What are tokens in AI, in simple terms?
What are tokens in AI? Tokens are the small pieces of text that an AI reads and writes. A token is usually a short word or part of a longer word. Tokens matter to a business because AI use is measured, limited and billed in them, much like electricity is billed in units.
Why AI counts in tokens, not words
An AI does not read whole sentences the way we do. Before it starts, your text is cut into small pieces, and each piece is a token. The answer is produced the same way, piece by piece.
Short, common words are one token each. Longer or rarer words are cut into two or three. Punctuation marks and numbers count too.
How much text is a token?
For English there is a handy rule of thumb: 100 tokens are about 75 words. A full page of text is roughly 500 to 700 tokens.
The rule is only rough. Other languages often need more tokens for the same meaning. Tables, long numbers and web addresses also use more than plain sentences.
Reading and writing are both counted
- Input tokens are everything the AI reads: your question, any document you pass along and, in a chat, the whole conversation so far.
- Output tokens are everything the AI writes back.
Providers usually charge more for output than for input. A long answer also takes longer to arrive, simply because there are more tokens to write.
A worked example: summing up a contract
Suppose you ask an AI to sum up a 3,000-word supplier contract in 150 words.
| Part of the request | Words | Tokens, roughly |
|---|---|---|
| Your question | 20 | 30 |
| The contract | 3,000 | 4,000 |
| The summary it writes | 150 | 200 |
| Total | about 4,230 |
Two things stand out. The contract makes up almost all of the total. What you send matters far more than how you word the question. And if you then ask "Which clause covers late delivery?" in the same chat, the contract is read again. The second question costs about as much as the first.
Where tokens are used without you noticing
- Long chats. Every new message carries all the earlier ones with it.
- Whole files for a small question. Forty pages are sent to find one date.
- AI that works in several steps. Each step reads the results of the steps before it.
Four ways to use fewer tokens
- Send the part that matters, not the whole document.
- Start a new chat when the subject changes.
- Ask for the length you need. "Three bullet points" costs less than an open question.
- Do not use AI for steps that a simple fixed rule can handle.
Do you have to count tokens yourself?
No. It is enough to know what drives the count: how much the AI has to read and how much it has to write. When you compare AI products, ask how they keep both small.
How Worknano handles this
Worknano is built so that the AI reads little and reads the right thing. The AI Assistant and the agents use semantic search, a search that finds text by its meaning. Instead of reading whole documents, they read only the passages that matter and answer from those. Making your content searchable and searching it does not use up your AI credits. Each agent has its own model setting, so short checks can run on a fast model. The plan itself is priced per company, never per user, so adding teammates does not change the bill.
- Which AI model does Worknano use?
- How pricing works in early access
- Semantic search: only the passages that matter
Create a free workspace and ask a question about your own documents.