A legal professional's handbook for reliable AI tools

Authoritative content forms the foundation of reliable GenAI systems. Heagerty identified multiple dimensions of authority in professional-grade GenAI.

"If you're looking at documents, there’s the user's uploaded content to consider — is the professional-grade GenAI tool grounding its answers in that authority? Are the answers that it’s providing based on and verifiable in that authority? Then you get into the issue of case law, where that content’s authority is given by the courts."

For legal professionals, the distinction between training GenAI on content versus giving it access to content is critical. Professional-grade GenAI tools such as CoCounsel use retrieval-augmented generation (RAG) to find relevant legal content from a limited and defined set of verified content, rather than relying on what it "remembers" from its training — which involves nearly all the information on the internet, accurate or not.

The gold standard for authoritative content in professional-grade GenAI includes:

• Primary legal authority — statutes, cases, regulations
• Secondary sources created by subject matter experts
• User-provided documents — briefs, contracts, memoranda
• Editorial enhancements from legal publishers — headnotes, annotations, practice guides

This multilayered approach to authority helps ensure GenAI responses are grounded in specific legal sources rather than in the model's general approximation of legal concepts.

The internet offers abundant information, but with varying degrees of reliability. As Heagerty points out, "If you go to Google, you'll find content for sure, but you won't know where it's coming from. It could be from myfunblog.com or it could be from a trusted news source like Reuters, or anywhere between."

Professional-grade legal GenAI, on the other hand, should only use trusted content with established authority in the field.

"If you're looking at content available to you within a professional-grade solution such as CoCounsel, it should come with the trustworthiness of a specific, vetted source of content — in the case of CoCounsel that’s Westlaw or Practical Law, or really all of Thomson Reuters. Professional-grade solution users should be able to rely on what's being provided as true and not have to guess where it came from and whether it’s true."

For legal professionals, this distinction is crucial. While consumer GenAI tools can provide plausible-sounding answers, they often lack:

• Citation to primary authority
• Verification against current legal precedent
• Recognition of jurisdictional differences
• Updates reflecting recent statutory changes

The American Bar Association has already documented cases where attorneys faced sanctions for submitting work product based on hallucinated case citations from consumer-grade GenAI platforms. Professional-grade GenAI legal solutions mitigate this risk by grounding all responses in verifiable legal authority.

Heagerty highlighted several higher-level key differences between professional-grade GenAI and consumer tools.

"Professional-grade GenAI skills should work within well-defined guardrails that are more extensive than you would get with a general-use chat assistant," Heagerty explained. "By kicking the work users want to have done to one of these skills, it should be able to do more discrete tasks, more repeatable tasks. With CoCounsel, we're therefore able to test the skills and make sure we're getting quality output that’s as free of hallucinations as possible. 

“We do that every day. We run CoCounsel skills through hundreds and thousands of tests to make sure the responses we're getting are consistent and accurate and grounded in the content provided."

Skills, in the context of GenAI, are the diverse range of capabilities the technology can accomplish, such as document automation, drafting, and research.

Unlike general-use GenAI tools, professional-grade solutions such as CoCounsel should:

• Use specialized skills with discrete, tested functions.
• Ground responses in authoritative Thomson Reuters content.
• Undergo rigorous testing against attorney-created benchmarks.
• Make it easy to verify sources for GenAI-generated results.

For legal professionals, these skills include document comparison, researching the law within Westlaw's databases, and drafting clauses using language from Practical Law. Unlike general-use GenAI, these specialized skills are designed specifically for legal workflows, utilize authoritative legal content, and are tested rigorously to meet the standards legal professionals require.

"None of the foundational LLMs that power these GenAI solutions does that," Heagerty noted. "They don't have access to the expert-created content, for instance, that Thomson Reuters does. The result is a level of authority that just isn’t available outside a professional-grade solution."

For legal professionals, these differences translate to greater reliability, reduced risk, and more efficient workflows. While general-use GenAI can provide a starting point for simple work tasks such as writing emails or summarizing meeting notes, only professional-grade GenAI tools offer the verification pathways necessary for legal work product.

Hear how attorneys are using GenAI to streamline their workflows and focus on clients.

Join the webinar

Professional-grade GenAI is a tool for human experts to work more effectively, and it only works well when human experts design and test it. Heagerty described his team’s comprehensive approach.

"We put together quite voluminous suites of tests where we give the GenAI different tasks, research-based or analysis, and repeat those tests. Then our subject matter experts — trained licensed attorneys — create ideal responses. We then run the skill and see whether the GenAI’s output matches the ideal response." 

This rigorous testing regime includes:

• Subject matter experts creating benchmark responses.
• Continuous iteration until consistently accurate results are achieved.
• Nightly automated testing to ensure ongoing reliability.
• Immediate investigation of any customer-reported inaccuracies.

This degree of rigor is vital when creating a solution that legal professionals can rely on. Of course, just as any attorney would check the work of a junior associate, any attorney using GenAI should also review the results. While this testing regime keeps the incidence of errors astonishingly low, it cannot eliminate the possibility entirely. This is all part of using GenAI wisely.

Like most complex systems, CoCounsel follows the 80-20 rule, where achieving the final 20% of quality requires disproportionate effort. Heagerty explains.

"Getting that first 80%, I don’t want to say it's easy, but it's easier for sure. But that last 20% is what we really do feel is most important in building skills. We want to do everything we can to get as close as possible to getting the right answer every time.”

This pursuit of excellence is particularly important in professional-grade GenAI, where edge cases often involve novel legal questions with significant implications. The challenges include:

• Ensuring consistent performance across varied legal scenarios.
• Addressing complex legal queries with multiple relevant precedents.
• Maintaining accuracy when legal authority is in conflict.
• Updating systems as new legal developments emerge.

Legal professionals should recognize that no GenAI system will be perfect, but professional-grade tools invest heavily in minimizing errors, especially in high-stakes legal contexts.

Heagerty made a crucial distinction regarding GenAI and expertise. "There's a difference between training the model on the content and giving the model access to the content."

CoCounsel's approach illustrates this difference. "The models CoCounsel uses are not specifically trained on Thomson Reuters content, but they do have access to the content, so its answers are grounded in that authoritative data source. But CoCounsel isn’t relying on what the model remembers of the content way back when it was being trained. It finds the content, reads it, and then uses that in the specific response it gives."

This brings up an important point for legal professionals — GenAI augments legal expertise by providing access to relevant information but does not replace the need for human legal professionals. Of course, that augmentation is incredibly powerful and has transformational impacts, including:

• Processing vast amounts of legal information quickly.
• Surfacing relevant precedents that might be overlooked.
• Drafting preliminary responses to routine legal questions.

The critical aspects of legal practice remain firmly in the domain of human attorneys, such as: 

• Strategic decision-making in litigation, transactional and advisory work.
• Understanding client objectives beyond stated facts.
• Ethical judgment and professional responsibility.
• Contextual application of legal principles to novel situations.
• Persuasive advocacy before courts and tribunals.

The most effective approach combines professional expertise with GenAI assistance, using each for its comparative advantages.

The concept of agentic AI has begun making its way into the legal industry. Agentic AI uses GenAI to create autonomous agents with specific goals, allowing them to work independently with minimal human oversight and check their own work before giving it back to the legal professional for final review.

As AI evolves toward being able to handle more complex workflows and be incorporated with agentic systems, Heagerty sees authoritative content becoming even more critical. 

"With agentic AI, you're getting into workflows, and whether that's the agents themselves stringing skills together, or it's more prescribed, you create more points of failure, and so the more we're grounded in authoritative content, the more we're protected against any deviations there."

For legal professionals, this trend can lead to developments including: 

• More sophisticated document automation grounded in authoritative legal content.
• AI assistance in drafting litigation or transactional documents based on verified legal authority.
• Automated discovery tools that reliably ground responses in primary sources.
• Workflow automation that connects discrete legal tasks while maintaining authority.

As these tools evolve, the distinction between professional-grade and consumer AI will likely grow, with legal-specific tools increasingly focused on authoritative content as their competitive advantage.

The future developments planned at Thomson Reuters are bright, with a distinct focus on more complex workflows in litigation and transactional work. 

"Right now, we're looking at getting more into complicated workflows, and part of what we're looking at is drafting for litigation. Creating a complaint to file with the court usually begins with documents your client has brought you, from which you identify potential causes of action to develop a fact pattern.”  

“Without access to authoritative content, such as we have at Thomson Reuters, both from primary authority like courts or statutes, as well as from our attorney-editors’ reviews, I just don't think it’d be possible to get a high-quality final complaint."

For legal professionals, these developments promise to transform how routine legal work is accomplished with:

• Automated first drafts of complaints based on client facts.
• Discovery request generation grounded in relevant legal standards.
• Response drafting that incorporates applicable objections and precedents.
• Fact pattern analysis that identifies relevant legal issues.

As AI continues to evolve, the distinction between consumer-grade and professional-grade tools increasingly centers on authoritative content. For legal professionals, choosing AI tools incorporating verifiable legal authority isn't just about getting better answers — it's about meeting ethical obligations and professional standards.

The future of AI for legal lies not in replacing attorney judgment but in augmenting it with tools. These tools can quickly access, analyze, and synthesize the vast corpus of legal authority that underlies our legal system. By grounding AI responses in this authority, professional-grade GenAI tools like CoCounsel help ensure technology enhances rather than undermines the quality of legal services.