Ten Questions We’re Asked About Lighthouse AI Search (and Our Answers)

Summary: Amid unprecedented data volumes and complexity, Lighthouse AI Search is helping legal teams move from million-doc datasets to defensible insights in minutes, not hours. In this post, we answer the 10 most common questions about how Lighthouse AI Search works, how it keeps your data secure, and how it’s redefining what’s possible in eDiscovery.

‍

We recently hosted the webinar, The Evolution of Answers: Introducing Lighthouse AI Search, where we showed how teams can move from million-doc datasets to grounded answers in a matter of minutes, with a high degree of accuracy, and without Boolean gymnastics. Along with a live demo of AI Search, our expert presenters led a lively round of Q&A—and here, we’re sharing 10 of those audience questions about AI Search functionality, accuracy, and security, plus presenter responses.
‍

Question 1: How long does a search take in AI Search?

On average, we see two to three minutes. Results progress through three phases—matching documents to your question, organizing those matches into answer groups, and generating summaries—so you can watch progress in real time and review results as they land.
‍

Question 2: Can we run multiple searches at once in AI Search?

Yes. There’s effectively no practical cap; teams can run large numbers of concurrent questions (into the thousands) and then review as results complete.
‍

Question 3: Is data in AI Search safe (and kept private)?

Yes. AI Search operates in a closed environment and only references the dataset you load for that matter. In accordance with guardrails we’ve put in place, it does not search the open internet, nor does it share data across cases. If an answer isn’t in your dataset, it won’t go elsewhere.
‍

Question 4: How do you minimize hallucinations in AI Search?

Every answer is grounded in your indexed documents and returned with exemplar source documents and highlights. As related to question three, AI Search is confined to sourcing answers from your document set, so the risk of hallucination is very low. You can click through and verify the supporting evidence immediately, keeping a human-in-the-loop for defensibility.
‍

Question 5: Can we scope a search to a subset (e.g., board-level comms) in AI Search?

Scoped searching is on the roadmap and launching soon. Today, you can target subsets by how you phrase the question (e.g., “What was the board discussing about X?”) to focus answers on specific people, timeframes, or topics.
‍

Question 6: Do “bad” questions waste time or budget in AI Search?

No. We don’t charge per search, so you can iterate freely. There are no “bad” questions because even those that don’t yield fruitful answers can give you clues about where you need to go next in your search. If a question doesn’t land, rephrase and try again—or lean on our expert search consultants to help craft prompts for your use case.
‍

Question 7: Will AI Search return every relevant document?

AI Search is a question-and-answer tool—not a linear review engine. It returns synthesized answers with exemplar documents that support them. From there, you can move items into downstream Relativity workflows (batching, review, seed sets for TAR, adding families/threads/near-dupes) as needed.
‍

Question 8: Does AI Search only work in Relativity?

Today, yes. AI Search is integrated with Relativity Server, including coding panels and one-click access to the Relativity viewer. Integration with RelOne is in development.
‍

Question 9: Is there lead time beyond loading data in AI Search?

A little. After loading data to a Relativity workspace, select the saved search you want indexed and build the AI Search index. It’s quick. We recommend standard reductions (dedupe, email threading, junk removal) but not keyword culling, since culling can hide important concepts you don’t yet know to look for.
‍

Question 10: Can we export results easily from AI Search?

Yes. While AI Search returns exemplars (not “all docs”), you can mass-code those documents and push them into your Relativity workflows for export, batching to reviewers, or using them as seed sets, then expand to related families, threads, and near-dupes.

Watch the full webinar, including the business case for AI Search and comprehensive demo, and learn more about Lighthouse AI Search. Have a question that isn’t covered here? Contact us and drop your question in the message, and we’re happy to get back to you directly.