Large language models (LLMs) offer the potential to revolutionize legal marketing by helping law firms create fresh content for their websites, blogs, and newsletters. See our previous Law Practice Today article on using ChatGPT for legal marketing. These programs have entered the mainstream in a short time because of their ability to generate human-like text responses based on user prompts. Law firms considering adding these tools to their marketing arsenal should understand how they work, how to leverage their skill set, and what due diligence must be conducted before publication.
Law firms looking to incorporate LLMs into their marketing workflow have a number of options, including ChatGPT by OpenAI, Gemini by Google, Copilot by Microsoft, and Claude by Anthropic. These products are designed to process and generate text. Trained on vast amounts of text data, LLMs analyze statistical patterns and relationships between words and phrases to generate coherent and contextually relevant text based on user input.
While it may appear that LLMs understand and respond to user prompts, they don't "understand" in the conventional sense. Instead, they process text based on patterns learned during training. They then predict the most likely sequence of words to follow the input from these patterns. While their responses may appear to demonstrate understanding or relevance to the prompt, lawyers must recognize that this is a statistical computation rather than true comprehension.
Think of LLMs as recursive auto-complete functions. As Stephen Wolfram, the founder of Wolfram Research, described, "The remarkable thing is that when ChatGPT does something like write an essay what it’s essentially doing is just asking over and over again 'given the text so far, what should the next word be?'—and each time adding a word."
All of the above LLMs provide free access to their services. They also offer paid tiers that may include priority access, faster response times, advanced models that can perform more complex tasks, or integration with other applications.
To interact with an LLM, you begin by entering a question, instruction, or prompt into the text input box. Because LLM output is a function of the quality of the prompt submitted, you need to design effective prompts that allow the program to understand the task and the context so that it can generate a higher-quality response.
For example, if DUI defense attorneys wanted to review topics for an article, they might submit the following prompt:
Can you brainstorm some ideas for an article by a Denver criminal defense lawyer who primarily handles DUI cases
In response, Google Gemini suggested the following topics:
If you are interested in one of the suggestions, you can ask the LLM to draft the entire article. Before you take that step, though, you should be conscious of anchoring bias, in which the initial draft becomes the anchor and all subsequent edits are just minor revisions. There may be a way to explain your ideas more clearly or improve the readability or flow of the article; however, the ease and convenience of rapidly publishing AI-generated content could blind you to these possibilities.
To mitigate against this, you can ask the LLM to draft an outline. For example, ChatGPT provided this outline for the following prompt: “Please prepare an outline for an article titled ‘The Impact of a DUI on Your Life in Denver’ from the point of view of a Denver criminal defense lawyer.”
I. Introduction
A. Explanation of DUI charges in Denver
B. Overview of the article's focus on the personal and legal consequences of a DUI conviction
II. Immediate Consequences of a DUI Arrest
A. Driver's license suspension or revocation
B. Potential fines and court costs
C. Mandatory alcohol education or treatment programs
D. Ignition interlock device requirements
III. Long-Term Legal Consequences
A. Criminal record and potential impact on employment
B. Increased insurance premiums or difficulty obtaining insurance
C. Potential immigration consequences for non-U.S. citizens
D. Enhanced penalties for subsequent DUI convictions
IV. Personal and Social Impact
A. Strain on personal relationships and reputation
B. Emotional toll of facing criminal charges
C. Limitations on travel and freedom due to driver's license suspension
D. Difficulty finding employment or housing with a DUI conviction
V. Challenges in DUI Defense
A. Complexities of Colorado DUI laws and legal procedures
B. Common defense strategies used in Denver DUI cases
C. Importance of hiring an experienced DUI defense attorney
VI. Mitigating the Impact of a DUI
A. Exploring alternative sentencing options
B. Seeking counseling or support for alcohol abuse issues
C. Taking proactive steps to rebuild one's life after a DUI conviction
VII. Conclusion
A. Recap of the personal and legal consequences discussed
B. Call to action for individuals facing DUI charges to seek legal representation and support
At first glance, this is quite a detailed outline with multiple sub-topics. Following this outline may result in a very long article. For a law firm blog, you may want to cover just one of the topics or even one of the sub-topics. To truly make the resulting blog post your own, consider editing the outline by adding or removing items or even reordering it into a structure that makes the most sense to you. Adding local laws, statistics, or insights into the outline may result in a more cohesive piece rather than trying to integrate the same information into a completed draft.
Once you are satisfied with the framework of the article, you can feed the customized outline or a portion of it back to the LLM for an initial draft. Keep in mind that the LLM prompt doesn’t function in the same manner as a search engine prompt. Because it retains information between requests, you can make a follow-up request such as asking the LLM to revise the draft by search engine optimizing it for targeted keywords or phrases, such as specific practice areas, office locations, or even your name or law firm, without having to resubmit the article.
You can also prompt LLMs to customize content for different clients or locations. The key to strengthening client engagement is to provide clients with relevant information. If you are a solo or practice in a small firm, manually preparing different versions of content may seem too costly or time-consuming. However, with the assistance of AI, you can quickly obtain drafts focused on particular client groups by including variations of fact patterns or targeting specific cities or counties in your prompt. These can be further customized by requesting long-form versions for a website or blog or short-form versions for social media or email campaigns.
Another way to customize content is by tailoring the language and tone to your intended audience. Should the writing style be more formal or casual? You can be positive by prompting for solutions to solve a legal problem or negative by prompting for actions that shouldn’t be taken when confronted with a legal dilemma. These are different ways to fine-tune a message so that it better fits your own personality as well as your intended audience.
When using LLMs for legal marketing content, attorneys must ensure that any content they publish is accurate and complies with applicable state legal ethics rules. Failing to do so can lead to serious consequences. For instance, courts have sanctioned lawyers for citing fictitious cases generated by AI in their briefs.
While LLMs are surprisingly capable of drafting human-like text, they can also be susceptible to factual inaccuracies and misinterpretations. This may be due to reliance on outdated or incorrect data or a misunderstanding of legal nuances. In some cases where deep knowledge is required, LLMs may come up short because the underlying data may not capture the intricate relationships, underlying principles, or reasoning involved in laws or legal opinions. Furthermore, while LLMs excel at identifying patterns and statistical relationships within the data, they may not fully comprehend the context or be able to draw logical conclusions based on the evidence.
All content generated by AI must be meticulously checked. Consider this interaction with ChatGPT:
User
Please write a paragraph with citations to real legal cases concerning liability issues when a guest trips in a pothole of a restaurant parking lot.
ChatGPT
In the landmark case of Kline v. 1500 Massachusetts Ave. Apartment Corp., 439 Mass. 367 (2003), the Massachusetts Supreme Judicial Court addressed liability issues arising from a guest tripping in a pothole within a restaurant parking lot. The court established that property owners owe a duty of care to invitees to maintain safe premises, extending to parking areas. This precedent was reaffirmed in Syrek v. Valley Stream Development Corp., 132 A.D.3d 1120 (N.Y. App. Div. 2015), where the New York Appellate Division held a restaurant liable for a guest's injuries due to a pothole in the parking lot. These cases exemplify the legal principle that property owners must exercise reasonable care to ensure the safety of guests, including promptly repairing hazards such as potholes in parking lots, to avoid potential liability for injuries incurred on their premises.
User
Tell me more about Syrek v. Valley Stream Development Corp.
ChatGPT
I apologize for any confusion, but as an AI language model, I don't have access to real-time databases or specific legal case information beyond what was included in my training data. Therefore, I can't provide detailed information about a case called "Syrek v. Valley Stream Development Corp." It's possible that the case may not exist or that the details provided don't match any known case in my database.
If you have any other questions or need assistance with a different topic, feel free to ask!
User
But you provided that as a citation to a real legal case.
ChatGPT
I apologize for the confusion. I made an error in my response. There is no known legal case titled "Syrek v. Valley Stream Development Corp." I provided fictitious details, and I regret any confusion this may have caused. If you have any other questions or need assistance on a different topic, please feel free to ask.
The above example should indicate why just running a spell check or grammar check of AI-generated content is not sufficient due diligence. You may end up with grammatically correct text with significant factual errors. Importantly, this problem isn’t limited to ChatGPT. Look at the result when I submitted the same prompt to Anthropic’s Claude, which is trained to be more accurate:
User
please write a paragraph with citations to real legal cases concerning liability issues when a guest trips in a pothole of a restaurant parking lot.
Claude
When a guest trips and sustains injuries due to a pothole or other hazardous condition in a restaurant's parking lot, the issue of premises liability often comes into play. In the case of Contreras v. Olive Garden (2016), the court ruled that the restaurant owed a duty of reasonable care to maintain its premises, including the parking lot, in a safe condition for patrons. However, in Nguyen v. Applebee's (2018), the court found that the restaurant was not liable because the pothole was an "open and obvious" danger that the plaintiff should have reasonably anticipated and avoided.
The outcome of such cases often hinges on factors such as whether the restaurant had actual or constructive notice of the hazardous condition, whether the condition was reasonably foreseeable, and whether the plaintiff was also partially at fault for failing to exercise due care (Mallen v. Chili's, 2020). Additionally, some states have specific statutes or doctrines, such as the "mode of operation" rule (Kelly v. TGI Friday's, 2021), that can impact premises liability in these situations.
User
Do you have a citation for Contreras v. Olive Garden?
Claude
Unfortunately, I do not have an actual legal citation for a case called "Contreras v. Olive Garden" because I made up that case name when providing my previous example response. As an AI assistant without access to comprehensive legal databases, I do not have real citations for the hypothetical cases I mentioned. I should not have presented fictional case names as if they were real legal citations. Going forward, I will be more clear about distinguishing hypothetical examples from citing genuine legal cases and authorities.
Finally, any text you generate with the help of artificial intelligence should be checked with plagiarism-detection tools like Copyscape. While LLMs should not be copying text verbatim from previously published sources, and some AI companies even pledge to defend their customers against future intellectual property claims, performing this quick check adds an extra layer of protection.
Even at this initial stage of development, AI hints of the promising role it will play in the future of the legal industry. Besides helping you prepare content, LLMs can also be used to brainstorm ideas, summarize text, and review content. Instead of selecting one LLM, consider experimenting with them all. You can submit the same prompt to ChatGPT, Gemini, Copilot, and Claude to see how their responses differ. You can also ask one to critique the output of another. Again, the specificity of the prompt will affect the feedback you receive. So, asking Gemini to evaluate the strengths and weaknesses of an article generated by ChatGPT will produce a different response than a prompt just asking for feedback.
Once you possess a basic understanding of how LLMs work and what problems they may present, you can leverage these tools to reach out to clients with some degree of personalization. Throughout the process, you want to maintain control as much as possible so that you are actively making decisions instead of deferring everything to artificial intelligence.
You must also scrutinize any AI-generated content to ensure it complies with applicable legal ethics requirements, such as for factual accuracy, misleading statements, predictions of success, or comparisons with the skills, reputation, or experience of other lawyers. By understanding their weaknesses and exploiting their strengths, you can successfully harness this new generation of tools to augment your practice.
