GEO Glossary
Training data
Training data is the text a language model learned its knowledge from. The model doesn’t know what happened after the training cutoff without web search.
At a glance
TRAINING DATA learned up to the cutoff date slow to influence, over months and years LIVE SEARCH retrieved at the moment of the question quick to influence, through current content
Example
If you ask a model without web search about a brand that has only existed for six months, it doesn’t know it. A brand that has been written about for years in trade articles, forums and lists, on the other hand, it often knows even without search.
In practice
What is written about you online today ends up in future training data. Mentions in trade media, best-of lists, forums and interviews therefore work twice: now through search and later through model knowledge.
How it fits together
From the training data an LLM learns which entities exist. If something is missing there, only Grounding via a live search helps.