Locating Ethos in the Age of AI DOWNLOAD PDF

Shortly after ChatGPT launched in late 2022, I was doing what most of us were doing: testing it, poking at it, trying to figure out what it could and couldn’t do. On a whim, I asked it to write a love poem for my husband. It produced something earnest, rhyming, and profoundly mediocre. I called him into the room. “I wrote you a poem,” I said, and read it aloud with as much feeling as I could muster.

He teared up.

Then I told him ChatGPT had written it. His face changed immediately. He wasn’t just disappointed. He was hurt. “But it was terrible!” I said, laughing. He didn’t laugh back.

I felt awful. So awful that I sat down and wrote him an actual poem. For the record, my poem was also terrible, but he loved it.

That exchange taught me an early lesson about the value of human authorship in the age of AI. It turns out, there are some kinds of writing you just shouldn’t delegate. When you delegate poetry-writing, you irrevocably change its nature such that the final product no longer shares the same meaning as the version that inspired it.

For a long time, I assumed that something similar was at work in legal writing, that it mattered whether a brief was written by a person or by AI. But when I looked more closely, the analogy began to break down. The credibility of legal writing almost never lies in the identity of the person who drafted the words. It rests instead on a shared account of professional trustworthiness, drawn upon by every licensed attorney and invoked by the lawyer’s signature. Generative AI did not create or destroy that arrangement. It simply made it visible.

Where we’ve been looking

Ethos is the credibility a speaker brings to an argument, the quality that makes an audience inclined to trust and be persuaded. The concept comes out of a world of live oratory, where much of a speaker’s ethos was quite literally embodied. 

Ancient rhetoricians disagreed, however, about where that credibility came from. Aristotle insisted that ethos should arise from the speech itself. Even a genuinely good person must demonstrate good sense, good character, and goodwill in the argument, because the audience’s trust should come from what is said rather than from prior reputation. Isocrates took the opposite position. For him, genuine credibility depended on the speaker’s actual character because an audience weighed the speaker’s life alongside his words. Ethos existed before the speaker ever opened his mouth. Cicero, the working advocate of the three, split the difference in practice, drawing on his standing and reputation while also crafting the character he presented in a given speech. Despite their disagreement, all three assumed credibility could attach to a particular human being standing before the audience.

Writing strips that human presence away, and the ancient disagreement followed ethos onto the page. A text cannot carry its author’s living presence. What it carries is a self, constructed out of word choice, tone, and structure, the figure Wayne Booth called the “implied author.” The reader never meets the writer. She meets a paper double and has to decide how much to trust it, and on what basis.

Legal writing scholarship has taken up all of these threads. The dominant answer has been textual. Credibility is built through the document itself, so we teach students careful analysis, precise citation, clean prose, and honest treatment of adverse authority. Michael Smith’s influential catalog of the trust-building choices available to a legal writer, sorted into Aristotle’s categories of intelligence, character, and goodwill, is the fullest version of the model. Others have brought the constructed self into legal writing, locating ethos in a “discoursal self” that exists within the four corners of the document. Still others describe ethos as a relationship of trust that develops between writer and reader over the course of a document. The accounts differ, but they share a premise that credibility is something the individual writer creates. Get the text right, and trust follows.

None of these accounts quite fits legal writing.

Not in the text

Start with the text. The underlying premise of ethos, and of rhetoric generally, is that the speaker has a choice about what to say. The less choice, the less rhetorical the document, and few genres offer less choice than the litigation brief. Take the appellate brief, the classroom’s classic persuasive assignment and the example I return to throughout this essay. An attorney drafting one is bound by rules of court on everything from what she may argue to page count and font size. She cannot even claim impartiality, because her ethical obligation is to advocate for one side.

Even the choices that remain are less free than they look. Much of what passes for ethos in legal writing is really just compliance with the rules of professional conduct. The Model Rules of Professional Conduct require a minimum threshold of ethos from every attorney, which makes evincing ethos almost synonymous with following the rules. Disclosing adverse authority does show candor, but the inference works only if the disclosure is voluntary. In the same way an apology feels insincere if forced, a disclosure made in compliance with a legally enforceable obligation says little more than that the writer follows the rules. That isn’t nothing. It is also hardly the imprimatur of honesty it would be if it weren’t required.

To be clear, choice does not vanish entirely, and neither does textual ethos. A brief writer still chooses her theme, her framing, her concessions, and how honestly she characterizes a bad record. A lawyer who declines to overclaim earns something real. For example, judges regularly report that accurate, candid briefs win their trust, while slippery ones forfeit it. The argument here is not that these choices carry no credibility. Rather, they cannot be the foundation. They adjust trust at the margin for a reader who already has some baseline reason to take the document seriously. Where that baseline comes from is the question this essay is trying to answer.

Meanwhile, ethos instruction has tended to overemphasize the part of credibility that is easiest to teach and to grade—proper punctuation, correct formatting, immaculate citation, met deadlines. Legal writing courses teach far more than mechanics, and the best instruction reaches framing, candor, and restraint. But mechanics are what the ethos label most often attaches to in practice, and that is quite the comedown for what Aristotle called almost the most effective means of persuasion a speaker possesses.

Mechanics are also the piece generative AI does best. The technology’s greatest strength may be its ability to produce clean, grammatically correct prose in seconds. Early on, that mastery was undercut by fabricated authority, but retrieval-backed platforms like LexisNexis’s Protégé and Thomson Reuters’ CoCounsel have reduced hallucinations, and some legal writing instructors are already trading full memos and briefs for quizzes, fill-in-the-blank outlines, team-written drafts, and greater emphasis on oral argument. This decentering of the text is a practical response to a practical problem, but it also tacitly concedes that what we were grading was not what made the document credible. 

Not in the writer

The writer fares no better, whether we mean the actual person of the Isocratean account or the constructed self of the more recent ones. Most legal documents do not have a sole author. They are the products of drafters, editors, formbooks, and templates. The “author” is whoever signs, and that has always been a polite fiction premised on the signer having verified everything beneath the signature. The fabricated-citation cases, whose numbers keep growing even as the tools improve, tell us how aspirational that premise has always been, as does the ordinary experience of anyone who has practiced law. If verification is aspirational, authorship fares worse. Clerks may draft much of what issues over judges’ names, and associates draft much of what issues over partners’. We have never really believed the signer wrote the words. We impute authorship anyway.

The reader’s side is stranger still. A judge does not simply absorb a brief and render a decision. She and her clerks can pull any cited source, test any proposition, and develop lines of reasoning neither party offered; in principle, nothing in a brief has to be taken on faith. She is also expected to decide on the law, not on the writer, so much so that it would be improper for her to admit that any part of a decision rested on who the attorney was. Self-negation may be the ultimate act of ethos in legal writing. The document becomes more credible the more its author disappears behind the reasoning. A discoursal self, constructed to vanish, is a strange home for credibility, and a writer-reader relationship in which the reader is instructed to ignore the writer is a strange thing to call the seat of trust.

In the credential

If the ethos of persuasive legal writing isn’t in the text and it isn’t in the attorney as author, then where is it? The answer is that the credential itself conveys ethos. Legal writing scholarship does not have a clean name for this category. Like personal character, it exists before the first word of the document is read. But unlike anything in the classical accounts, it is impersonal and transferable. When a judge picks up a brief signed by a licensed attorney she has never encountered, the judge is not extending credibility to that attorney personally. The attorney is drawing on a common account, built up and maintained by the profession as a whole. The signature works as synecdoche, with one member standing in for the whole.

The mechanism for drawing on that account is the signature. A signature on a brief is less like a byline than like a wedding vow. A byline claims authorship. A vow is what J.L. Austin called a “performative.” In other words, it does something in the saying, and it works only from the right lips (or pen) under the right conditions. If an AI says, “I now pronounce you married,” nothing happens. If an unlicensed person signs a brief, nothing happens either, or rather, something happens, but it is unauthorized practice rather than a filing. And what the attorney vows is not that she wrote the brief. It is that she stands behind it, that the factual contentions have evidentiary support, that the legal arguments are nonfrivolous, that the authorities are real. At the trial level, Rule 11 has said this in so many words for decades. We have been reading the signature as a claim of authorship when it was always a promise of answerability.

The vow also explains what trust is for in a system whose readers can check anything they want. They can check anything, but they cannot check everything. No chambers independently verify every record cite, every quotation, every representation about what happened below. Trust does its work in the zone where checking is impractical, and the certifications required in court rules are promises about exactly that zone.

If the signature never vowed authorship, though, why should anything change just because the drafter is now a machine rather than a clerk? In one sense, nothing should. Clerks, associates, formbooks, and templates have been producing lawyers’ words for as long as there have been lawyers, and the credibility of those documents survived because the signers stood behind them. Machine drafting does not change what the signature means.

What it changes is how easy the vow is to break. A clerk’s draft arrives with a human being who has done the reading. An AI draft can be produced, instantly and plausibly, without anyone having done the reading at all. That is what makes the fabricated-citation cases scandalous. The scandal is not that a machine wrote the words; someone other than the signer has been writing lawyers’ words all along. It is that the signer signed without standing behind the thing signed. The cases also expose the polite fiction described earlier: attorneys were never verifying everything beneath their signatures as thoroughly as we assumed. Now they will have to. With the other human checkpoints gone from the production chain, the signature is the only place where verification is guaranteed, and the standard for signing has never mattered more.

The profession’s response to the fabricated-citation cases has been loud: standing orders, ethics opinions, disciplinary referrals, front-page coverage of small-stakes mistakes. The harm in any single case is often modest, amounting to sanctions, an embarrassed lawyer, and a wasted afternoon of chambers time. The disproportionate response makes sense once you see collective ethos as a shared resource. Every member draws on the account; any member can drain it. Each fabricated citation that reaches a courtroom is a depletion event, a withdrawal against every lawyer’s credibility at once.

Where we go from here

If ethos lives in the credential, three implications follow from the advent of generative AI.

First, unauthorized-practice and certification rules will need to be rewritten around who signs rather than who drafts. The profession has long told two stories about its monopoly. The first is a competence story, in which the license marks who can do the work, and the second is an accountability story, in which the license constitutes a class of answerable persons. Generative AI dissolves the factual premise of the competence story because drafting competence is no longer scarce, or will not be for long. The accountability story, by contrast, survives generative AI untouched because what a judge trusts about a signed brief was never the signer’s rare drafting talent. It is that the signer can be sanctioned, disciplined, sued, disbarred. Rewriting the rules around the signature would let many more people, and machines, into the room where legal documents are produced while keeping intact the liability structure that makes those documents credible. And because certification regimes take years to rewrite, waiting for the technology to finish arriving means starting the rewrite too late.

Second, the duty behind the signature will need to be defined by verification capacity rather than by effort. A signature-centered regime carries a danger that scholars of automation call the “moral crumple zone”: the human who remains in an automated system absorbs blame for failures she had limited power to prevent. Effort-based standards make that happen, because effort can only be litigated backward from a known harm. Once a fabricated citation has been filed, the answer to “would more inquiry have caught it?” is always yes. A verification-capacity standard avoids the trap. It imposes a strict duty to check what is checkable (citations, quotations, the record) and, for the judgment calls that cannot be checked (issue selection, the theory of the case, what to concede), a duty of non-delegation. The zone that cannot be audited is the zone where the signer must actually be the author. This structure would not be an invention; securities law built its due-diligence defense the same way decades ago, calibrating each signer’s duty to what that signer could reasonably verify.

Third, legal writing courses matter more, not less. If we are initiating students into enforceable answerability rather than scarce skill, then the first-year writing course is where students learn what it means to stand behind a document: how to verify work they did not perform, what standard must be met before signing, and when the right answer is to refuse to sign at all. Drafting stays at the center because you cannot verify an analysis you could not have performed, and if students stop drafting, the profession loses, one generation on, its capacity to check. Nothing about the traditional course was wasted. The struggle to draft was building the capacity to verify all along. We just called it writing.

My husband did not care whether the poem was good. He cared whether it was mine. When I began thinking about generative AI and legal writing, I assumed the profession’s unease matched his. I no longer think that. The poem is safe from AI forever, because the maker’s identity is its content. Nobody will ever want an optimal love poem. A better-written poem from someone else, or something else, is not a superior poem; it is a different and worse act. The brief was never like that. It is judged by its effect on a reader who does not care who the actual writer is, and it never mattered whether one person wrote it, so long as someone stood behind it.

1. ^

    Aristotle, On Rhetoric: A Theory of Civic Discourse bk. I, ch. 2, at 1356a (George A. Kennedy trans., 2d ed. 2007) (identifying ethos, pathos, and logos as the three pisteis furnished through speech).

2. ^

    Id. (persuasion through character “should result from the speech, not from a previous opinion that the speaker is a certain kind of person”); see also id. bk. II, ch. 1, at 1378a (a speaker establishes ethos by demonstrating practical wisdom (phronesis), virtue (arete), and goodwill (eunoia)).

3. ^

    Isocrates,Antidosis § 278, in 1 Isocrates 246, 246 (Yun Lee Too trans., 2000) (“[W]ords seem more true when spoken by those who lead good lives than by those whose lives have been criticized and . . . proofs based on a person’s life have greater power than those provided by speech.”); see also William Benoit, Isocrates and Aristotle on Rhetoric, 20 Rhetoric Soc’y Q. 251, 257 (1990).

4. ^

    See Cicero, De Oratore bk. II, §§ 182–84 (E.W. Sutton & H. Rackham trans., Loeb Classical Library 1942).

5. ^

    Wayne C. Booth, The Rhetoric of Fiction 70–71 (2d ed. 1983); Roger D. Cherry, Ethos Versus Persona: Self-Representation in Written Discourse, 5 Written Commc'n 251, 260–63, 268–69 (1988).

6. ^

    Michael R. Smith, Advanced Legal Writing: Theories and Strategies in Persuasive Writing 125 (3d ed. 2013).

7. ^

    J. Christopher Rideout, Ethos, Character, and Discoursal Self in Persuasive Legal Writing, 21 Legal Writing: J. Legal Writing Inst. 19, 42–43 (2016).

8. ^

    Melissa H. Weresh, Morality, Trust, and Illusion: Ethos as Relationship, 9 Legal Commc'n & Rhetoric: JALWD 229, 234–35 (2012).

9. ^

    The everyday bulk of persuasive filing is trial-level motion practice, and the argument here applies to those documents as well. Given crowded dockets and thinner clerk support, some of it may even apply with more force at the trial level.

10. ^

     See Smith, supra note 6, at 125–34 (cataloguing the traits that convey ethos through the text; nearly every one is also a requirement embodied in the Model Rules of Professional Conduct).

11. ^

     See, e.g., Kristen Konrad Robbins-Tiscione, The Inside Scoop: What Federal Judges Really Think About the Way Lawyers Write, 8 Legal Writing 257, 277–78 (2002).

12. ^

     Indeed, until relatively recently, most authors who addressed the topic of ethos and pathos in legal argument focused on oral argument. Legal writing texts focused primarily on logos. See Weresh, supra note 8, at 232.

13. ^

     Eugene Garver, For the Sake of Argument: Practical Reasoning, Character, and the Ethics of Belief 7 (2004); Aristotle, supra note 1, bk. I, ch. 2, 1356a (a speaker’s “character may almost be called the most effective means of persuasion he possesses”); Isocrates, supra note 3, at 78–81; Smith, supra note 6, at 125 (“A strong argument can be made that ethos is more important to persuasive legal writing than either logical argument (logos) or appeals to emotion (pathos). In fact, the effectiveness of both emotional and logical arguments depends in large part on the perceptions of the advocate’s credibility.”); Benoit, supra note 3, at 257; Michael Frost, Ethos, Pathos & Legal Audience, 99 Dick. L. Rev. 85, 104 (1994); Rideout, supra note 7, at 19–20.

14. ^

     See, e.g., Carolyn V. Williams, Bracing for Impact: Revising Legal Writing Assessments Ahead of the Collision of Generative AI and the NextGen Bar Exam, 28 Legal Writing: J. Legal Writing Inst. 1, 46–74 (2024) (arguing for the incorporation of more of what Williams calls “traditional” assessment methods, in contrast to “performance” methods which rely on written products like memos and briefs as evidence of learning. Williams recommends alternative assignments such as multiple-choice quizzes and tests, final exams modeled after the new NextGen Bar, partial fill-in-the-blank outlines, oral exams, collaborative drafts, self-assessment, revising AI-written drafts, and even the development of creative projects like podcasts).

15. ^

     See, e.g., Park v. Kim, 91 F.4th 610, 615–16 (2d Cir. 2024) (referring an attorney for disciplinary proceedings after she submitted a reply brief citing a nonexistent case generated by ChatGPT and conceded that she had not confirmed the citation’s accuracy); Mata v. Avianca, Inc., 678 F. Supp. 3d 443, 466 (S.D.N.Y. 2023) (sanctioning attorneys $5,000 for filing a brief drafted by ChatGPT that cited entirely fabricated cases, which the supervising attorney signed without verification). For the continuing growth of such cases even as the tools improve, see Damien Charlotin, AI Hallucination Cases Database, https://www.damiencharlotin.com/hallucinations/.

16. ^

     Arguments from pathos face similar constraints. But see Smith, supra note 6, at 123–24 (arguing that credibility determinations can significantly affect how legal decision-makers receive both logical and emotional appeals).

17. ^

     The account offered here concerns lawyer-signed court filings, with the brief as the paradigm. Office memoranda, transactional documents, and demand letters occupy different rhetorical situations (no judicial reader, no Rule 11 certification) and are outside the frame, though the credential may do similar work in some of them. Judicial opinions raise a separate question: their credibility rests on the authority of the office rather than on license-backed answerability, and the ethos of judicial writing deserves its own treatment.

18. ^

     Neighboring literatures come close without quite arriving here. Rhetoric scholars writing about organizations describe an “institutional ethos” that individual speakers borrow from the entities they represent, and the sociology of professions has long treated licensure as a collective credibility project. See, e.g., Ruth Amossy, Ethos at the Crossroads of Disciplines: Rhetoric, Pragmatics, Sociology, 22 Poetics Today 1, 9, 20–21 (2001); Magali Sarfatti Larson, The Rise of Professionalism 40–42 (1977); Andrew Abbott, The System of Professions 59–61 (1988). 

19. ^

     Reputation still matters, of course. In a small local bar, judges know exactly which lawyers are careful, and they discount accordingly. But reputation is an adjustment to a default. See Rideout, supra note 7, at 41 n.7 (noting hypothetical exceptions to purely text-based ethos determinations where the writer has known characteristics beyond the text, such as an attorney known to the judge).

20. ^

     J.L. Austin, How to Do Things with Words 6–7 (2d ed. 1975).

21. ^

     Id. at 14–15. 

22. ^

     Fed. R. Civ. P. 11(b). Although Rule 11 does not apply to appeals, it has analogs in the Federal Rules of Appellate Procedure, especially Rule 38 governing frivolous appeals and Rules 46(b) and (c) governing discipline for conduct unbecoming an attorney or for failure to comply with an applicable rule. Fed. R. App. P. 38, 46(b)–(c).

23. ^

     For a collection of judicial standing orders on AI use, see the University of Chicago Library’s guide, Generative AI in Legal Research, Education, and Practice, https://guides.lib.uchicago.edu/AI/Practice; see also ABA Comm. on Ethics & Pro. Resp., Formal Op. 512 (2024) (addressing generative AI).

24. ^

     For the competence/rent framing and its critics, see generally, e.g., Clifford Winston, Robert W. Crandall & Vikram Maheshri, First Thing We Do, Let’s Deregulate All the Lawyers (2011); Deborah L. Rhode, Access to Justice (2004).

25. ^

     Madeleine Clare Elish, Moral Crumple Zones: Cautionary Tales in Human-Robot Interaction, 5 Engaging Sci., Tech. & Soc’y 40, 40–41 (2019).

26. ^

     Escott v. BarChris Constr. Corp., 283 F. Supp. 643, 681–90 (S.D.N.Y. 1968); see also Fed. R. Civ. P. 11 advisory committee’s note to 1993 amendment; Model Rules of Pro. Conduct r. 5.3 (Am. Bar Ass’n, amended 2012) (changing “nonlawyer assistants” to “nonlawyer assistance” so that the supervisory duty reaches non-human help).

27. ^

     The insight that writing is itself a mode of learning and analysis is well established in composition studies and has been applied to legal writing pedagogy. See generally, e.g., Donald M. Murray, Teach Writing as a Process Not Product, 71 Leaflet 11 (Nov. 1972), reprinted in Cross-Talk in Comp Theory: A Reader 3 (Victor Villanueva & Kristin L. Arola eds., 3d ed. 2011); Peter Elbow, Writing Without Teachers (2d ed. 1998); Janet Emig, Writing as a Mode of Learning, 28 C. Composition & Commc’n 122 (1977); Philip C. Kissam, Thinking (By Writing) About Legal Writing, 40 Vand. L. Rev. 135 (1987).