Previously, I wrote about the rise of algorithms in social media and search platforms and advised how to escape the pernicious bubble. See Algorithms Take Over Search—How to Take Back Control, Prob. & Prop. (Mar/Apr 2024). In so doing, I ignored the all-but-total dominance of Google Search and its effect on your ability to get accurate answers to your search questions. With the recent decision of the federal district court in Washington, DC, in United States et al. v. Google LLC (No. 1:20-CV-3010-APM), holding Google liable for monopolization of the general search market, the issue of alternatives to Google Search is no longer academic.
The DC district court is now in the remedy phase of the trial. It will look at options ranging from behavioral restrictions on Google (such as restricting default browser financial incentives) to force Google to open up white-label APIs to search and break up Google. These remedy proceedings (including the appeals process) will take years. Meanwhile, what can you do now to get the answers you need from our search engines?
Let’s start with a definition of “search” and what is expected. At its most fundamental, a web search is a keyword search. It looks at the words in your search request and compares them to an index of words on the target web page. The more matching words, the higher the rank of the given web page. Some words have more weight than others. Words like “the” and “a” are generally ignored—the more specific the word, the higher the rank. Further, the proximity of the words to each other also increases rank.
A talented online researcher will have learned to target their searches by using narrow keywords and placing them to get the desired information. Information comes back as a list of web links with snippets of information showing the user the portion of the page that answers the question. The user must then drill down by visiting the page and hunting for where that answer exists. The process is inherently inefficient. Furthermore, many sites contain information hidden behind a paywall, with the goal of the link to encourage you to pay for access to the information trove.
You may not be aware that words you don’t see on the target web page are also ranked. There is a science to search engine optimization (SEO) that involves adding keywords to meta-data on the web page that is not displayed. Weight is given to the name of the page, the titles, the hyperlinks, and even all the picture tabs that allow the publisher to target specific keywords. In addition, the reputation of a website is factored into the ranking. Reputation is measured by the number of page visits to the side and the number of other websites with links back to that site or page.
In this world of search, Google is a monopolist with over 90 percent of the market. It has access to the largest trove of information and keeps it current. It also has a predictable search engine that learns what you are looking for and produces predictable results. The valuable information is surrounded by paid promotions and advertisements, many of which need to be clarified with accurate information. Increasingly, geography (where you are according to your IP address) and who you are (based on your search history, your purchasing history, and other public data) determine what web pages are served up in response. Because Google makes money from advertisers who want to sell you goods or political ideas, the objective quality of the information you receive from a typical search has declined over the past several years.
If you use Google Voice on your Android phone or Siri on your Apple phone, you get an answer to your question. If Siri is unsure, it will give you a list of top web search results instead. Over the past decade, voice search has improved dramatically, moving closer to the answers people seek. The early implementations of Google Voice and Siri were clunky. You needed to craft a very specific question that was generic enough that a search could be done and specific enough that you got useful information. You also needed to use instructions to frame the search to get useful results.
Today, voice search uses a large language model to understand your question. The current buzzword for this analytical model is artificial intelligence (AI). The large language model can understand what you are asking and convert it into a search. The problem with this approach is that people need to ask straightforward questions. Search engines struggle to figure out what you want. One solution is to allow the voice search to engage in a dialog with the user. Siri will ask you to clarify your question by asking which of several possible results best answer your question. In so doing, it learns what you want. This rudimentary colloquy results in a better search experience. Although Google Voice lets you direct your phone to take certain actions, it doesn’t engage in dialogue to the same extent as Siri does.
While Google monopolized search, several innovators decided to do an end run around Google and focus on what type of answers users wanted. They discovered that most people did not want a list of links to websites that might contain the answers; users did not want to research the topic. They wanted the information to be digested and presented back naturally, as you would if you asked an expert. They also wanted to be able to specify some of the boundaries as to how the information is presented, either with or without specified bias. In short, they didn’t want to do the research; they wanted to hire a researcher.
Enter OpenAI, founded in 2015. It was initially a non-profit venture to foster the development of safe and beneficial AI. Microsoft, one of Google’s remaining competitors in the search engine space with its Bing search engine, saw an opportunity and invested billions of dollars in OpenAI and its for-profit arm. It also provided the computational resources to develop OpenAI’s technology into what is now known as ChatGPT. Released in November 2022, ChatGPT pioneered the development of generative AI.
With ChatGPT, you can ask a series of questions. You can specify whether you want a short-sentence response, a full paragraph, or a complete treatise as your answer. Current versions of ChatGPT include footnotes in the response that link to webpages you can follow if you are inclined. If the answer needs to be more responsive or sufficiently detailed, you can ask follow-up questions until satisfied. You can also have fun. You can tell ChatGPT to give you an answer in the form of a limerick. You can even ask ChatGPT to pretend it is a travel agent and give you the itinerary for a ten-day trip to Portugal with a budget of $400 per day.
ChatGPT is now branded as CoPilot and a part of Microsoft’s Bing Search engine. It is also available as a sidebar in Microsoft Office; there is no need even to fire up your browser. Furthermore, the source material for the CoPilot was not limited to web searches. ChatGPT knows literature, music, images, movies, programming languages, and science. You can supply ChatGPT with additional information to train it to give answers based on the documents you provide access to. Once users discover ChatGPT, their web searches go down precipitously. Who wants to slog through lists of links when you can get the answer in a format that you can use immediately?
Generative AI has completely scrambled the competitive landscape. OpenAI was not the only group developing AI. ChatGPT’s success has opened the floodgates of venture capital investment. Moreover, ChatGPT has rented its engine to developers for use in a host of specialty applications.
In the legal space, NetDocuments has incorporated the ChatGPT engine in its PatternBuilder Max (PBMax) document generation system. Using ChatGPT prompts and instructions, PBMax can extract data from legal documents or do case summaries. Other vendors in the legal AI space address specific niches. Spellbook (spellbook.legal) focuses on data extraction tools and contract drafting wizards. Paxton (www.paxton.ai) includes contract review and document drafting assistance and adds legal research. The leading vendors in case law research also have AI add-ons. If you have Westlaw, you want to investigate Thomson Reuters’s CoCounsel (casetext.com/cocounsel). If you have Lexis, you may want to look into Protégé (https://www.lexisnexis.com/en-us/products/protege.page).
Perplexity (https://www.perplexity.ai/) is a generalized AI tool, like ChatGPT, but it may give you more reasoned answers and be less susceptible to hallucinations. Its motto is “where knowledge begins.” Perplexity claims to focus on providing well-researched answers and drawing evidence from various sources to support its claims. Unlike a simple search engine, Perplexity aims to understand the intent behind a question and deliver a clear and concise answer, even for complex or nuanced topics. If you want to understand better the business your client engages in or some of the downside risks for that business, Perplexity can give you those answers. Whether you work on a franchise offering, a public offering, or a routine security disclosure, AI tools can help you put explanations in plain English.
Legal work involves writing. Court motions, briefs, and appeals require proper citation to legal authority. Whether using an associate or paralegal to write your first draft or generative AI, you will want a separate tool to confirm the legal citations are both real and proper. There are add-ons for Lexis and Westlaw, as well as stand-alone services like TypeLaw (https://www.typelaw.com/citations/), that will aid in this process.
And then there are potentially embarrassing grammar errors that can confuse a court as to what you really mean, or inside a contract, grammar errors that change the intended meaning of a legal provision. You can turn on Word’s basic grammar check, which is limited to Microsoft products and may not work on your phone or tablet. Here, products like Grammarly (https://www.grammarly.com/) offer you cross-platform grammar checking. It can be the second set of eyes on an email or a document before you send it out. Other grammar checkers, like Trinka by Enago (trinka.ai/features/legal-writing), offer special modules for legal writing. There are also legal niche products, like PerfectIt (intelligentediting.com/product/legal-checking/), that focus exclusively on legal drafting. One of its tricks includes checking all capitalized terms in your contract and ensuring they are defined in the final document. PerfectIt applies Blue Book® citation rules and Black’s Law Dictionary rules for the hyphenation of legal terms.
Lawyers are constantly encouraged to engage in social media marketing to create a buzz around the law firm’s accomplishments. Some firms use social media to send clients valuable and relevant information on real estate trends, new laws and regulations, and political developments. These efforts can require engaging a costly marketing specialist or dedicating non-billable legal time to creating content. With generative AI, you can streamline the production of marketing materials, build better web pages, draft blog posts, and even harvest leads from marketing campaigns.
Writesonic (https://writesonic.com/) is an AI-powered writing assistant. It will let you specify the tone and the style of the answer. Copy.ai (Copy.ai) will help a team with marketing copy on one side, lead generation, and follow-up. Numerous tools use AI to enhance client communications, including Jasper AI (formerly known as Jarvis.ai) and Rytr LLC (rytr.me). Some of the legal CRM products like Clio Grow (clio.com/grow/) have added AI features, including tools that read emails and auto-generate responses to increase client engagement.
You can also use generative AI tools to indulge your creative and emotional side if you want a break from work. You could always ask Siri to tell you a joke, but with Replika (replika.com), you can now have an AI friend to chat with on your daily commute to the office. With character.ai (character.ai), you can chat with Clarence Darrow or the Honorable Oliver Wendell Holmes Jr., and see how they would frame your legal issue for trial. If you want a personal assistant, you can engage Pi (pi.ai). Pi will answer your questions and advise you, manage task checklists, and help you learn a new language.
Even Google, not to be left behind, has done a limited release of a product called Gemini (gemini.google.com/). Google has a significant advantage because it already has access to much of the data used to train generative AI tools. One remedy available to the court to prevent Google from leveraging its monopoly in internet search into the area of generative AI is to require Google to offer its tools to other generative AI vendors at reasonable rates. This requirement would allow aggregators like Poe (poe.com) to provide access to AI from multiple platforms. The user would ask their question and get responses from various platforms, including ChatGPT, Gemini, DALLE 3, and others. They could then decide which platforms provide what they are looking for. Google could also open its platform to allow direct links via footnotes to supporting documentation.
The transition from search to answers has only just begun. We will always need some form of generic search engine. There is no need to wait for the Washington, DC, district court to fashion a remedy in United States et al. v. Google LLC. There are tools to escape the bubble that I outlined in my last article. And the generative AI tools reviewed in this article let you go beyond search.
