Patterns Emerging from SCOTUS OT '25 Oral Arguments and Opinion Authorships

What signals who will author opinions for the Court? Details from this term start to provide tell tale signs.

In Case v. Montana, Justice Gorsuch arrived at oral argument with a specific legal theory already in hand. He pressed counsel on whether a private citizen historically could enter another person’s property to prevent serious harm. He conveyed that officers should have no fewer rights than a private citizen would. He insisted that any emergency-entry privilege is bounded — it authorizes entry, not a general license to search the premises. When his separate concurrence appeared, it tracked those questions almost exactly: the Fourth Amendment’s emergency-aid exception, he wrote, is grounded in the common law of necessity and the historic privilege of private citizens to enter property to avert serious physical harm.

That alignment between argument and opinion is not incidental. Across OT2025’s argued cases decided so far, the oral-argument transcripts reveal two distinct signatures of opinion ownership. For majority opinions, the clearest signal is early conversational centrality — which justice occupies the exchange in its opening phase. For concurrences and dissents, the stronger signal is linguistic: whether the framing a justice advances at argument later reappears in the written opinion. Neither signal is a reliable point prediction of authorship. Together, they suggest the transcript is doing something more than documenting advocacy theater. It is recording an early, partially visible stage of the opinion-writing process.

The dataset created for the previous Legalytics article covers 41 argued cases and 15,363 speaking turns. The majority-authorship analysis draws on the 15 OT2025 cases with majority opinions available at the time of the run; the separate-opinion analysis uses 22 concurrences and dissents across 12 slip opinions with separate opinions.

For majority opinions, timing matters more than volume

The conventional assumption — that the justice who dominates the argument is most likely to write — turns out to be weaker than a subtler alternative. The better predictor of majority authorship is not how much a justice speaks overall, but whether that justice is active in the argument’s first 25 turns (see the figure above). Justices who are most present in that early window end up being the majority author at a substantially higher rate than those who are merely the most voluminous speakers across the whole argument (early share places the author in the top three 60 percent of the time, versus 47 percent for raw turn volume, with a permutation p-value of .011 for early share).

The distinction has an intuitive explanation. Early in the argument, the justices are still establishing what the case is actually about — which doctrinal frame will govern, which facts are legally relevant, which party has the harder question. A justice who is especially active in that formative phase may be doing more than asking questions. They may be taking ownership of the legal problem. The justice who talks most across the full argument may simply be an engaged questioner. Those are different things.

One complication is worth flagging. Because the in the post-Covid setup the justices are recognized in seniority order at the outset, earlier first turn is not a pure measure of initiative — it partly reflects the institutional seating sequence. The data bears this out: a justice’s order of first entry is statistically associated with authorship, but when used alone to pick a specific author it performs poorly (the author lands in the top three on this measure in only seven of fifteen cases). The early-share measure is more useful precisely because it captures sustained presence in the opening exchange rather than just the accident of when a justice first gets called upon.

The case-level record illustrates where the signal is sharpest. Justice Sotomayor’s majorities in Galette v. New Jersey Transit Corp., Bowe v. United States, and Hain Celestial Group v. Palmquist each display the early-presence profile clearly. In five others — Berk v. Choy (Barrett), Villarreal v. Texas (Jackson), Bost v. Illinois State Board of Elections (Roberts), GEO Group v. Menocal (Kagan), and Urias-Orellana v. Bondi (Jackson) — the eventual author ranked second rather than first, still within the authorship field the measure identifies.

The counterexamples are equally instructive. In USPS v. Konan, Justice Thomas authored the majority but ranked seventh on total turns and ninth on word share – perhaps not surprising for Thomas. Chief Justice Roberts authored Learning Resources v. Trump while ranking sixth on turns and eighth on word share. Neither justice looked dominant by volume. Both make more sense under an early-centrality account than under a talkativeness account.

Raw speaking volume still matters — but differently for different justices

Volume is not irrelevant. Majority authors do tend to occupy more of the Court’s speaking space than their colleagues in a given case (turn share and word share each place the eventual author in the top three roughly 47 percent of the time). But the relationship between volume and authorship is complicated by the fact that the justices have very different baseline speaking styles, and authorship does not show up the same way across all of them.

Justice Jackson is the most verbally prominent justice in the decided-case set by a considerable margin, averaging nearly 18 percent of justices’ turns and nearly 23 percent of justices’ words. Justice Gorsuch is also highly active at argument. Justice Thomas, by contrast, averages around 4 percent of turns and 3 percent of words — a gap that makes raw-volume comparisons across justices almost meaningless as an authorship signal. Thomas entering a conversation at all is a different data point than Jackson doing so.

This means that authorship does not always look like a spike in absolute terms. For justices who are already high-output at argument, authorship tends to appear as a relative increase over their own baseline — more present, earlier, than in the cases they do not write. For justices like Thomas, whose baseline is quiet, authorship may instead reflect structural position in the exchange: entering early in the seniority sequence into a case that is factually or doctrinally in his wheelhouse, without necessarily dominating the verbal record.

One justice-level anomaly is sharp enough to deserve its own note. Justice Kagan wins the longest-average-turn metric in eight of fifteen cases — more than any other justice by a wide margin — but authors only three majority opinions in the dataset, and in two of those three she does not win the long-turn measure at all. Her argumentative style runs to extended analytical turns that look, in transcript, like dominance. But extended turns are not the same as case ownership. Kagan at argument is often working through a problem aloud, or laying out a doctrinal landscape, in a way that does not translate into the early conversational centrality that majority authorship tends to require. It is a useful reminder that the signals in the data are about structural position, not intellectual output.

The best simple composite predictor in the file — combining opening-position order and share of justice turns — identifies the exact majority author in about one of four cases and places the author in the top three in roughly half (mean reciprocal rank ≈ .46). Those figures are useful not as a prediction engine but as a field-narrower: oral argument, read carefully, reduces the authorship space before the opinion is assigned or circulated.

Majority authors range broadly; separate authors drill down

One of the cleaner findings in the data concerns not how much a justice speaks, but how that speaking is distributed across the two sides of the argument. The model for majority authorship shows that concentrating disproportionate attention on the first advocate — the petitioner — is a negative predictor of writing the majority opinion (odds ratio approximately 0.42, p ≈ .040). The practical implication is that justices who are posturing, consciously or not, for the majority opinion tend to engage both advocates. They need the full record to write for the Court.

When the outcome measure shifts from majority authorship to any-opinion authorship — including concurrences and dissents — the sign flips. First-advocate concentration correctly identifies an opinion author in six of fifteen cases at the top rank, up from two of fifteen for majority authorship alone. And in each of those six hits, the winning justice is not the majority author. The justices flagged by this measure — Gorsuch in Bowe, Alito in Villarreal and GEO Group, Barrett in Bost — are in every instance a separate writer. The pattern has an intuitive logic which has already been shown to be significant in the political science literature: a justice who has already decided which side’s argument is deficient may spend the argument pressing that side harder, leaving a behavioral trace that the majority author — who must write to hold a coalition — cannot afford to leave.

Concurrences and dissents leave a different kind of trace

The majority-opinion signal is structural. The separate-opinion signal is textual, and it is considerably stronger.

When the question shifts from who wrote the majority to who wrote any opinion, the metrics that perform best change: sustained word volume and linguistic presence across the full argument become more predictive than early conversational centrality. This makes sense on its face. Separate writers may not be concerned with framing the Court’s common path. They are staking out a position of their own, and they may begin doing so at argument.

That is what the text-similarity analysis captures. A justice’s oral-argument language — the specific doctrinal vocabulary, the framing of the legal issue, the choice of which facts to emphasize — can be compared against the eventual opinion text. For majority opinions, this comparison is a modest signal at best: the justice whose argument language is closest to the majority opinion turns out to be the author about one-third of the time at rank one (40 percent top-three). Majority opinions are coalition products, shaped by assignment, circulation, and the need to hold five votes. Individual voice gets sanded down in the process.

Separate opinions preserve individual voice in a way that majority opinions cannot. They do not need other justices’ buy-ins. For concurrences and dissents, the text-similarity measure identifies the author at rank one in roughly 45 to 55 percent of cases depending on specification, and places the author in the top three in 60 to 65 percent (see the figure above).De-styled variants — which strip out case-specific proper nouns and procedural boilerplate to focus on structural legal language — perform comparably, suggesting the signal is conceptual rather than merely lexical.

The Gorsuch concurrence in Case v. Montana is the clearest single illustration of why. His argument questions were not just topically related to his concurrence. They were structurally identical. He organized the oral exchange around the common-law necessity privilege; he organized the concurrence around the same principle. When the text-similarity measure works well, that is what it is detecting — not shared vocabulary but shared legal architecture, visible first in the questions a justice chooses to ask and later in the opinion she chooses to write.

The measure’s ceiling — and what explains it

The cases where the similarity analysis fails are not random, and the pattern of failure is itself informative. Two structural features of the current Court account for most of the noise.

The first is the Thomas problem — though it is partly soluble. Justice Thomas authored five separate opinions and the full-text similarity measure misses him every time: his average cosine similarity score when he authored a separate opinion (.054) sits below his average score in cases where he did not write (.067). There is essentially no linguistic signal to detect because there is almost no transcript text to analyze. Raw volume, as noted above, simply does not work for Thomas.

But content does. When the analysis shifts from how much Thomas speaks to what he speaks about, the picture changes. Thomas’s opening questions in argument concentrate on jurisdictional, procedural, and statutory-text issues at rates that are distinctively high relative to his colleagues. Measured against those content features, the data correctly flags Thomas as an opinion author in all six of the cases where he wrote separately and jurisdictional or textual questions were tracked. His questions about standing, scope, and statutory meaning are, in effect, a signature — one that volume-based measures cannot read but content-based measures can. The Thomas problem is not that oral argument contains no information about his separate writing. It is that the information is encoded in the type of his questions rather than their number.

The second structural issue cuts the opposite way. Because Justice Jackson speaks at such high volume across nearly every argument, her linguistic fingerprint is everywhere in the transcript whether or not she writes. That dominance produces false positives: in five separate-opinion cases, the similarity measure points to Jackson as the likely author when someone else wrote — Barrett and Kagan in Learning Resources, Sotomayor in USPS v. Konan, Barrett again in Bost, and Alito in GEO Group. Jackson’s high baseline cosine in non-authored separate-opinion cases — the highest of any justice — effectively creates noise around authors who are less verbally prominent. Together, the Thomas and Jackson patterns define the measure’s practical ceiling: it works best when the eventual author speaks enough to leave a distinctive trace but not so much that the trace is indistinguishable from general argument engagement.

Learning Resources v. Trump is the hardest single test in these decisions, and worth examining on its own terms. The case generated six separate opinions — concurrences by Barrett, Gorsuch, Jackson, and Kagan, plus dissents by Kavanaugh and Thomas. That is the most complex opinion ecology of any case this term. The similarity measure correctly identifies three of the six authors at rank one: Gorsuch, Jackson, and Kavanaugh. The three misses — Barrett (ranked seventh), Kagan (ranked fourth), and Thomas (ranked sixth) — are each explicable. Thomas is the Thomas problem. Barrett and Kagan both wrote relatively brief opinions on grounds that overlapped substantially with Jackson’s far longer concurrence, which the measure naturally scored highest. The result is not a failure of the method so much as a reflection of what happens when individual legal voices converge on a shared doctrinal concern: the loudest expression of that concern absorbs the similarity signal.

What the data can and cannot establish

The most important qualification is that this analysis does not directly account for ideology or coalition structure. Some of the early-centrality signal for majority authorship likely reflects the behavior of justices who are already positioned near the center of the likely coalition — justices who know, or suspect, where the majority is heading and are already working to define it. Some of the linguistic continuity in separate opinions may reflect ideological distinctiveness rather than rhetorical individuality as such. Disentangling those effects would require a larger sample and a structural model of assignment and coalition formation.

A second limit is the sample size. Fifteen decided argued cases is a meaningful start, but it is not a basis for strong distributional claims. The separate-opinion similarity results are drawn from a subset of that already modest universe. The patterns here are real enough to be worth taking seriously, but can be explained more as an artifact of behavior from this term than something that necessarily traces across multiple terms.

What the data does establish is simpler and, for Court watchers, more immediately useful. Oral argument is not mere performance. The transcript contains information about who is taking ownership of a case — information that begins to appear before any opinion is assigned, circulated, or announced. For majority opinions, that ownership tends to look like early conversational presence and broad engagement across both sides of the argument. For separate writings, it tends to look like the persistence of a justice’s individual frame from the bench into the opinion — and, for quieter justices, the particular doctrinal register of the few questions they do ask. The Gorsuch concurrence in Case v. Montana is the clearest example this term of what that pattern looks like when it works: a justice who organized the argument around a specific legal principle and then organized his opinion around the same one.

Oral argument, on this account, is allocation as much as performance — one of the places where the justices begin, in public and in real time, to decide who will own the case.