Special Tokens You Never See
Part of the Fundamentals section of Coddy's AI Prompts journey — lesson 12 of 23.
Beyond your text and the system prompt, the AI uses special tokens — invisible markers that structure the conversation. They look something like <|user|>, <|assistant|>, or <|end|>, though the exact format varies by model.
These tokens mark where each message starts and ends, and who's speaking. Without them, the model wouldn't know whose turn it is. You never type them and you don't see them, but they consume space in the context window.
This is why even an empty conversation isn't zero tokens, and why token counts sometimes seem slightly higher than expected. In long back-and-forth conversations, these markers add up — each message boundary requires its own set.
You don't need to manage special tokens directly. But knowing they exist helps explain where those "extra" tokens are coming from.
Cheat sheet
AI conversations use special tokens — invisible markers that structure the conversation, such as <|user|>, <|assistant|>, or <|end|>.
These tokens mark message boundaries and identify speakers. They consume space in the context window, which is why:
- Empty conversations aren't zero tokens
- Token counts may seem higher than expected
- Long conversations accumulate overhead from message boundaries
You don't manage special tokens directly, but understanding them explains where "extra" tokens come from.
Try it yourself
This lesson doesn't include a code challenge.
Test yourselfThis lesson includes a short quiz. Start the lesson to answer it and track your progress.
All lessons in Fundamentals
2How AI Reads You — Tokens
What Are Tokens?Words ≠ TokensWhy Token Counts MatterThe Context WindowTokens in Other LanguagesSpecial Tokens You Never See3Writing Your First Prompts
AI Talks Markdown (By Default)Controlling the Output FormatControlling Tone and StyleWhen the AI Ignores You