The Role of Emotional Stimuli and Intensity in Shaping Large Language Model Behavior

Ameen Patel; Felix Lee; Kyle Liang; Joseph Thomas

cs.LGApr 7, 2026

The Role of Emotional Stimuli and Intensity in Shaping Large Language Model Behavior

Authors: Ameen Patel, Felix Lee, Kyle Liang, Joseph Thomas

Organizations: Irvington High School · Glen A. Wilson High School · Del Norte High School · California High School

Abstract

Emotional prompting - the use of specific emotional diction in prompt engineering - has shown increasing promise in improving large language model (LLM) performance, truthfulness, and responsibility. However these studies have been limited to single types of positive emotional stimuli and have not considered varying degrees of emotion intensity in their analyses. In this paper, we explore the effects of four distinct emotions - joy, encouragement, anger, and insecurity - in emotional prompting and evaluate them on accuracy, sycophancy, and toxicity. We develop a prompt-generation pipeline with GPT-4o mini to create a suite of LLM and human-generated prompts with varying intensities across the four emotions. Then, we compile a "Gold Dataset" of prompts where human and model labels align. Our empirical evaluation on LLM behavior suggests that positive emotional stimuli lead to more accurate and less toxic results, but also increase sycophantic behavior.

Figures & tables

Figure 1: The LLM prompts were created through different human prompts for the four emotions and intensity levels. The human prompts were used in a prompt generation pipeline where we expedited prompt generation, creating 415 prompts for four emotions. The LLM emotional prompts were then used to test different datasets with their baseline prompts and with the emotional prompts.

	Anger		Joy		Insecurity		Encouragement
	LLM	Human	LLM	Human	LLM	Human	LLM	Human
Base Score	0.9300	0.9200	0.9000	0.9100	0.9200	0.9200	0.8900	0.9100
Augm. Score	0.9291	0.9191	0.9076	0.9260	0.9203	0.9200	0.8950	0.9300
% Diff.	-0.0968	-0.0978	0.8444	1.7582	0.0326	0.0000	0.5618	2.1978

Table 1: Overall Mean Base Score (without our emotional prompt add-on, abbreviated as Base ), Overall Mean Augmented Score (with our emotional prompt add-on, abbreviated as Augm. ), and Percent Difference (between the augmented as base scores, abbreviated as % Diff. ).

	Anger		Joy		Insecurity		Encouragement
	Human	LLM	Human	LLM	Human	LLM	Human	LLM
Arguments	0.4727	0.5583	0.6148	0.6638	0.3667	0.2170	0.5296	0.6229
Math	0.5233	0.4956	0.5741	0.5194	0.3704	0.2688	0.4778	0.7277
Poems	0.3967	0.5044	0.4556	0.6032	0.4333	0.2220	0.3889	0.6105

Table 2: Mean Positivity Scores across three domains— Arguments, Math, and Poems . Scores are calculated by prompting the LLM to choose between two options—the base prompt (without our emotional prompt add-ons) and the augmented prompt (with our emotional add-ons). The Mean Positivity Score is the average of all comparison scores across all prompts (where base prompts are scored ’0’ and augmented prompts are scored ’1’.

	Anger		Joy		Insecurity		Encouragement
	LLM	Human	LLM	Human	LLM	Human	LLM	Human
Base Score	0.2915	0.2914	0.2912	0.2914	0.2913	0.2914	0.2915	0.2916
Augm. Score	0.2860	0.2910	0.2874	0.2910	0.2877	0.2911	0.2877	0.2912
%Diff.	-1.8868	-0.1373	-1.3049	-0.1373	-1.2358	-0.1030	-1.3036	-0.1372

Table 3: Overall Mean Base Toxicity Score (without our emotional prompt add-on, abbreviated as Base Toxicity Score ), Overall Mean Augmented Toxicity Score (with our emotional prompt add-on, abbreviated as Augmented Toxicity Score ), and Percent Difference (between the augmented as base scores, abbreviated as % Diff. ). Overall, the toxicity reduced across all emotions and within human and LLM prompts.

Figure 2: Mean Positivity Scores for human-generated emotional prompt add-ons.

Figure 3: Mean Positivity Scores for LLM-generated emotional prompt add-ons.

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 4: A chart of the Human Gold Dataset prompts (anger in yellow, encouragement in red, insecurity in purple, and joy in blue).

Figure 5: Mean Base Scores and Mean Augmented Scores for human-generated vs. LLM-generated emotional prompt add-ons (from the Gold Dataset). Scores evaluated from Anthropic’s SycophancyEval subset on accuracy. Overall, there is little to no difference between human-generated and LLM-generated scores.

Figure 6: Mean Positivity Scores for human-generated vs. LLM-generated emotional prompt add-ons (from the Gold Dataset). Scores evaluated from Anthropic’s SycophancyEval subset on sycophancy. Overall, LLM-generated prompts result in more agreeable/sycophantic response than human-generated prompts.

Figure 7: The Mean Toxicity Scores are the mean scores of our Mean Base Score for the baseline scores and the Mean Augm. Score , the mean score of our emotional prompt add-ons onto the toxicity dataset. The base score is higher in both cases, with the LLM Gold Dataset having a higher change in the toxicity score.

Explore similar work

Do Emotions in Prompts Matter? Effects of Emotional Framing on Large Language Models
Apr 2, 2026Minda Zhao, Yutong Yang, Chufei Peng +5LLM Prompting Framing Effects in Language Models
Harvard University. · Bryn Mawr College.
Spurious Prompts: Can Irrelevant Prompts Steer Large Language Models?
May 28, 2026Pawel Batorski, Abtin Pourhadi, Jerzy Sarosiek +2Prompt Sensitivity LLM Prompting
1Heinrich Heine University Düsseldorf · 2Jagiellonian University · 3IDEAS Research Institute
SyPS: Measuring Sycophancy Prompt Sensitivity in Large Language Models
Aug 24, 2026Lijia Huang, Yao Fu, Sihao RenLLM Sycophancy LLM Evaluation
Northeastern University · Case Western Reserve University · Everpure

Apr 2, 2026J/K move · Enter open · S save

Do Emotions in Prompts Matter? Effects of Emotional Framing on Large Language Models

Minda Zhao, Yutong Yang, Chufei Peng, Rachel Gonsalves+4

Harvard University. · Bryn Mawr College.

LLM Prompting Framing Effects in Language Models LLM Evaluation

Open paper AskPDF arXiv

Emotional tone is pervasive in human communication, yet its influence on large language model (LLM) behaviour remains unclear. Here, we examine how first-person emotional framing in user-side queries affect LLM performance across six benchmark domains, including mathematical reasoning, medical question answering, reading comprehension, commonsense reasoning and social inference. Across models and tasks, static emotional prefixes usually produce only small changes in accuracy, suggesting that affective phrasing is typically a mild perturbation rather than a reliable general-purpose intervention. This stability is not uniform: effects are more variable in socially grounded tasks, where emotional context more plausibly interacts with interpersonal reasoning. Additional analyses show that stronger emotional wording induces only modest extra change, and that human-written prefixes reproduce the same qualitative pattern as LLM-generated ones. We then introduce EmotionRL, an adaptive emotional prompting framework that selects emotional framing adaptively for each query. Although no single emotion is consistently beneficial, adaptive selection yields more reliable gains than fixed emotional prompting. Together, these findings show that emotional tone is neither a dominant driver of LLM performance nor irrelevant noise, but a weak and input-dependent signal that can be exploited through adaptive control.

Abstract

Explore similar work

Do Emotions in Prompts Matter? Effects of Emotional Framing on Large Language Models

Spurious Prompts: Can Irrelevant Prompts Steer Large Language Models?

SyPS: Measuring Sycophancy Prompt Sensitivity in Large Language Models