What are tokens and why does AI hallucinate?
In short
AI systems process language not in words but in tokens, small text building blocks. They predict the most likely next token each time. That explains why AI sounds fluent yet can confidently claim false things, that is, hallucinate.
What a token is
A token is the smallest processing unit of a language model: often a word, sometimes just a syllable or a punctuation mark. The model breaks every text into such tokens and computes internally with numbers, not letters. The length of inputs and answers is also measured in tokens.
How AI generates language from it
At its core a language model does one thing: it predicts the most likely next token based on everything before. Token by token, fluent text emerges. The model does not understand content like a human, it computes what sounds plausible.
Why this turns into hallucinations
Because the goal is "plausible" and not "true", a model can fill gaps with convincing but false content. Such invented, linguistically flawless statements are called hallucinations. When reliable information is missing, the model guesses rather than admitting it does not know.
What helps against it
Hallucinations drop when the model relies on real sources. Through grounding, a live web search, it pulls in current, verifiable content. For brands that means: the clearer, more consistent and more findable your facts are online, the more likely AI represents you correctly.
What this means for your visibility
Hallucinations about your brand are a risk, but also a hint: where AI guesses, it lacks reliable sources about you. You can close exactly these gaps with consistent, evidenced content.
Key Takeaways
- Tokens are the small building blocks in which AI processes language.
- AI predicts the most likely next token, it does not check for truth.
- Hallucinations are plausible-sounding but false statements.
- Reliable, findable sources lower the risk and improve how you are represented.
Frequently asked questions
Are tokens the same as words?
Not quite. A token is often a word but can also be just a syllable or a punctuation mark.
Can hallucinations be switched off entirely?
No. But they can be reduced significantly by having the model rely on reliable sources.
What does this have to do with my brand?
If clear information about you is missing, AI is more likely to invent something. Consistent, evidenced content prevents false statements.
About the author
Christoph Schempershofe, founder of VISIBILIS. Since his studies he has combined marketing with technology: from websites and brand building through search engine marketing to AI visibility (GEO). As a lecturer at FOM and IU he teaches marketing, online and search engine marketing and content management systems.