What "Tokens" Mean and Why They Show Up on Your AI Bill
By Stacey Tallitsch | May 16, 2026
If you have looked at the pricing page for any AI tool in the last year, you have seen the word "tokens." OpenAI prices ChatGPT API access in tokens. Anthropic prices Claude in tokens. Google prices Gemini in tokens. The number is on your bill, but nobody at any of these companies stops to explain what a token actually is.
You have a business to run. You did not sign up to learn a new vocabulary. But tokens are the unit AI companies use to bill you. If you want to know what your AI tools are going to cost month to month, you need a clear picture of what one actually is.
This post explains what a token is in plain language, walks through an example using a real small business, and clears up the 3 things people get wrong about them.
A token is a piece of a word
Here is the short version. A token is a small chunk of text that AI software counts as one unit of work. It is not exactly a letter, and it is not exactly a word. It is something in between.
The simplest way to think about it: most short common words are 1 token. Longer or less common words get broken into 2 or 3 tokens. Punctuation marks count. Spaces sometimes count.
A rough rule that AI vendors themselves use: 1 token equals about 4 characters of English text, or roughly 75% of an average word. So 100 tokens is about 75 words. 1,000 tokens is about 750 words, or a short page of text.
The taxi meter is a useful comparison. When you take a taxi, the meter runs while the car is moving and while it is waiting. You pay by the unit, and the meter does not care whether your trip is to the dry cleaner or to the airport. It just counts. AI tokens work the same way. The software counts tokens, you pay per token, and the meter does not care what you are doing.
What this looks like for a real small business
Take a small landscaping business that uses ChatGPT to draft customer follow-up emails. The owner pastes in the customer's information and the recent job details, then asks the AI to draft a thank-you email and a follow-up reminder.
The pasted-in information is the "input." A typical input for this kind of email might be about 300 words, which is roughly 400 tokens.
The email the AI writes back is the "output." A friendly email of about 2 short paragraphs is maybe 150 words, or close to 200 tokens.
So one email costs about 600 tokens total. At current public pricing for a mid-tier AI model, 600 tokens is well under a tenth of a cent. If the business sends 100 follow-up emails a month, the AI portion of the bill is somewhere around 5 cents.
That math is the part nobody is showing you. The token vocabulary is intimidating, but the actual cost for typical small business workloads is almost always tiny. Most operators worry about runaway AI bills based on what they read about big company spending, and the worry rarely matches what their own usage looks like.
For an even cleaner explanation of what an AI system is doing when it processes your text, OpenAI's help center has a short article on tokens that is worth bookmarking if you want to verify the math yourself.
What people get wrong about tokens
3 things trip people up the first time they look at AI billing.
Tokens are not just words. Some short common words like "and," "the," and "of" are 1 token each. Longer or unusual words get split. The word "implementation" might be 2 or 3 tokens depending on the AI model. Made-up words, brand names, and unusual industry terms tend to break into more tokens than regular words do. This means a post written in plain English costs slightly fewer tokens than the same length post written in dense industry jargon.
Both your input and the AI's output count. When you paste in a long document and ask the AI a 1 sentence question, you pay for the long document going in plus the answer coming out. Output tokens are usually priced higher than input tokens, sometimes 3 or 4 times as much. So when you are looking at a pricing page, look at both numbers, not just the cheap one.
Long conversations stack tokens. This is the one that surprises people the most. When you have a long back and forth chat with ChatGPT or Claude, the AI does not actually "remember" the conversation. Each time you send a new message, the entire conversation history is sent back to the AI as part of your input. A 30 message conversation can have a much larger input cost on the 30th message than it did on the first one, even if your new message is short.
For most small business uses, single tasks like drafting an individual email or summarizing a single document, this stacking does not matter much. But if you build an AI agent that holds long conversations or processes many documents in a row, the token math compounds. Worth knowing before you commit to a tool that bills this way.
What "context window" means and why it matters
While you are looking at AI pricing pages, you will see another term that is closely related: "context window."
A context window is the maximum amount of text the AI can consider at one time. It is measured in tokens. If a model has a 128,000 token context window, that means roughly 96,000 words of text can be in front of the AI at once: your current input, the prior conversation, plus whatever document you pasted in.
For most small business uses, you will never hit the context window limit. But if you are using AI to process a long contract, a long meeting transcript, or a full year of customer communications, knowing the context window matters. Past that limit, the AI starts forgetting or summarizing instead of holding everything in working memory. The technology that makes this work is the same large language model architecture underneath every major AI assistant.
Why this matters for your business
3 practical things come from understanding tokens.
First, when you read an AI vendor's pricing page, you can now budget. Multiply your expected monthly usage in words by 1.3 to get tokens, then multiply by the per token price. The number you get is usually a fraction of what your gut estimate would have been.
Second, you can compare AI tools honestly. Some vendors price in tokens, some in monthly subscriptions, some in API calls. When you see token pricing, you now know what a token is, and you can convert it back to "how much for 100 emails" or "how much for processing a stack of invoices."
Third, you can recognize when an AI tool is going to be expensive before you commit to it. A tool that pulls in your full customer database every time you ask it a question is going to burn tokens fast. A tool that asks you 1 focused question at a time will not.
When tokens stop mattering
Once you understand what a token is, you can stop tracking them. Most small business AI usage will run well under monthly subscription tier limits. ChatGPT, Claude, and Gemini all offer subscription plans that fold token costs into a flat rate, and for the vast majority of operators that is the right plan. You only need to think about tokens directly if you are building something custom or hitting unusual usage levels.
But the next time you see "tokens" on a pricing page or in a vendor's documentation, you will know what they mean. That is the floor for being able to evaluate AI tools as a business owner. Not learning to code, not memorizing the technology, just knowing enough of the vocabulary to make decisions without being talked over.
-- Stacey | The Standalone
About the Author
Stacey Tallitsch runs The Standalone, an AI Implementation Diagnostic practice for small business owners. He has 30 years of experience in technology and has written 21 books on systems thinking and decision-making. More than 30,000 students have learned from his online courses.
- Stacey Tallitsch, The Standalone