Similar wording does not, by itself, tell you whether people are thinking alike. A preprint revised on 29 September 2026, “The Hitchhiker’s Guide to Monoculture: AI Homogenizes Syntax, Not (Necessarily) Semantics”, reports that distinction in Kaggle code submissions.
Its author, Gordon Burtch, finds growing similarity in literal syntax since AI assistants spread, but no measured convergence in problem-solving approach or intent. If that split holds, a policy aimed at protecting diversity of thought needs more than evidence that people write in similar ways.
The preprint that splits the question
Burtch’s argument, stated in the abstract, is that “convergence in language need not imply convergence in ideas, and existing evidence rarely distinguishes between the two.” The paper’s design tries to enforce that distinction. It examines Kaggle contest submissions from 2019 to mid-2026, a domain chosen because AI coding assistants diffused there early and because code lets you separate surface form from underlying strategy more cleanly than prose does.
The study uses different measures for different questions. Term-frequency-inverse-document-frequency (TF-IDF) n-gram representations track surface syntax: which tokens appear, and in which patterns. Voyage code-3 retrieval embeddings are used to represent intent and approach: what a submission is trying to do.
Within a contest, the submissions became more similar on the syntax measure, but not detectably more similar on the intent measure. That result is author-reported. The paper’s conclusion is that shared AI tools “are homogenizing how Kaggle contestants express their solutions, not the problem-solving strategies they employ.”
The title points to a concrete example. The paper documents submissions converging on the random seed value 42, which it reports as consistent with LLMs reinforcing a programming-culture convention associated with Douglas Adams’ novel. In the paper’s account, that is a shared mannerism: it changes how the code looks without changing what it computes.
Keep the limits attached to the result. All quantitative findings are author-reported, and the published abstract does not give detailed methods or effect sizes. The absence of measured convergence in ideas also comes from code, not prose. A contest with a verifiable objective gives researchers a clearer way to separate expression from strategy than an essay, policy memo, or brainstorm with no scoring function.
Three claims hiding in one word
Before writing a rule about AI and homogenization, decide which change you are trying to address. Policy documents routinely mix together changes in expression, ideas, and decisions. Each has different evidence and calls for a different response.
Syntactic homogenization is convergence in surface expression: vocabulary, phrasing, n-gram distributions, stylistic tics. This is the claim with the most evidence. The Kaggle preprint reports it in code, and the prose-domain evidence points the same direction. A cross-journal analysis of AI tool usage in academic writing cites a 2026 study by Aydın and colleagues finding that AI-generated academic texts tend to exhibit poor readability and relatively high similarity rates. That is a surface-level finding: readability scores and text similarity, not idea content.
Semantic homogenization is convergence in ideas: the approaches people take, the arguments they make, the hypotheses they generate. This is the claim most AI-writing bans implicitly assert, and no reviewed study measures it directly in prose. The Kaggle preprint reports a null here, in its own domain, with its own metrics.
Algorithmic monoculture means many decision-makers use the same model and their decisions become correlated. The concern is about shared outcomes, rather than shared phrasing, and it has a separate research literature.
Treating evidence about wording as evidence about ideas can send a policy in the wrong direction. A similarity score on student essays measures syntax. Calling it a diversity audit changes the claim without changing what was measured.
The evidence menu
The metric needs to match the concern. Surface similarity, diversity of ideas, and correlated decisions need different study designs. They also support different interventions:
| Claim | What the metric must capture | Representative evidence | Intervention it can support |
|---|---|---|---|
| Syntactic homogenization | Surface similarity: n-grams, TF-IDF, readability | Hitchhiker’s Guide preprint (syntax convergence, author-reported); Aydın et al. 2026 similarity rates, cited secondhand in arXiv:2502.00632 | Disclosure norms, style guidance, editing standards |
| Semantic homogenization | Approach and intent: embeddings over solution strategy, rubric-scored argument diversity | Hitchhiker’s Guide preprint (Voyage code-3 embeddings; null result, author-reported); small L2 argumentative-writing experiment | Bans or diversity interventions, only if such evidence exists |
| Algorithmic monoculture | Correlated decision outcomes across agents | Kleinberg and Raghavan, PNAS 2021; price-of-anarchy bound; the Critics paper | Model-diversity requirements in screening and allocation pipelines |
The prose evidence is limited. The strongest prose-specific item is a 10-participant experiment on ChatGPT-guided argumentative writing for second-language learners. Its abstract reports better clarity, logical coherence, and use of evidence in the guided group. The unguided control group did better on language mechanics and articulation of main arguments. At n=10, with only abstract-level reporting, this suggests ways to measure prose beyond wording; it cannot support a policy on its own.
The counter-case: monoculture models do find real harm
The strongest “what”-level result here concerns decisions, not writing. Kleinberg and Raghavan’s PNAS 2021 paper on algorithmic monoculture and social welfare models firms screening applicants. Each firm can use its own noisy ranking or adopt a shared algorithm that is more accurate for that firm in isolation.
In the model, every firm can rationally choose the shared algorithm while their collective decisions become worse on average. The mechanism is a Braess’-paradox-style effect, and it has been experimentally verified for more than two firms. No outside shock is needed: the harm exists in equilibrium. Each firm gets a more accurate ranking, but the shared algorithm values the same candidates in the same way, so correlated selection offsets part of the gain.
An average can also miss what happens to a particular person. The same paper describes applicants being locked out of the market when every employer or lender uses the same screen, even if average decision quality holds up. A stable accuracy average would not reveal that exclusion.
This literature is why “monoculture” deserves its own row in the taxonomy rather than serving as a synonym for text similarity. The harm Kleinberg and Raghavan model is invisible to any wording-similarity metric: it lives in correlated outcomes, not correlated phrasing. Further reading: consensus failure in AI search.
But the harm is bounded, and contested
Before any policy team cites Kleinberg and Raghavan as settled proof that shared models are dangerous, they should read the 2026 replies. “Price of Anarchy of Algorithmic Monoculture” generalizes the earlier model and reports what its authors describe as the first worst-case bound on the social welfare loss from monoculture: a tight constant of 2 on the price of anarchy. In the authors’ framing, monoculture, and decentralized optimization more generally, is close to optimal. The earlier paper exhibits settings where harm occurs; this one bounds how bad the harm can get within the model class.
A second paper, “Algorithmic Monoculture and its Critics”, formalizes the main objections to monoculture, covering systemic exclusion, agency and gaming, and information aggregation and exploration, and concludes that monoculture is “less problematic than its critics have supposed”: commonly cited objections fail, and the ones with force are not decisive. It cites Kleinberg et al. (2025) for the claim that as long as firms can rationally choose whether to solicit new information, societal outcomes cannot be dramatically worse under monoculture. It also flags an assumption both welfare models share and calls it implausible: that decision-makers’ errors are independent under polyculture. If everyone’s fallback judgment is correlated in the same direction anyway, the diversity benefit of separate decision-making shrinks.
The evidence leaves a policy team with qualified findings. Harm occurs in some settings (2021). The worst case within the model is bounded at a factor of 2 (April 2026). Researchers also dispute the realism of the models’ key independence assumption (2026). A risk memo can describe the mechanisms without presenting their real-world magnitude as settled.
Measurement traps: when the number answers a different question
An evaluation can produce correct numbers and still answer the wrong question. A September 2026 audit of an AI agent pipeline reports 139 routing cases, 112 passes and 27 failures (arXiv:2609.12017), with zero gating failures. The arithmetic was correct, but known gaps had been exempted from the gate. The evaluation therefore measured something different from what its labels suggested. As the authors put it: “Agent evaluations can be numerically correct while measuring a different construct from the one implied by their labels.”
That is precisely what a wording-similarity diversity audit does. The similarity score is real. The label “idea diversity” is not what it measures. The fix is not a better similarity score; it is a scope definition that names the claim before the metric is chosen. Audit-ecosystem design guidance from AIES 2022 puts the requirement plainly: an audit’s scope should be sufficiently delineated, not too broad and not too narrow, or its results become difficult to interpret or translate into enforcement and changed practice. A screening-tool audit scoped to “the model’s outputs” when the claimed harm is correlated rejection decisions across employers is too narrow. A campus-wide “AI impact on student thinking” audit scoped to everything is too broad. Both produce numbers that cannot carry a policy.
Decision guide: match the lever to the evidenced harm
The practical rule that falls out of the evidence menu:
- If your evidence is similarity scores, readability metrics, or style-drift observations, your claim is syntactic. The levers that fit are disclosure norms and style guidance. The disclosure lever already exists in practice: the cross-journal analysis describes AI usage declarations in academic publishing as an emerging transparency norm “conceptually akin to conflict of interest (COI) statements.” That norm manages surface convergence and provenance without asserting anything about thought.
- If you want to ban AI writing or fund a diversity intervention on the grounds that “AI makes everyone think alike,” you are making a semantic claim. No reviewed evidence establishes that claim in prose, and the one direct test in code returned a null. The burden is on semantic-level evidence: intent-embedding studies, rubric-scored argument diversity, longitudinal ideation measures. Commission that measurement before the ban, not after.
- If your concern is many institutions routing decisions through the same model, you are in monoculture territory. The relevant analysis is decision-outcome correlation, the welfare debate runs from demonstrated harm instances to a tight bound of 2, and the levers are structural: model diversity requirements, independent fallback evaluation, applicant-level recourse for the lockout cases Kleinberg and Raghavan identify.
Use the claim’s name in the policy itself. A rapid evidence scan of 13 AI risk frameworks and 831 mitigations finds widely used risk terms referring to different actors, actions and mechanisms. “Homogenization” and “monoculture” can fragment in just this way. A reader should not have to guess whether a rule concerns wording, ideas, or shared decisions.
Limits and falsifiers
Everything above the monoculture sections rests on one author-reported preprint whose semantic null covers Kaggle contest code from 2019 to mid-2026, measured with TF-IDF n-grams and Voyage code-3 embeddings, and whose full methods and effect sizes are not in the published abstract. Absence of idea convergence in contest code does not establish absence in essays, summaries, or open-ended ideation, and the v3 revision date (29 September 2026) means the paper has not yet faced extended peer scrutiny. The prose-domain counter-evidence is thin in both directions: no reviewed study demonstrates idea-level convergence in human writing, and the creativity literature even contains the counter-position, in “How AI Generates Creativity from Inauthenticity”, that AI widens rather than narrows creative scope, which is an argument rather than a measurement.
What would settle the semantic question in prose is not mysterious: pre-registered studies that measure approach and intent directly, using embedding spaces or rubrics designed for ideas rather than tokens, across domains where the objective is not externally fixed the way a Kaggle leaderboard fixes it. Until those exist, the defensible policy position is narrower than most current drafts. Wording convergence is real and documented. Thought convergence is asserted and unmeasured. Monoculture harm is modeled, bounded, and contested. Draft each rule against the claim its evidence actually supports, and treat any audit that reports a similarity score as a diversity verdict as a construct error, because that is what it is.
Join the discussion
Share a useful perspective or ask a question about this article.