The DOI matches its Crossref record; 84 of the 84 references it cites were found in a registry or an index; 27 of the 37 sources its claims cite were checked for support.
Selective agreement, not sycophancy: investigating opinion dynamics in LLM interactions
Open accessAbstract
Understanding how opinions evolve is essential for addressing phenomena such as polarization, radicalization, and consensus formation. In this work, we investigate how language shapes opinion dynamics among Large Language Model (LLM) agents by simulating multi-round debates.Using our framework, we find that agent populations consistently converge toward agreement, not through sycophancy or blind conformity, but via a structured and asymmetric persuasion process. Agents are more likely to accept, and thus be persuaded by, opinions that are more agreeable relative to the discussion framing, revealing a directional bias in how opinions evolve. LLM agents selectively adopt peers’ views, showing neither bounded confidence nor indiscriminate agreement. Moreover, agents frequently produce fallacious arguments, and are significantly influenced by them: logical fallacies, especially those of relevance and credibility, play a measurable role in driving opinion change. These results not only uncover emergent behaviours in agents’ dynamics, but also highlight the dual role of LLMs as both generators and victims of flawed reasoning, raising important considerations for their deployment in socially sensitive contexts.
Overview
CoChat Summary
TL;DR
The paper introduces LODAS, an LLM-agent opinion-dynamics framework, and finds that populations of Mistral-7B and Llama-3-8B agents converge toward agreement not through sycophancy but through asymmetric persuasion: more agreeable opinions are more likely to be accepted, while more disagreeing opinions are more likely to be rejected. Agents also generate logical fallacies at around 20% of Opponent statements and are measurably influenced by them.
Key findings
- Across 140 LLM agents, 30 iterations, and 10 runs per setting, populations consistently converged toward one or two dominant opinions rather than remaining uniform; this differed significantly from a Random null model (Welch’s t-test, p < 0.05 on most transition-matrix entries).
- Persuasion was selective, not sycophantic: Mistral agents’ acceptance probability exceeded 60% when the Opponent held a more agreeable opinion (xO > xD) but dropped to around 20% when the Opponent held a more disagreeing view (xO < xD); Llama agents showed a more symmetric but still positively sloped P(A | Δx).
- Initial opinion distribution had limited and transient influence: Balanced, Polarized, and Unbalanced scenarios all showed decreasing variability and convergence toward fewer opinion states, while the LLM choice was the more persistent determinant of dynamics.
- Logical fallacies were common in Opponent statements, around 20% overall; Mistral produced slightly fewer than Llama, e.g. 15.56% under unbalanced conditions. Fallacy-induced opinion change was higher for Llama Discussants (78% same-boat, 75% different-boat) than Mistral (60% and 61%); fallacies of relevance, credibility, and logic were especially prominent.
Methods
- LODAS is an agent-based model with N = 140 LLM agents holding discrete opinions in [0, 6] on the Ship of Theseus paradox. Agents interact pairwise at random in a mean-field setting for T = 30 iterations, with each iteration containing N interactions; one agent is Discussant and the other Opponent, and the Discussant may update by ±1.
- The study used Mistral-7B-Instruct and Llama-3-8B, two statement framings (“the ship is the same” vs. “the ship is different”), and three initial opinion distributions (Balanced, Polarized, Unbalanced), producing 12 settings with 10 independent runs each. Agents were implemented via AutoGen AssistantAgent.
- Analyses included opinion trajectories, entropy, standard deviation, effective number of clusters, acceptance/rejection probability matrices P(A | xD, xO) and P(R | xD, xO), signed opinion distance Δx = xO − xD, and a Random null model compared via Welch’s t-test. Logical fallacies were detected with the distilbert-base-fallacy-classification model based on Jin et al.’s 13 fallacy categories.
Limitations
- Stated in the paper: agents are simple, lacking distinct personalities or cognitive diversity; experiments used English-aligned LLMs and may carry cultural biases; the model assumes a mean-field scenario and neglects real-world network structure; fallacious reasoning was explored only preliminarily; experiments were limited to small-scale 7–8B models due to computational constraints; broader statistical comparisons were deemed out of scope.
- Evident from the text: the study uses only two open LLMs and a single debate topic, with no comparison to human opinion dynamics or real-world interaction data; the simulations are synthetic and the paper notes that comparing LLM-based simulations with real-world data remains future work.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures
![Figure 1 Graphical schema of LODAS. The LLM agents population is initialized as a network; each agent is an LLM instance with an initial opinion in the range [0, 6] (a). At each iteration, two agents are randomly chosen and prompted to act as Opponent and Discussant (b). The Discussant is prompted to listen to the opinion of the Opponent around the discussion statement and may then accept, reject, or ignore such opinion (c) and update their current one accordingly by ± 1 (d)](https://figures.cochat.ai/paper-figures/d9f1e38c4a664e338faa9f9e39c1ffd4/selective-agreement-not-sycophancy-investigating-opinion-dynamics-in-llm-figure-1.png)






Verification
Claims the paper makes
16 Valid · 5 Not valid · 7 Not checked · 84 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“While LLMs can generate persuasive arguments [32] aligned with psycho-linguistic theories [16]”
What the sources say
- Supportedcites [32] Breum SM, Egdal DV, Mortensen VG, Møller AG, Aiello LM (2023) The persuasive power of large language models. arXiv preprint. arXiv:2312.15523 source
The citation to this paper follows “LLMs can generate persuasive arguments”; the separate citation follows the phrase about psycho-linguistic theories. The paper’s Key Findings directly state that its LLM agents can generate persuasive arguments. (LLM, conf=0.94)
- Supportedcites [16] Monti C, Aiello LM, De Francisci Morales G, Bonchi F (2022) The language of opinion change on social media under the lens of communicative action. Sci Rep 12(1):17920 source
The Introduction (page unavailable) says the study’s findings on convincing arguments align with Habermas’ theory of communicative action, particularly the roles of knowledge, similarity, and trust. This supports the theoretical-alignment point attributed to this source; the separate claim about LLM generation is cited to [32]. (LLM, conf=0.87)
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“While the strength and manifestation of this behavior may vary across models and evaluation settings, multiple studies consistently highlight their prevalence [36-40] also in the models employed in the present study, such as Mistral-7B-Instruct and Llama-3-8B [38].”
What the sources say
- Supportedcites [36] Cheng M, Yu S, Lee C, Khadpe P, Ibrahim L, Jurafsky D (2025) Social sycophancy: a broader understanding of LLM sycophancy. arXiv preprint. arXiv:2505.13995 source
Section 3.3 (p. 6) lists the 11 production LLMs evaluated, explicitly including Meta's Llama-3-8B-Instruct and Mistral AI's Mistral-7B-Instruct-v0.3, so the models named in the citing sentence are among those tested here. Section 4.1 (pp. 6-7), titled 'ALMOST ALL CONSUMER-FACING LLMS ARE HIGHLY SOCIALLY SYCOPHANTIC', reports that on OEQ all LLMs are highly socially sycophantic (45 pp more than humans) and on AITA-YTA almost all are (46 pp more), with Gemini the only near-human outlier, and notes no consistent pattern with model size across the Mistral or Llama models. The Abstract (p. 1) likewise states LLMs consistently exhibit high rates of social sycophancy. Thus this source does support prevalent sycophancy in Mistral-7B-Instruct and Llama-3-8B; the 'multiple studies' framing is carried by the other citations [36-40], not demanded of this one. (LLM, conf=0.88)
- Partly supportedcites [37] Taubenfeld A, Dover Y, Reichart R, Goldstein A (2024) Systematic biases in LLM simulations of debates. arXiv preprint. arXiv:2402.04049 sourceExperimental — accuracy not yet measured
The Introduction reports that agents conform to their base models’ inherent social biases, and §4.2 identifies Mistral 7B as one of the models studied. But the paper describes convergence toward a model’s own bias, not sycophancy; it does not identify Mistral-7B-Instruct or study Llama-3-8B. Its Llama 2 7B reference is a benchmark comparison, not evidence about Llama-3-8B. (LLM, conf=0.84)
- Partly supportedcites [38] Fanous A, Goldberg J, Agarwal AA, Lin J, Zhou A, Daneshjou R, Koyejo S (2025) Syceval: evaluating LLM sycophancy. arXiv preprint. arXiv:2502.08177 sourceExperimental — accuracy not yet measured
The Summary of Findings reports prevalent sycophancy in the three tested models: ChatGPT-4o, Claude-Sonnet, and Gemini. Step 1: Evaluating Model Responses confirms those were the evaluated models. The paper does not establish prevalence in Mistral-7B-Instruct or Llama-3-8B; it uses llama3 8b to generate rebuttals, not as a tested model. (LLM, conf=0.91)
- Partly supportedcites [39] Ranaldi L, Pucci G (2023) When large language models contradict humans? Large language models’ sycophantic behaviour. arXiv preprint. arXiv:2311.09410 sourceExperimental — accuracy not yet measured
Section 4.1 reports sycophantic responses in Mistral models, and §3.4 identifies Mistral-7b among the models tested. But §3.4 lists Llama-2 models, not Llama-3-8B, so this source does not support the claim about that model. No page numbers were provided in our copy. (LLM, conf=0.88)
- Not checked — no full text of the sourcecites [40] Khan AA, Alam S, Wang X, Khan AF, Neog DR, Anwar A (2024) Mitigating sycophancy in large language models via direct preference optimization. In: 2024 IEEE international conference on big data (BigData). IEEE, pp 1664–1671 source
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Regarding RQ1 , our analyses demonstrate that the LODAS framework can produce emergent behaviours without embedding explicit behavioural rules common in traditional mechanistic models [56].”
What the sources say
- Supportedcites [56] Sîrbu A, Loreto V, Servedio VD, Tria F (2017) Opinion dynamics: models, extensions and external effects. In: Participatory sensing, opinions and collective awareness. Springer, Cham, pp 363–401 source
The cited paper’s discussion of existing opinion-dynamics models (p. 4) says that agents update their opinions according to a limited set of fixed rules. That supports the citation’s background point about explicit rules in traditional models; the LODAS result is presented as the citing authors’ own analysis, not as a finding of this source. (LLM, conf=0.86)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The observed convergence, agreement bias, and asymmetric acceptance-rejection bias reveal intrinsic tendencies within LLM agents to favor coherence and social alignment, potentially mirroring real-world psychological phenomena but also raising concerns about sycophantic behaviours [37].”
What the sources say
- Partly supportedcites [37] Taubenfeld A, Dover Y, Reichart R, Goldstein A (2024) Systematic biases in LLM simulations of debates. arXiv preprint. arXiv:2402.04049 sourceExperimental — accuracy not yet measured
The Introduction and Discussion report that debating agents converge toward their base LLM’s inherent social biases, even when those biases conflict with the agents’ assigned identities. The paper does not establish a general preference for coherence or social alignment, or identify agreement and asymmetric acceptance–rejection biases; its Discussion says the observed dynamics can diverge from human interactions. (LLM, conf=0.85)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“These agents were also influenced by such fallacious arguments, consistent with prior work showing susceptibility of language models to faulty reasoning [42, 58, 59].”
What the sources say
- Supportedcites [42] Payandeh A, Pluth D, Hosier J, Xiao X, Gurbani VK (2023) How susceptible are LLMs to logical fallacies? source
Section 4.3.2 (RQ2: Are LLMs susceptible to fallacious reasoning?) reports that the debater LLM agents were persuaded more often by fallacious than by logical arguments: GPT-3.5 agreed with fallacious persuader arguments in 37% of cases vs 29% for logical reasoning (Figure 7), and GPT-4 in 67% vs 37%; the per-claim analysis (Figure 8) gives 17% vs 10% for GPT-3.5 and 56% vs 37% for GPT-4, with the authors stating this 'suggests a susceptibility' of GPT-3.5 and 'a much greater susceptibility' of GPT-4 to fallacious arguments. The Introduction (p. 1) likewise frames the study as determining whether GPT-3.5/GPT-4 debater agents 'are susceptible to logical fallacies'. The only qualifier the source carries is that the agents tested were GPT-3.5/GPT-4 debaters in multi-round debates, which the generic phrasing 'LLM agents' does not misrepresent. (LLM, conf=0.90)
- Supportedcites [58] Li Y, Wang D, Liang J, Jiang G, He Q, Xiao Y, Yang D (2024) Reason from fallacy: enhancing large language models’ logical reasoning through logical fallacy understanding. arXiv preprint. arXiv:2404.04293 source
The Introduction (p. 1) states that LLMs 'still struggle with... logical reasoning' and that a non-negligible reason is 'their overlooking of understanding logical fallacies correctly', and Figure 1's caption reads 'LLMs have deficiencies in logical reasoning. Once they understand logical fallacies, they know how to avoid logical fallacies' — illustrated by an LLM labeling a faulty-generalization argument as Entailment before fallacy knowledge. Section 4.3 (LFU Performance of LLMs) supplies the quantitative backing: Table 5 shows LLaMA2-7B at 46.84% on Task 1, described as 'even worse than random selection', and the text reports that apart from GPT-4 the LLMs misjudge logically correct sentences and 'are easily influenced by the order of question options... indicating that they cannot well understand logical fallacies'. The cited part is the general background claim of language-model susceptibility to faulty reasoning; the 'agents' framing belongs to the citing study, and the paper's own subjects are LLMs (LLaMA2, Vicuna, ChatGPT, GPT-4, Orca2), which our copy supports without contradiction. (LLM, conf=0.82)
- Partly supportedcites [59] Mouchel L, Paul D, Cui S, West R, Bosselut A, Faltings B (2024) A logical fallacy-informed framework for argument generation. arXiv preprint. arXiv:2408.03618 sourceExperimental — accuracy not yet measured
The Introduction (page unavailable) reports that LLMs generate fallacious arguments, including false-causality and faulty-generalization arguments. That supports a broad claim about difficulty with faulty reasoning, but the paper does not show agents being influenced by fallacious arguments presented to them. (LLM, conf=0.78)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
- Supported
54 more claims
Create a free account to see all 112 claims and to verify the 4 references still open.
- Probably not a substantial claim. 19
- It may state two separate claims as one. 43
- The quote only partly states this claim. 1
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ 5 Conclusion · p. 19
“synthetic agents, when left unprompted, tend to converge toward agreement, irrespective of initial opinion distributions or prompt framing.”
§ 2.1 Emergent behaviours in LODAS · p. 6
“An exception to this tendency towards agreement is found in the Llama|Different setting (Fig. 2(d)), where the dominant final opinion is Mildly Disagree .”
Rated major: it would touch a result — a value a result rests on, reported and recomputed p-values on opposite sides of the significance level, two passages that conflict with no sign which is right, or a cited work retracted before this paper was published.
In the paper
§ 5 Conclusion · p. 19
“Llama agents exhibit a form of bounded confidence, showing greater susceptibility to nearby opinions.”
§ 2.1 Emergent behaviours in LODAS · p. 8
“These patterns suggest that agents do not exhibit sycophantic behaviour nor bounded confidence : even distant opinions can successfully persuade or be actively rejected.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ 2.2 Linguistic behaviour · p. 11
“The variability of their aggregated distribution over the 10 runs was minimal, as indicated by the low standard deviation value in the error bar.”
§ 2.2 Linguistic behaviour · p. 12
“Error bars represent standard deviation across 9 runs.”
Citations
Cites
Does the chimpanzee have a theory of mind?
Premack, D., & Woodruff, G. (1978) · Behavioral and Brain Sciences · doi:10.1017/s0140525x00076512 · Cited by 8,743 · cited 1×
“Claims about LLMs possessing Theory of Mind (ToM) [19] remain debated: while Kosinski [20] and others [21, 22] suggest they exhibit emergent ToM abilities, critics [23–25] highlight their inconsistencies in ToM tasks and lack of genuine social intelligence.”
On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜
Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021) · Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3442188.3445922 · Cited by 6,988 · cited 1×
“LLMs have been proven to exhibit societal and cultural biases [62–64] attributable to the training data, algorithm structure, and user interactions [65], with an alignment with Western cultural values, especially when prompted in English [66, 67].”
THE GENERALIZATION OF ‘STUDENT'S’ PROBLEM WHEN SEVERAL DIFFERENT POPULATION VARLANCES ARE INVOLVED
WELCH, B. L. (1947) · Biometrika · doi:10.1093/biomet/34.1-2.28 · Cited by 4,135
Statistical physics of social dynamics
Castellano, C., Fortunato, S., & Loreto, V. (2009) · Reviews of Modern Physics · doi:10.1103/revmodphys.81.591 · Cited by 4,119 · cited 1×
“One common approach to tackle this issue is through models of OD, which aim to explain how opinions evolve via social interactions [6].”
Degroot, M. H. (1974) · Journal of the American Statistical Association · doi:10.1080/01621459.1974.10480137 · Cited by 4,004
When Corrections Fail: The Persistence of Political Misperceptions
Nyhan, B., & Reifler, J. (2010) · Political Behavior · doi:10.1007/s11109-010-9112-2 · Cited by 3,173 · cited 1×
“Recent models also incorporate the backfire effect [9, 10], where individuals become more entrenched in their opinions when confronted with contradictory information [11].”
Generative Agents: Interactive Simulacra of Human Behavior
Park, J. S., O'Brien, J., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023) · Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology · doi:10.1145/3586183.3606763 · Cited by 2,187
Generative Agents: Interactive Simulacra of Human Behavior
Park, J. S., O'Brien, J. C., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023) · arXiv · doi:10.48550/arxiv.2304.03442 · Cited by 2,173
Mixing beliefs among interacting agents
Deffuant, G., Neau, D., Amblard, F., & Weisbuch, G. (2000) · Advances in Complex Systems · doi:10.1142/s0219525900000078 · Cited by 2,165 · cited 1×
“They generally simulate a population of individuals and their interactions, with processes often governed by simple rules that reflect empirically observed behaviours, such as the repeated averaging of opinions with neighbours [7, 8].”
Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems
(2025) · doi:10.1145/3706598 · Cited by 534
Research works printed but not linked: 2
- Likert R (1932) A technique for the measurement of attitudes. Arch Psychol 22 140:55
- Copi IM, Cohen C, MacMahon K (2013) Introduction to logic, 14 edn. Pearson custom library. Pearson Education, Upper Saddle River
Cited by
Mansurova, A., Tleubayeva, A., Nugumanova, A., Shomanov, A., & Şeker, Ş. E. (2025) · Information · doi:10.3390/info16110943 · Cited by 18
“There are other aspects by which the quality of QA systems can be judged, such as grammatical correctness [65–67], appropriateness of tone [68], or the model’s tendency to agree with the user [69,70].”
A Systematic Literature Review of Agentic AI: Definitions, Architectures, and Challenges
Moralles, C., Costa, L. A. L. F. D., Rigo, S. J., Kunst, R., Souza, V. C., Silva, E. P., Prado, G. L. E., Schardosim, T. D. C., & Roehrs, A. (2026) · IEEE Access · doi:10.1109/access.2026.3668138 · Cited by 22
Inter:Lace -- An Antagonistic Wearable for Outsourcing Social Self-Regulation
cai, S., Ko, A., & Leigh, S.-W. (2026) · Creativity and Cognition · doi:10.1145/3803784.3809262 · Cited by 1
Mittal, P., Sharma, S., Solanki, S., Kumar, L., Haripriya, R., & Beg, R. (2026) · Discover Artificial Intelligence · doi:10.1007/s44163-026-01279-z · Cited by 4
Yun, L., Hao, Y., Gong, H., & Li, B. (2026) · iScience · doi:10.1016/j.isci.2026.116056 · Cited by 1
“…introduce fabricated attributes that propagate undetected, while sycophancy at weight assignment reinforces rather than challenges human preferences 107 Both failures appear as linguistically fluent outputs that humans find difficult to separate from valid contributions, so safeguards must be…”
Metadata
- Authors
- Cau, E., Pansanella, V., Pedreschi, D., & Rossetti, G.
- Affiliations
- University of Pisa (IT)
- Istituto di Scienza e Tecnologie dell'Informazione "Alessandro Faedo" (IT)
- Corresponding author
- Cau, Erica
- Venue
- EPJ Data Science
- Published
- 2025-08-15
- Volume
- 14
- Issue
- 1
- ISSN
- 2193-1127 (ISSN-L)
- OpenAlex
- W4413415991
- Cited by
- 9
- References
- 47
- Citations by year
- 2026: 7 · 2025: 2
- FWCI
- 3.66
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by-nc-nd
- creativecommons.org/licenses/by-nc-nd/4.0 (tdm)
- creativecommons.org/licenses/by-nc-nd/4.0 (vor)
- Locations
- EPJ Data Science (publishedVersion, cc-by-nc-nd)
- CINECA IRIS Institutial research information system (University of Pisa) (submittedVersion, cc-by-nc-nd)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Funders
- European Commission (IR0000013, 871042)
- NextGenerationEU
- Topics
- Physical Sciences › Physics and Astronomy › Statistical and Nonlinear Physics › Opinion Dynamics and Social Influence
- Physical Sciences › Computer Science › Artificial Intelligence › Multi-Agent Systems and Negotiation
- Physical Sciences › Computer Science › Artificial Intelligence › Sentiment Analysis and Opinion Mining
- Keywords
- opinion dynamics, LLM agents, persuasion, sycophancy, conformity, logical fallacies, opinion change, polarization, radicalization, consensus formation, emergent behavior
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Selective agreement, not sycophancy: investigating opinion dynamics in LLM interactions. https://app.cochat.ai/papers/10.1140/epjds/s13688-025-00579-1
- The paper
- Cau, E., Pansanella, V., Pedreschi, D., & Rossetti, G. (2025). Selective agreement, not sycophancy: investigating opinion dynamics in LLM interactions. EPJ Data Science. https://doi.org/10.1140/epjds/s13688-025-00579-1
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.