Swipe to see each slide.
The short answer
A token is a small piece of text, such as a word, part of a word or a punctuation mark, that an AI model breaks your writing into before it processes it. Tokens matter because many AI services charge by the number sent and generated, and a model can only handle a limited number at once.
Three things to remember
- 01Different models can split the same text differently, so token counts vary.
- 02Input tokens are what you send, including instructions and documents, and output tokens are what the model writes back.
- 03Long chat histories, big pasted documents and long answers use tokens fast, so send only what is relevant.
Every slide, in words
- 01
What is a token?
The small pieces of text AI models actually process
AI models do not read text exactly the way people do. They break text into smaller units called tokens.
Example
"The customer needs a refund." might become pieces like:
The | customer | needs | a | refund | .
A token may be a whole word, part of a word, punctuation, or another text unit
Different models and tokenizers can split the same text differently
Tokens are the units models count, process, and often bill for.
- 02
Why do tokens matter?
Tokens affect cost, speed, and how much text a model can handle
Cost: Many AI APIs charge based on the number of input and output tokens.
Context: A model can only process a limited number of tokens at once.
Speed: More tokens usually mean more work for the model to process.
Output: Long answers consume more output tokens and can cost more.
Token count is one of the basic measurements behind AI usage.
- 03
Input tokens vs. output tokens
The model counts what you send and what it generates
Input tokens
What goes into the model:
Your prompt
System instructions
Conversation history
Retrieved documents
Tool results or other context
Output tokens
What the model generates:
The answer
A summary
Code
Structured JSON
Tool-call instructions
Total usage is often the combination of input tokens plus output tokens.
- 04
What uses up tokens fast?
Long context can grow quickly, even before the model answers
Long conversation histories carried into every new request
Large PDFs, manuals, or documents pasted into the prompt
RAG systems that retrieve too many chunks
Verbose system instructions and duplicated context
Very long generated answers
Tool outputs, logs, or data that are sent back into the model
More context is not always better context.
Efficient AI systems try to send the model the right information, not every piece of information.
- 05
The small-business takeaway
Tokens are why AI usage is not really "free text"
You do not need to count tokens by hand. But you should understand what drives them.
Shorter, focused prompts can reduce unnecessary usage
RAG should retrieve only the context that is actually relevant
Huge conversation histories can increase cost and latency
Model pricing may treat input and output tokens differently
Token limits are one reason context-window size matters
Tokens = the basic unit of AI text processing.
Once you understand tokens, context windows and AI pricing make a lot more sense.
More in part 2
- 1What is generative AI?Written
- 2What is a foundation model?
- 3ChatGPT, Claude, Gemini and Copilot: what's the difference?Written
- 4Free or paid AI: what do you actually get?Written
- 5What is a token?You are here
- 6What is a context window?
- 7What is inference?
- 8What is temperature?
- 9What is a system prompt?
- 10What is prompt engineering?
- 11What does a good prompt look like? Ten you can copyWritten
- 12How do you use AI for email and everyday writing?Written
- 13How do you use AI to summarise documents and meetings?Written
- 14How do you research with AI without getting burned?Written
- 15What are custom instructions and projects, and why set them up?Written
- 16What is an AI hallucination?
- 17What is multimodal AI?
- 18What are cloud AI platforms?
- 19How does AI actually learn?
- 20What is chain of thought?
- 21Why does thinking longer make it smarter?
- 22What is reasoning effort?
Next step
Where does this fit your business?
Six quick questions about your business, then your AI plan: what to start with, what a person still checks, and the lessons that go with it. Or bring it straight to a free thirty-minute call.