Law firms have moved from AI exploration to integration – what's working and what's not?
When ChatGPT was launched two years ago, it took the world by storm. Lawyers rapidly adopted it while law firms clamoured for solutions, and we entered a media-induced hype cycle around generative AI. Based on recent comprehensive surveys of large law firms and considering the technological progress of the last two years, I see clear patterns emerging around generative AI.
Aligning with what I have observed over the last two years, the 2025 Citi Hildebrandt Client Advisory report sets out six significant learnings from a survey of large US firms on their use of generative AI:
Generative AI follows a well-known pattern in technology called the Gartner Hype Cycle. The Gartner Hype Cycle starts with a technology trigger (generative AI in this case), which then surges into new technological developments and eventually moves into the peak of inflated expectations, now seen with the proliferation of new tools along with some early announcements from law firms on their use, successes, and failures with technology.
The legal industry is flooded with new generative AI tool offerings. To give you an idea of the size of the explosion in generative AI, in October 2023, Legal Tech Fund published a generative AI landscape chart with 96 different solutions. Just nine months later, LegalTech Fund published another early-stage legal tech generative AI landscape with over 300 solutions. With so many available AI options, lawyers and law firms are overwhelmed and left to wonder which ones to select and which ones to pilot.
When we were first introduced to generative AI solutions, there was a naive optimism that these tools would seamlessly integrate into legal workflows and provide immediate and perfect results. However, it quickly became evident that accuracy and reliability posed significant challenges, with the phenomenon of "hallucinations" undermining trust in the technology. Soon after, the concept of “prompting,” providing an initial text input or instruction that guides a tool like ChatGPT, emerged as a critical skill.
Effective use of these tools required more than just the right inputs; it also needed tailored training to unlock their full potential. Even two years later, while some workflow-specific solutions have reduced the need for intensive prompting, legal tech's most powerful AI tools still demand skillful prompts to deliver optimal results.
Moreover, integrating these tools into legal practice isn't just about teaching the software; it's about deeply understanding lawyers' workflows and adapting the tools to meet their needs. The scale of this task makes it impractical to provide one-on-one support for every lawyer in a firm. As a result, I have heard several firms saying they are now exploring creative and scalable solutions to train their teams in the skills to use generative AI effectively in legal practice.
A lawyer recently asked me whether they could use generative AI to assist them in an area of law they had little experience with. They were curious whether the AI software incorporated appropriate case law, legal documents, and/or legislation to review a particular contract comprehensively.
I responded that while some technologies may incorporate legal knowledge, they still fail to offer an adequate evaluation at the level that a lawyer with expertise would bring. Tech tools are helpful but should not be considered foolproof: It’s important to have human eyes look over generative AI results and confirm their accuracy.
I saw Vals.AI’s recent presentation of their legal AI benchmarking study. They found that Large Language Models (LLMs) did better than lawyers in three areas:
Similarly, the Strategic Knowledge & Innovation Legal Leaders Summit (SKILLS), an annual conference of knowledge management and innovation leaders, identified summarization as the dominant current use case in law firms, with 62 respondents ranking it as their top choice, followed by code creation and answering legal questions (13 top rankings) in their 2025 survey.
The takeaway is that current legal AI is good for certain tasks but not all legal tasks. Summarizing legal documents appears to be a popular way to use generative AI in law and has a high degree of accuracy.
It would be wise to move from considering to implementing generative AI in 2025 and build up our skills. The data from the SKILLS survey suggests that the legal industry is entering a phase where the focus has shifted from whether to implement AI to how to implement and integrate generative AI effectively into legal practice.
This is supported by data from the Thompson Reuters 2025 State of the US Legal Market report, which found that technology spending in large law firms increased in 2024, reflecting the high cost of generative AI. Further, the SKILLS survey found that 78 percent of firms actively engage with public LLMs, while 66 percent develop internal solutions. This multi-platform approach suggests that firms are applying different AI solutions to meet specific needs within the firms.
Generative AI tools have a lot of power, but we need to know how to use them. Make 2025 a year of learning the tools and improving your skills. I have found a marked improvement in the number of queries and the general use of generative AI within our firm compared to last year.
If you haven’t already, don’t wait to build your knowledge and your skills – start today, and a year from now, you may be an expert in using generative AI, too.
