they measure the actual work the model does.

From my own experience connecting AI tools to websites and small projects, this was the first surprise: the bill counts tokens, not questions. A one-line question is nearly free. The same question with a 20-page document pasted under it costs many times more, because every word of that document becomes tokens the model has to read.

How to see tokens for yourself

The easiest way to make all this real is to try OpenAI’s free Tokenizer tool. Paste in any sentence and it shows you exactly how the text splits into coloured chunks, with a live token count. Two minutes with it teaches you more than any definition.

Quick tip: if an AI chat starts forgetting details or giving weaker answers, your conversation has probably outgrown its context window. Start a new chat and paste in only the information that still matters.

Common Questions

How many words is 1,000 tokens?

Roughly 750 words in English, using OpenAI’s rule of thumb. The exact number depends on your language and word choices, since longer and rarer words split into more tokens.

Do spaces and punctuation count as tokens?

Yes. Spaces usually attach to the word that follows them, and punctuation marks often become their own tokens. Everything you type contributes to the token count.

Why do AI plans talk about tokens instead of messages?

Because messages vary hugely in size. A short question and a pasted 50-page report are both “one message,” but the report takes far more computing work. Counting tokens is the fair way to measure that work.

Final takeaway

Tokens are the small chunks of text every AI model actually reads and writes. They explain the limits on your chats, the reason long conversations drift, and the way AI pricing works. You never have to count them by hand. But once you know they exist, AI tools stop feeling mysterious and start feeling like something you can plan around.