Introducing Everlaw AI Deep Dive

When a legal matter starts, smart litigators move quickly.

There’s an onrush of discoverable data, millions of documents, dozens of data types, and countless hours of human review. Without the right tools, critical insights can stay buried.

In the past, legal teams have conquered this deluge one document at a time, bit by bit, hunting for the key pieces of evidence to start building an argument around. It’s expensive, time-consuming, and fiercely entrenched in the status quo.

But what if, instead of sifting through documents to find the answer, the answer was clear from the very beginning?

The newest addition to the Everlaw^AI Assistant suite of generative AI tools, Deep Dive (formerly code-named Project Query), lets legal professionals turn the case preparation process on its head, enabling users to start with the answer, and then verify how the model got there. This tool enables partners and senior attorneys to engage with the data directly, earlier in the discovery process. Rather than waiting for first-level review to surface insights, partners can now leverage Deep Dive to immediately start asking questions and building case strategy.

Everlaw AI Deep Dive was built to recognize when it lacks sufficient information to accurately answer a question in order to avoid the risk of generating misleading answers, and only responds when relevant content is found. Every answer is backed by direct citations from the document corpus for easy verification and further investigation.

During our five month long closed beta program, we worked closely with a select group of customers to learn how they’re using Deep Dive, and incorporated their feedback to make it even better.

Now, with the end of our closed beta program, we’re excited to announce that Deep Dive is now available in an open beta. This will allow even more users to experience and provide feedback as we move closer to a full release.

Deep Dive Overview

With the next evolution of generative AI tools created for legal professionals, Everlaw wanted to take advantage of a powerful new implementation of this technology.

Large language models, or LLMs, are the basis for many of the generative AI systems we use today, from ChatGPT to Google Gemini and more.

The way these systems are able to learn is by predicting the next word in a sentence. Or, to put it more simply, filling in the blank.

Take, for example, the sentence “The sky is blue.” Here’s how you would feed this sentence into an LLM to help it understand where each word fits, so it’s more likely to guess correctly.

“___ sky is blue.”

“The ___ is blue.”

“The sky ___ blue.”

“The sky is ___.”

When the model predicts correctly, you let it know its guess was correct, and when the model predicts incorrectly, you update the values to ensure it has a higher chance of making the right choice next time. When this exercise is spread over hundreds of billions of times, the model becomes incredibly efficient at making the correct guess. 

And while LLMs are proficient at answering low-stakes questions, they can also be inconsistent, sometimes copying word-for-word straight from a webpage, other times making up facts, better known as hallucinating.

RAG is a framework that enhances LLMs by retrieving relevant information from a knowledge base before generating a response. It helps the model get more accurate answers by leveraging external sources of information to supplement the LLM’s internal knowledge. This ensures the LLM has the most up to date information, and allows users to review the sources the LLM used to generate its response.

However, when it comes to the law, RAG has been tricky to implement due to the complex nature of legal documents, which can often contain multiple different forms of data that are unstructured, and that need to be compared and contrasted with other data sets.

The release of Deep Dive solves this by applying RAG in a way that’s specifically built for the complexities of legal data. Legal teams can ask natural-language questions of their document corpus just as they would with a colleague, and quickly surface accurate, verifiable insights.

By leveraging the latest LLM reasoning models, Deep Dive is able to essentially mirror how humans think through a problem by breaking the issue down into smaller, more logical steps. Users can ask the model open-ended questions the same way they’d ask a colleague.

And because Deep Dive is grounded in the documents you instruct it to reference, its only responds when relevant content is found. Ask Deep Dive if the sky is blue and, if no documents refer to a blue sky, it simply say it cannot answer.

With Deep Dive, partners and attorneys can easily explore new matters as soon as they come in, identify key issues and entities within data sets, effectively prepare for depositions and trials and more. Here are some use cases in which Deep Dive can be especially helpful.

Early Matter Exploration

One of the places where Deep Dive excels is in exploring a data set at the beginning of a matter.

For example, if a firm receives a whistleblower complaint alleging financial misconduct by a client executive, they can leverage Deep Dive to quickly understand the most important facts of the case and surface key documents.

A question like “What did employees say about expense reporting?” can help identify sentiments during that time period, and bring relevant discussions to the forefront. With subsequent follow-up questions, the scope of review can be narrowed to begin formulating an argument.

Key Issue and Entity Identification

Identifying key issues and entities within a document set is a crucial step in figuring out the major players and exploring the substance of a case.

Once a firm is brought into a large antitrust matter, for example, there are years of corporate communications and regulatory filings to get through. With Deep Dive, they can ask questions to help get a comprehensive overview of the case.

They can ask “Who was involved in internal discussions about competitors?” and receive a list of documents that highlight recurring names and departments that otherwise would’ve taken hours of work to discover. This allows the case team to quickly identify the key players in the matter so they can know who to pay attention to going forward.

Deposition Preparation and Trial Support

Preparing for a deposition is a detailed and high-stakes process in which every eventuality needs to be planned for. Attorneys have to know the facts of the case inside and out and be ready to change course at a moment’s notice.

For example, when preparing a witness outline for an upcoming deposition involving disputed contract negotiations, Deep Dive can be used to skip the manual review process entirely. A user can ask “What does Jane Smith say about the contract terms?” and immediately receive relevant documents that include exactly what they’re looking for.

Built-In Safety and Reliability

As with all of Everlaw’s feature releases, time is taken to create tools that are safe and reliable for all legal professionals, no matter their technological expertise.

For Deep Dive specifically, Everlaw has implemented additional security measures to ensure that no client data is stored or retained by the LLM.

Everlaw has prioritized accuracy and honesty in Deep Dive’s output, building in structural safeguards to ensure the system admits when it doesn't have enough relevant information rather than trying to please users with potentially incorrect answers. Additionally, all responses include precise, citation-backed answers directly from the corpus, allowing for easy verification.

For example, during a review of documents in a breach of contract case, if an attorney asks Deep Dive, "Was there an agreed-upon cap on damages?", and no document in the corpus clearly addresses a damages cap, Deep Dive won’t invent an answer or suggest something that isn’t there. It’ll return a clear message indicating there’s not enough information to respond confidently—avoiding the risk of hallucinated claims.

Everlaw’s AI Philosophy

Everlaw’s approach has always been about building for the long term.

Everlaw^AI Assistant’s suite of generative AI tools were made to provide users with the utmost control, confidence, and privacy and security that customers have come to expect from Everlaw. It covers many aspects of the case preparation process, from review (finding important evidence that may never have been uncovered or take many more days and dollars to find) to case building (telling more powerful stories) and beyond.

Each feature is iterated on many times, refining the user experience to better provide value and avoid pitfalls. With each new release, Everlaw has stuck to its core generative AI principles, and kept respect for users at the forefront.

Next-Gen AI for the Legal Profession

Deep Dive offers an opportunity for legal professionals to put the days of endless searches and sifting through irrelevant data behind them, and instead focus on winning cases and getting the best outcomes for clients.

“Pinpointing facts in a vast corpus is gold and doing it in seconds is game-changing,” Steven Delaney, the Litigation Support Director, at Benesch, said. “Deep Dive feels more elegant than keyword search, delivering answers—not just documents—instantly. By getting straight to the facts, we can save time, prioritize strategy and reshape the discovery workflow.”

Ryan O’Leary, research director at IDC, echoed this sentiment, adding that its practicality will help make a difference for users.

“Everlaw continues to raise the bar in GenAI innovation with Deep Dive,” he said. “It’s one of the most practical AI applications I’ve seen in discovery—enabling investigation before review in seconds. With its ease of use, Deep Dive promises to instantly boost efficiency and deliver a significant impact for legal teams.”

In a profession where perfection is often the goal, Deep Dive offers a new path forward—where a simple question leads directly to the truth.

If you’re interested in learning more about Deep Dive or Everlaw^AI Assistant, join the Deep Dive beta or request a demo today!

Justin Smith is a Senior Content Marketing Manager at Everlaw. He focuses on the ways AI is transforming the practice of law, the future of ediscovery, and how legal teams are adapting to a rapidly changing industry.