Grounding

Grounding is making an AI model base its answer on specific source material, such as the documents in a matter, and cite that material, instead of answering from its general training.

Last updated

What it means for a UK law firm

Grounding changes what a solicitor has to check. An ungrounded answer has to be verified from scratch; a grounded answer points to the passage it relies on, so the reviewer can confirm the point by reading the source.

Grounding reduces AI hallucinations but does not remove them. A grounded model can still misread a passage, or cite a passage that does not support the point made. A well-grounded system also says when its sources do not answer the question, instead of filling the gap from general knowledge.

Example

Illustrative scenario

A solicitor asks whether a share purchase agreement requires a counterparty's consent on a change of control. A grounded answer quotes the relevant clause and links to it. The solicitor opens the clause, reads the definitions it relies on and confirms the answer.

How Chancery handles it

In Chancery, every answer about documents cites the passage it relies on, and says so when the documents do not answer the question. Questions can be asked of one matter, all of your matters, your precedent library, or a matter and your precedents together; see Ask your matter.

Related terms

This glossary entry is general information, not legal advice. See the full legal AI glossary for other terms.

Give your firm the same advantage as the top 20, at a price that makes sense for yours.