---

 In 2025, after more than 35 years of using traditional legal research platforms, I made a major change in how I perform legal research. My practice involves deep, complex, and often theoretical applications of U.S. tax law, especially U.S. international tax matters. 

I moved my law office to my home and decided to eliminate an expensive online legal research service. The annual cost had reached approximately $8,500. Instead, I began using large language model platforms, principally ChatGPT and Gemini, at a combined cost of about $40 per month. 

The result surprised me. 

For my purposes, the traditional large-scale legal research platforms are no longer worth keeping. Their search model is stale and outdated. Their search capabilities are awful. Their content, once you can find it, is often excellent, and their citations are trustworthy. But the old research model has been overtaken by a faster, more flexible, and more analytically useful approach. 

1. Speed and Practical Usefulness 

With the traditional legal research platform, I often had to begin with Boolean searches, a relic developed in the 1960s that obtained widespread adoption in the 1970s, more than 50 years ago. Yet my high-priced legal research platform still relied on it at least through 2024. For the type of legal research I perform, I found Boolean searches virtually worthless. Finding the information in the platform could take 10 minutes or more and often required scrolling through pages of information to locate the right section. 

With ChatGPT or Gemini, I can describe the facts, identify the legal issues, and ask for an integrated legal analysis. In many cases, a comprehensive first response appears in under two minutes. Of course, that response is only the starting point, akin to a brainstorming session rather than a finished project, but it is incredibly useful. It is surprisingly good at finding hidden issues. 

 It gives me: 

1. A structured issue list; 
2. A summary of the governing legal principles; 
3. Potential authorities to verify; 
4. Arguments on both sides; 
5. Practical risks and consequences; and 
6. A roadmap for deeper research. 

2. Integrated Analysis 

Real legal problems rarely present themselves as isolated issues. A single client matter may involve contract law, fiduciary duties, tax consequences, procedural rules, evidentiary issues, limitations periods, damages, disclosure obligations, and litigation risk. Those issues often interact. 

The strength of LLM platforms is that they can process multiple issues together. They can analyze the fact pattern as a whole, identify hidden issues, present alternative theories, and explain how changing one fact may change the legal result. 

For example, I can ask an LLM to analyze: 

“Whether a foreign transfer should be treated as a gift, compensation, trust distribution, capital transaction, or reimbursement, and what reporting forms, penalty risks, and disclosure positions may follow from each characterization.” 

An integrated analysis is extremely helpful and eliminates time-consuming research that otherwise is often required with the old legal research methods. 

3. The Essential Warning: Verify Everything 

The major drawback is citation reliability. 

A lawyer must verify every citation. Statutes, regulations, cases, administrative guidance, procedural rules, quotations, and case holdings all must be checked. 

Sometimes an LLM will cite a case that is not on point. Sometimes it will describe a real case too broadly. Sometimes it will identify the correct legal principle but attach it to the wrong authority. Sometimes the “perfect” case simply does not exist. 

That is not a minor concern. It is by far the biggest drawback to using LLMs. For that reason, use ChatGPT and Gemini as powerful legal research assistants, challenge them with follow-up questions, and never, ever treat them as final authorities. They excel at issue spotting, organizing analysis, developing arguments, and identifying possible sources. But they are not lawyers, and they should never be considered anything more than researchers. 

The proper standard is simple: Treat the answers as a starting point, and never trust the citations without verification. 

4. Online Free Resources Fill Much of the Gap 

One reason the traditional paid services have become less necessary is that lawyers now have access to excellent online resources outside those platforms. 

For example, Google Scholar contains federal and state court decisions and is easy to search. It also has a very useful feature to check a case’s history and verify whether the case is still valid. There are other online platforms, including Justia, CourtListener, Cornell Law Information Institute, FindLaw, and Caselaw Access Project, available for double-checking and verifying LLM-generated research. 

I will address Google Scholar in a separate article because it deserves its own discussion. 

5. Better Instructions Produce Better Research 

The quality of the LLM answer depends heavily on the quality of the instruction. 

A vague prompt produces a vague answer. A detailed legal research instruction produces a much better result. For general legal research, I suggest using an instruction similar to this: 

General Legal Research Instruction for LLM Platforms 

Please provide a careful legal analysis, not merely a surface-level summary. Identify the governing legal issues, relevant statutes, regulations, cases, administrative guidance, procedural rules, and practical consequences. 

Analyze the facts in an integrated way. Do not treat each issue in isolation if the issues affect one another. Identify factual uncertainties and explain how different facts could change the result. 

Where appropriate, discuss: 

1. The strongest arguments supporting the client’s position; 
2. The strongest opposing arguments; 
3. How a court, agency, regulator, or opposing party might characterize the facts differently; 
4. The difference between positions that are well-supported, reasonable, aggressive, uncertain, or weak; 
5. Procedural risks, penalties, deadlines, burden of proof, evidentiary problems, and disclosure considerations; 
6. Practical drafting, negotiation, litigation, or compliance consequences. 

Please cite only real legal authorities. Do not invent cases, statutes, regulations, administrative rulings, quotations, or citations. If you are uncertain whether an authority exists or whether it stands for the stated proposition, say so clearly. 

For each important authority, provide enough identifying information so that I can independently verify it. Where possible, include citations to the relevant statute, regulation, case, agency ruling, or official source. 

Do not overstate the conclusion. Distinguish settled law from arguable positions. If there is no clear authority directly on point, say so and explain the closest available authorities by analogy. 

Use a conservative professional tone suitable for review by a lawyer, court, government agency, client, or opposing counsel. 

6. My Practical Conclusion 

The legal research process has changed. For my practice, the large traditional paid research platforms are no longer worth the cost. They are expensive, slow, and tied to an older search model. LLM platforms are faster, more flexible, and better at the early stages of legal analysis. 

But lawyer beware: Do not trust, and always verify. The future is not traditional legal research versus AI. It is LLM-assisted legal research, lawyer-verified authorities, and professional judgment. That is the new model, and it works.