Best Practices for AI Contract Drafting and Analytics

by David Silbert


If LinkedIn’s news feed is any indication of the state of AI in the legal industry, then it’s time for attorneys to embrace a new aspect of practicing law. Whether you practice law with a firm or in-house, attorneys need to master the art of legal writing for an AI audience. “What,” you might ask, is an “AI audience”? The literal answer to that question is AI software – the tools that apply machine learning, latent semantic indexing, and natural language processes, like Seal Software, to read contracts and report on critical legal and business intelligence. But the better answer is simply the client. Whether it’s compliance, transactional, or business-driven, every company is looking to increase the speed and accuracy of its contract analysis, while driving down costs within a predictable spend model.

The choice is simple:  Either learn to draft documents for your clients’ new AI model or draft documents that impede their business efforts. Assuming you want to be on the right side of that equation, here’s a basic primer to help you write for your new audience.

Garbage In, Garbage Out

If a document is not drafted with AI in mind, its utility is diminished. The most essential and most obvious way to enable AI is to ensure that all documents are digital or OCR ready. That means that poorly scanned image files (while still usable) will leave valuable data on the table. Instead, presenting a properly scanned document or a digitally friendly text or PDF file ensures that the client will have the best chance to leverage the text in your document. The best practice is to use documents that have been saved to directly to PDF, or (at a minimum) been scanned in at 300 dpi (scanning at 300 dpi is not officially a standard for OCR, but it is considered the gold standard). This simple, upstream step will pay dividends to the client’s ability to leverage their AI tools downstream.

The Right Way, or the Wrong Way

The best way to learn is by doing. So, let’s explore a model commercial marketing provision to see the right way and the wrong way to leverage AI. This is an extreme example, but it addresses a number of significant drafting conventions. 

Ex. A – Good AI Drafting

Marketing. Vendor shall not use Company’s name, logo, or marks without Company’s prior written consent. Notwithstanding the foregoing, Vendor may use Company’s name, logo, or marks if they are used in marketing materials among a list of other entities for whom the Vendor provides services or in the context of otherwise publicly available information. Vendor may not use Company’s name, logo, or marks in marketing materials that promote the Company’s competitors (including, but not limited to Acme, Inc., GloboTech, LLC, and MegaCorp GmbH). Vendor is prohibited from using Company’s name, logo, or marks in the following regions:  the European Union, Africa, and South America.

Ex. B – Bad AI Drafting

Marketing – Consent Required. Vendor shall not use Company’s name, logo, or marks (“IP”) without Company’s prior written consent. 

Marketing – Exceptions. Notwithstanding the foregoing, Vendor may use Company’s IP:

  • if they are used in marketing materials among a list of other entities for whom the Vendor provides services, or
  • in the context of otherwise publicly available information.

Marketing – Competitors. Vendor may not use Company’s IP in marketing materials that promote the Company’s competitors, including, but not limited to:

  • Acme, Inc.,
  • GloboTech, LLC, and
  • MegaCorp GmbH.

Marketing – Geographic Limitations. Vendor may only use Company’s IP in the following geographic regions:



North America


South America




Middle East


Asia Pacific


European Union




At first glance, many attorneys are going to prefer the conventions laid out in Ex. B – clear headers, defined terms, lists, and charts. To a human audience, these conventions and their markers provide a clear road map to complex topics. But to an AI audience, these conventions only complicate or eliminate comprehension. Let’s address them one-by-one.

I’m Good, I Don’t Need a Break

Excessive line breaks and page breaks can inhibit AI from recognizing relevant concepts. Some breaks are unavoidable. Indeed, there is only so much room on a page. But excessive or stylistic breaks in text, can mean missed concepts downstream. AI tools look at words and concepts at the phrase, sentence, paragraph, and section level of a document. Inserting artificial breaks, like those in Ex. B, reduce the tools’ ability to evaluate related concepts. By contrast, the single paragraph in Ex. A enables AI tools to see that the paragraph (without any line breaks) is entirely related to the concept of marketing. Moreover, sophisticated users of AI tools can create analytics that will recognize the individual, related concepts (e.g., exceptions) within the paragraph itself. 

Nowhere is the issue of orphaned concepts more apparent than the use of the bulleted list in Ex. B, which appears to AI tools as words floating in space. By contrast, Ex. A keeps those names linked to the language explaining that those entities are the Company’s competitors. And sophisticated AI practitioners can create analytics that will automatically report the specific names of the competitors identified in the agreement when prepared in the style of Ex. A. That data is lost in Ex. B. Simply put, breaking up concepts inhibits AI tools from making those connections and reporting granular, related concepts.

This Isn’t Soccer, Headers Don’t Score Points

AI tools either don’t or shouldn’t leverage headers to identify concepts (with limited exceptions). The first actual provision in Ex. B is the perfect example of a useless header for AI. Nowhere in the operating clause is the word “marketing” used. By contrast, the paragraph as a whole in Ex. A contains language pertaining to names, logos, marks, and marketing materials. AI tools will see those concepts grouped together in that paragraph and know that the language pertains to marketing rights. 

AI Isn’t Hi-Def

Too many defined terms are also problematic, particularly when paired with excessive line breaks. Note that the use of the defined term “IP” in Ex. B only saves the drafter 9 words. But it hides the key words of names, logos, and marks in each of its subsequent uses. Defined terms are valuable and essential but should be used only when drafting truly demands it. Otherwise, you will have defined terms in orphaned concepts where AI tools will not automatically connect the dots.

AI Navigates Better in Uncharted Territory

To be clear, charts have their place in complex contracts. However, it might not have the same placement when drafting an AI contract. From elaborate rent schedules to extensive carve outs in appendices, attorneys rightfully leverage charts to consolidate detailed information into a simplified view. Unfortunately, AI tools see charts as … well … charts. And the information contained therein is largely inaccessible. Fortunately, AI tools are constantly evolving to learn new tricks, charts included. And AI tools now have the ability to extract a chart or table, recognizing the individual cells therein, so that the data can be transferred to a database in the same format. But AI tools still lack the ability to recognize the meaning contained in those charts. One can certainly argue that the table in Ex. B contains more information than the parallel language in Ex. A. But does it really? The drafting in Ex. A contains the same inferences as to permitted regions without actually stating them. But more to the point, Ex. A allows AI tools to extract the concept and even report out the specific geographies where usage is prohibited. Ex. B does not.

Old Dogs, New Tricks

AI in the legal space is only increasing. And savvy practitioners already leverage these tools. Incorporating best practices upstream at the drafting stage, will produce significant downstream benefits. Attorneys who know how to draft for an AI audience distinguish themselves from their competition and demonstrate value to their clients, not only in their legal acumen but in their ability to anticipate their clients’ business needs. Attorneys don’t need Ph.Ds. in AI to add value. They just need to write for an AI audience – a new trick any old dog can master.

Leave a comment