What Is a Token in AI, and Why Do Tools Charge by It?
By Stacey Tallitsch | October 23, 2026
You are looking at the pricing page for an AI tool, and it says something like "$0.01 per 1,000 tokens." You stare at it. What is a token? How many tokens does a normal email use? Is this going to cost you five dollars a month or five hundred? The page does not say, because it assumes you already know.
You do not need to feel behind. The word "token" is one of those terms the AI world uses constantly and explains almost never. So let me give you the plain version. Once you understand what is a token in AI, those pricing pages stop being a guessing game, and you can actually tell what a tool will cost you.
The short version: a token is a chunk of text
A token is a small piece of text. Most of the time it is about three quarters of a word. Some short words are one token. Longer words get split into two or three tokens. Spaces and punctuation can count too.
Here is the only math you need. Roughly 100 tokens is about 75 words. So a short customer email of around 150 words is about 200 tokens. A long page of text, say 750 words, is about 1,000 tokens. OpenAI, the company behind ChatGPT, lays this out in a short help article on what tokens are, and the rule of thumb holds across most tools.
Why do AI tools work in tokens instead of words? Because this is how the tool actually reads. An AI language tool does not see whole sentences the way you do. It breaks your text into these little chunks and reads them one at a time. If you want the fuller picture of how these tools work under the hood, I covered it in what a large language model actually is. For today, you only need the one idea: tokens are the chunks, and chunks are how the tool counts.
Think of it like a taxi meter
The cleanest way to picture token pricing is a taxi meter. When you get in a cab, the meter runs based on distance. You do not pay a flat fee per trip. You pay for how far you go.
Token pricing works the same way. The meter runs on text. The more text involved, the more it costs. A quick question costs very little. A long back and forth, or asking the tool to read a big document, costs more because more text ran through the meter.
And here is a part that surprises people. Both sides of the conversation count. The text you type in counts, and the text the tool writes back counts too. So if you ask a short question and the tool writes a long three page answer, most of your cost came from the answer, not the question. The meter runs in both directions.
It also helps to know that each token is cheap. On most usage based tools, you are paying a tiny fraction of a penny per token. So even a long request, with a big document going in and a full page coming back, often costs a few cents, not a few dollars. The numbers only add up when you are running hundreds or thousands of these requests, which is exactly what happens when AI is built into your business and handling work all day. One owner tapping away at a chat window is rarely the thing that runs up a bill.
Why your "simple" request sometimes costs more than you think
Say you run a small accounting practice. You paste in a long client email full of questions, plus three prior emails for background, and you ask the AI to draft a reply. That is a lot of text going in. Then it writes a detailed reply, which is a lot of text coming out. All of that is tokens, and all of it runs the meter.
Now compare that to a quick ask like "write me a one line thank you text for a client." Tiny input, tiny output, almost no cost.
This is not a reason to be stingy. It is just good to know where the money goes. When you keep your requests clear and only paste in what the tool actually needs, you get better answers and a smaller bill at the same time. That is one more reason it pays to understand how to write a good prompt, since a tight, clear request uses fewer tokens than a rambling one.
The part that trips up most owners: two different ways you pay
This is where real confusion sets in, so let me clear it up. There are two common ways you pay for AI, and they are not the same thing.
The first is a flat monthly plan. You pay one set price per month, like a gym membership, and you use the tool through its website or app as much as you want within fair limits. ChatGPT, Claude, and Gemini all offer plans like this. If you are on one of these, you are mostly not paying per token at all. You pay your monthly price and you are done. For day to day use by one owner, this is usually the simplest path.
The second is usage based pricing, where you pay per token. This is how the "behind the scenes" versions of these tools are sold, the kind a software company or a contractor hooks into your systems. If you hire someone to build you an AI phone answering tool, or wire AI into your booking software, the bill underneath it is very often per token. That is when the taxi meter is really running on your dime.
So when you see "per token" pricing, ask one question. Is this a flat monthly tool I just log into, or is this something running in the background of my business that charges by usage? The answer tells you whether tokens are something you need to watch closely or something you can mostly ignore.
When this actually matters for your business
If you are a one person operation using a chat tool through its website on a flat monthly plan, you can mostly forget tokens exist. You pay your monthly fee, you use the tool, that is the whole story. Do not let the jargon scare you off a tool that would help you.
Tokens start to matter in two situations. The first is when you move from a flat plan to anything that bills by usage, because now the length of every request affects the bill. The second is when you are deciding whether to build or buy an AI tool for your business, and a vendor quotes you token pricing. Knowing roughly how many tokens your real work uses lets you sanity check their numbers instead of taking them on faith. If you are weighing one of those decisions, it connects directly to the bigger question of whether AI will really save your business money.
The main thing to carry away is this. A token is just a chunk of text, about three quarters of a word. Some tools charge a flat monthly fee and you never think about tokens. Other tools charge by the token, and then the rule is simple: more text, more cost, on both the question and the answer. That is the whole idea. The pricing page was never as complicated as it looked. It just forgot to explain itself.
-- 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