Asking a question
Ask prepares your data for smart search by organizing it in a modern system built for speed and context.
When you type a question, Ask:
Interprets your query to understand intent using semantic expansion.
Searches across your chosen range of documents for the closest matches in meaning.
Orders results by relevance, showing the strongest matches first.
Summarizes and creates an easy-to-read narrative from the most relevant documents.
You can also guide the response with extra instructions—for example, asking Ask to build a Boolean search from the top results or to return the answer in another language.
Semantic similarity
Semantic similarity is about matching text by meaning rather than exact words. Two sentences can look completely different on the surface but say the same thing:
The vehicle collided with a stationary object.
The car crashed into something that wasn't moving.
These sentences use different words but mean the same thing — that's high semantic similarity. Ask uses this approach because legal searches are about concepts, not keywords. Searching for "terminations" in a wrongful termination case should also pull up documents that say "firing," "layoffs," or "pink slip."
Important
Remember: a semantic search result means “conceptually nearby” not "guilty."
Similarity score
Ask’s similarity score is a number between 0–100% that tells you how closely a document matches your query. Ask uses this score to rank results and surface references, then group documents into high, medium, and low tiers so you can focus on the most relevant material first. For more information, see How to Use Ask.
Similarity score measures closeness, not relevancy. A human reviewer's job is to look at that 35% score and say: "Yes, I can see why the model flagged this and no, it doesn't actually help my case." Some documents will be close enough for the model to flag but not close enough to matter. That's not a mistake — it's the model casting a wide net. You're the one who decides what's worth keeping.