The post below tackles other vices of AI-generated prose (building on this post) and shows you how to avoid them — but first, take an hour to learn how you should use AI:
I can show you legal-AI workflows you haven’t seen before — ones requiring no software beyond a subscription to Claude or ChatGPT. Join me for a live Zoom presentation and Q&A.
AI Workflows for Lawyers · November 19 · Noon–1 p.m. Pacific · $149
The best AI models can write fluent, cohesive, and grammatically flawless text — a computing revolution — but a look beneath that polished surface reveals a tendency to generate bland, vague sentences that sometimes border on the meaningless.
As in my earlier post about AI’s characteristic sentence patterns, the legal-brief quotations below are drawn from the 56 amicus briefs filed in In re OpenAI Copyright Litigation; as you’ll see, some of the amici’s attorneys probably took advantage of OpenAI’s wares when weighing in on the case. (My firm, Rushing McCarl LLP, wrote an AI-free amicus brief for ten AI startups —including Splice, Boomy, and ProRata) in support of the plaintiffs.
Four signs that you’re reading AI slop.
1. Abstract nouns and vague verbs
Static sentences — sentences built on verbs describing states, such as is or lacks — are often necessary because many sentences, including this one, describe situations or things rather than events. But such sentences are less likely to have the forward momentum, salience, and visualizability that make readers want to continue.
Those virtues are better achieved by sentences about characters performing actions (ELW § 4.4) — sentences that show someone doing something. Active sentences put that action in the verb, while static ones encase it in nouns called nominalizations.
AI tends to write static sentences packed with abstractions, making the sentence’s substance difficult to grasp and impossible to visualize:
• A general-purpose AI system is the product of organizational choices.
• The distinction is model specific.
• That shared structure explains why corpus breadth matters to the development of general capabilities.
• The distinction between preserving information and publishing all information is essential.
• Training-data provenance is a foundational accountability mechanism.
Perhaps to avoid the monotony of sentences built on is and are, AI sometimes swaps these for vague relational verbs: involve, reflect, emphasize, represent, indicate, suggest, relate to, facilitate, encompass, underscore, connect, and encapsulate. Pairing such verbs with abstract nouns leads to sentences that are superficially coherent but unintelligible:
• Across many works, recurring patterns can contribute to general capabilities for comparing and synthesizing reports.
• The Universal Guidelines for AI likewise connect transparency to accountability and human responsibility.
Tacking on throwaway modifiers doesn’t clarify the sentence:
One of the most significant ways they have done so has been to hew closely to the statute’s insistent focus on the particular works allegedly copied and whether the alleged infringer had created a competing substitute of that work.
2. Low-information topic sentences
AI often generates sentences that are devoid of substantive content. It especially favors using these as paragraph-opening topic sentences — which is the opposite of the content-first, conclusions-plus-reasons approach that I teach.
Empty topic sentences like these — many of which contain backreferences (pointing phrases) and are not self-contained — make text unskimmable:
• Generative AI training fits within that framework.
• The work of CAIDP Global Academic Network members reinforces this point.
• History underscores the wisdom of that rule.
These sentences center on undefended, generalized evaluations or characterizations containing no new information. They take the reader’s attention for granted, assuming that the reader will stick around to learn the reasons supporting an assertion even though the text doesn’t bother to preview them.
3. Artificial cohesion through excessive backreferences
Cohesion and coherence are what make a single text hang together as a unit pursuing one rhetorical goal; without them, the text would amount to a jumble of disconnected sentences and paragraphs. (Cohesion refers to a text’s verbal surface while coherence refers to its underlying meaning.)
Backward-pointing cohesive ties, more commonly known as transitions, are especially useful for promoting both values by connecting one sentence and paragraph to another. As I said in Elegant Legal Writing: “Think about adjacent sentences and paragraphs as links in a chain, with each link pointing backward to previous information.”
Top AI models write cohesive prose but at the expense of substance:
That lesson is especially important here.
That sentence points backward to the “lesson” and forward to the reasons for the lesson’s importance — one cohesive device pointing in each direction. But the sentence doesn’t say anything, so it’s useless. Any sentence or clause that doesn’t have its own propositional content and communicative purpose should be cut.1
Attorneys editing AI drafts sometimes compound the problem of AI vacuity by emphasizing an AI-generated sentence’s vague announcement with hyperbole, piling on even more emphasis to a sentence that doesn’t deserve it:
The implications of Plaintiffs’ theory would be extraordinary.
That sentence supplies no information except that the writer for some reason finds unspecified implications of Plaintiffs’ theory (abstractions of an abstraction) to be extraordinary. I suspect the label extraordinary was added by the attorney. AI models, which hedge their writing (often, typically, etc.) even more compulsively than attorneys do, prefer sleepier and unfalsifiable modifiers like essential:
• Those decisions helped pave the way for search engines that are now essential to education, commerce, research, and civic participation.
• That decision dramatically expanded public knowledge by enabling access to information once buried in distant or forgotten collections.
Many AI-generated sentences also misplace emphasis. The last stressed syllable in a sentence is its most emphatic position (ELW § 4.18). The closing words of most sentences should usually convey novel information — as the given-new principle leads readers to expect. Many AI-generated sentences instead end in contentless abstractions (framework, participation) and characterizations (important and extraordinary), flouting that principle.
4. Jargon and glibness
AI prose sometimes reflects the jargon of its tech-industry makers, which can create a grating and smug tone. My running list includes key as an adjective; land, delve, and swallow as metaphorical verbs; and load-bearing, edge case, failure mode, and shape as metaphors. Some terms, like utilize, leverage (v.), and impact (v.), are preexisting pathogens traceable to pre-AI airport books about management. Others, like furniture (n., artifacts) and stack (v., combine), are novelties that AI has at once introduced into common discourse and turned into embarrassing clichés.
Here’s an example from a Claude-generated revision memo:
Filing furniture is unfinished. The Conclusion is a placeholder. All cross-references and the Table of Authorities show stale results, and nine authorities lack TA fields.
Generating such revision memos is an ideal use of AI, and one I teach, since the AI’s response doesn’t touch your prose and won’t reach your readers. But that doesn’t make it any less annoying.
Further reading
Ryan McCarl, How Not to Write Like AI: Sentence Patterns, Elegant Legal Writing (Aug. 24, 2026), link.
Ryan McCarl, Can You Fool Me with AI-Generated Prose?, Elegant Legal Writing (Aug. 24, 2026), link.
Ryan McCarl, AI Workflows and Ground Rules for Attorneys, Elegant Legal Writing (Aug. 31, 2026), link.
Sam Kriss, Why Does A.I. Write Like … That?, N.Y. Times Mag. (Dec. 3, 2025), link.
My book Elegant Legal Writing is available on Amazon and Audible. Please help me sustain the book’s momentum by rating it five stars and sharing my posts.
Please share this post with your networks:
Thanks for reading Elegant Legal Writing! Subscribe for free to receive new posts and support my work.
Connect with Ryan McCarl: Order the Elegant Legal Writing book | LinkedIn | Rushing McCarl LLP | Linktree | Second Stage (blog) | Bluesky | Substack | Twitter (X) | Threads
Photo by Emiliano Vittoriosi on Unsplash.
Relatedly, a reader encountering a pointing phrase like “that lesson,” especially in a topic sentence, should instantly know what it refers to without needing to mentally backtrack. This becomes harder to do when the pointing phrase refers to something abstract.
