How Much Is One Million AI Tokens?

Business owners often hear that AI costs are based on “tokens.”

That sounds technical.

Here is the simple version.

A token is a small piece of text. In everyday English, one token is roughly three-quarters of a word.

So one million tokens is about 750,000 words.

That is roughly:

• 7 to 10 average business books or novels
• About 1,500 printed pages
• Around 5,000 normal business emails
• More than 1,000 typical back-and-forth AI conversations

That is a lot more text than most people imagine.

Tokens are how AI measures text

When you type a question into an AI system, the system does not simply see words the way we do.

It breaks the text into smaller pieces called tokens.

The words you send count as tokens. The instructions behind the scenes count as tokens. The documents or business context sent with your question count as tokens. The answer you receive back also counts as tokens.

That is why token cost includes both sides of the conversation:

• the information going in
• the answer coming out

This does not mean business owners need to become token experts. But it does help to understand the scale.

One million tokens sounds abstract. About 1,500 pages feels more real.

The bigger lesson is not the price

Many people focus first on the cost per million tokens.

That is understandable. Cost matters.

But for most small and midsize businesses, the more important question is:

What information is the AI being given, and why?

Imagine your business has years of useful knowledge spread across:

• SOPs
• proposals
• policies
• meeting notes
• customer FAQs
• product information
• project files
• employee handbooks

All of that information could easily add up to millions of tokens.

But a good AI system should not dump all of that into every question.

If someone asks, “What should we include in a proposal for this service?” the AI probably does not need the employee handbook, every past meeting note, and every policy document.

It needs the few pieces of information that are actually relevant.

Good AI systems send the right context

This is where design matters.

A thoughtfully designed AI workflow retrieves the right information before asking the AI to answer.

For one question, that might mean a few paragraphs from an SOP.

For another, it might mean a customer profile, a proposal template, and notes from a recent meeting.

For a more complex question, it might mean several related documents.

The goal is not to send the AI everything the business knows.

The goal is to send enough relevant context for the AI to be useful without burying it in noise.

That matters for three reasons.

First, it can reduce cost.

Second, it can make the system faster.

Third, it can improve the quality of the answer.

More context is not always better. The right context is better.

A simple business example

Suppose an employee asks an AI assistant:

“How do we handle a customer who wants to cancel before their contract term ends?”

A poorly designed system might search too broadly or send a large pile of unrelated company documents.

That uses more tokens. It may also confuse the answer.

A better system would look for the relevant contract policy, customer service guidance, escalation rules, and perhaps a standard response template.

Then the AI can help draft a useful answer while a person still reviews the final response.

That is practical AI.

Not magic. Not hype. Not replacing judgment.

Just the right information, at the right time, for the right business question.

Why this matters for AI cost

AI APIs are often less expensive than people expect.

For example, if a model costs $2 per million input tokens, then processing one million input tokens costs $2.

A short request using a few thousand tokens may cost less than a penny.

A longer request with more business context may cost a few cents.

The exact price depends on the model, the provider, the amount of input, and the length of the output.

But the principle is simple:

Token cost is not only about the model you choose. It is also about how carefully the workflow is designed.

A careless system can send too much information, cost more, run slower, and still miss the point.

A mindful system sends what matters.

The takeaway

One million tokens is roughly 750,000 words, or about 1,500 printed pages.

That is the size of several books.

But most business questions do not need several books of information.

They need the few pages, paragraphs, or records that actually matter.

That is the difference between using AI as a toy and using AI as a practical business tool.

The value is not in giving AI more information.

The value is in giving it the right information, for a question that is actually worth answering.