Grounding
Connecting a generated answer to real, checkable information.
Technical Term
Full explanation
Grounding is how an AI system connects a generated response to external, verifiable information, rather than relying only on patterns learned during training.
The distinction is between two different sources an answer can draw from: a model's training data, frozen at whatever point training ended, and live, external information retrieved at the moment of the query, current search results, a connected database, a specific set of documents. Grounding is the mechanism that connects the answer to the second kind.
This is directly why current, well-structured content matters, separate from whether a brand was ever part of a model's original training data. A system with retrieval-based grounding can find and cite a page published yesterday; a system relying only on frozen training knowledge cannot, no matter how well the page is written.
Why this matters
Grounded answers draw on sources that exist now, which is what makes current, well-structured published evidence worth building.
Example
A model with no grounding, asked about a company's current pricing, might answer from outdated training data or decline to guess. The same question asked through a system that grounds its answer with a live search first can retrieve and cite the company's current pricing page directly, even if that page was published after the model's training ended.
Where this sits
Grounding belongs to Retrieval, Evidence & Consensus, one of the seven layers of the Aingworth Glossary. Information exists, then it is retrieved, then sources are selected, then they corroborate one another, then consensus emerges. This is where Aingworth's Consensus Over Rankings thesis lives.
Continue with: You Are Using AI in the Wrong Place.
Related terms
Ready to see where your brand stands?
Start with the free diagnostic to see which condition of the Recognition Triangle is leaking, or read the frameworks these terms belong to.