Thesis subject proposed within PR[AI]RIE-PSAI – Fellows call 2026

Thesis supervisor: Nicolas Baumard (DR CNRS, Institut Jean Nicod, DEC, ENS-PSL)

Co-supervisor: Edgar Dubourg (Post-doctoral researcher, Institut Curie, PSL)

1. Context, positioning, and objective

Narratives rank among the most effective tools for learning and behavioral change. Research across psychology (Bandura et al., 1961; Bruner, 1991; Mar et al., 2021), economics (DellaVigna & La Ferrara, 2015; Riley et al., 2017), and public health (Murphy et al., 2011; Singhal, 2004) has consistently demonstrated the power of stories to educate, persuade, and motivate. A recent meta-analysis covering more than 33,000 participants found that people recall and comprehend narrative texts better than expository ones (Mar et al., 2021). In educational contexts, narratives built around clear goals, obstacles, and resolutions facilitate both comprehension and retention (Thorndyke, 1977; Bullock et al., 2021).

A major limitation of existing narrative-based interventions, however, is that they typically adopt a ‘one-size-fits-all’ approach: the same story is delivered to every learner, regardless of their motivational profile. This represents a missed opportunity. Research in health communication has shown that personalized messages (i.e., those aligned with recipients’ individual characteristics) outperform generic ones on measures of attention, comprehension, retention, and intention to act (Noar et al., 2007; Kreuter et al., 1999; Avenel et al., 2026a). A meta-analysis of 702 experimental tests confirmed that matching messages to recipients’ motivational orientations reliably increases persuasion (Joyal-Desmarais et al., 2022; Avenel et al., 2026b; see also: Hirsh et al., 2012; Matz et al., 2017).

Yet personalization has so far been studied primarily at the level of short persuasive messages, not full narratives. The present thesis addresses this gap by developing an AI-driven framework for personalizing narratives. The core scientific question is: Can we use large language models (LLMs) to systematically tailor narrative content to individual motivational profiles, and does this personalization improve learning outcomes in educational settings?

2. Theoretical framework: from motivational alignment to narrative personalization

This thesis builds on two converging lines of research. The first concerns the cognitive mechanisms that make narratives compelling. We adopt the vicarious learning framework (Olsson et al., 2020; Olsson & Phelps, 2007), which hypothesizes that stories engage audiences because they reproduce the informational structure that enables humans to learn from observing others: an intentional agent pursuing a goal, encountering obstacles, acting under uncertainty, and experiencing causal consequences. Narratives thus function as simulated environments for vicarious problem-solving (Dore, 2020; Dubourg & Baumard, 2023; Meltzoff et al., 2012).

The second line of research concerns the empirical basis for matching motivational content to individual profiles. Work on the attractiveness of stories has identified a list of motivational content features, defined as psychologically meaningful cues embedded in messages that activate specific motivational systems (e.g., curiosity, threat, parental care, disgust, social bonding; Sperber & Hirschfeld, 2004; Dubourg et al., 2024). In a large-scale study of anti-smoking advertisements (N = 2,622 daily smokers, 7,866 ad evaluations), Avenel et al. (2026b) demonstrated that the persuasive impact of these content features depends on person-specific alignment between motivational cues and recipients’ personality traits.

In this thesis, we will therefore test the motivational alignment hypothesis: the hypothesis that the match between a protagonist’s goals and the observer’s own motivational priorities increase narrative interest and behavioral change. Across two preregistered experiments supervised by the advisors of this project (N = 953), motivational alignment reliably predicted interest in a story, while gender-, race-, and age-similarity between the audience and the protagonist did not. This finding reframes narrative engagement: what matters is not who the protagonist resembles, but what goals they pursue relative to the observer’s concerns (Maunoir et al., 2026; see also Maccoby & Wilson, 1957; Dore, 2022). For personalization, this implies that effective tailoring should target the motivational content of narratives, not merely surface-level demographic features.

3. Scientific approach and work programme

The thesis is structured around three complementary work packages (WPs) that progressively move from computational tool-building to large-scale validation and applied intervention. 

WP1. Computational: LLM-based annotation and generation of motivationally tailored narratives

Building on validated methods for using LLMs to annotate cultural and narrative content at scale (Dubourg et al., 2024, 2025; Gilardi et al., 2023; Bongini et al., 2023), this WP develops a computational pipeline to (a) automatically annotate educational narratives for their motivational content features (curiosity, threat, parental care, social bonding, etc.), (b) score their degree of motivational alignment with a given learner profile, and (c) generate or adapt narratives to maximize alignment for specific motivational profiles. The annotation method follows the approach validated in Dubourg et al. (2025, 2024), where LLMs were used to systematically score over 65,000 works across multiple dimensions using structured scales, with demonstrated convergent and discriminant validity against human ratings and alternative computational methods. 

Here, the LLM will score each narrative segment for the presence and intensity of motivational content features, using the taxonomy developed by Avenel et al. (2026b). This WP will also develop a brief motivational profiling questionnaire for learners, drawing on the Big Five personality inventory and motivational priority scales (Dubourg et al., 2024; Marengo et al., 2021; Del Giudice, 2024).

WP2. Observational: large-scale validation of motivational alignment in educational content

This WP tests whether motivational alignment predicts learning-related engagement in large, ecologically valid corpora of educational narratives. Using the annotation pipeline from WP1, we will score motivational content features across corpora of existing narratives. We will then examine whether content whose motivational profile better matches the known preferences of its target audience is associated with higher engagement metrics (completion rates, comprehension scores, self-reported interest). A key test will model learner drop-off within stories as a function of the preceding section’s motivational alignment score. Our prediction: sections with higher motivational alignment will be associated with lower abandonment rates and better downstream performance. Cross-domain comparisons (e.g., formal education, therapeutic patient education) will assess the generalizability of the framework.

WP3. Applied: AI-personalized narratives for educational outcomes

The final WP applies the framework to real educational settings. Using the LLM-based generation capabilities from WP1 and the validated alignment–engagement associations from WP2, we will conduct randomized controlled trials in which learners receive either (a) a standard educational narrative, (b) a narrative personalized to their motivational profile, or (c) a narrative personalized to a mismatched profile (active control). Primary outcomes will include comprehension, retention (tested after a delay), and intention to act on the information. We will test this in educational contexts broadly defined, including formal learning (e.g., science education modules) and therapeutic education (e.g., patient education for chronic disease self-management, a domain where narrative-based approaches are increasingly used but rarely personalized; Avenel et al., 2026a). This applied WP directly leverages the AI tools developed in WP1, demonstrating how LLMs can serve not only as analytical instruments but as engines for scalable, evidence-based narrative personalization (see also Costello et al., 2024; Altay et al., 2023).

4. Relevance to PR[AI]RIE-PSAI

This project lies at the intersection of artificial intelligence and cognitive science. It uses AI (LLMs) not as a black box but as a scientifically validated instrument for measuring and manipulating psychological variables in narratives, building on methods already published by members of this team. It combines fundamental research on narrative cognition with applied objectives in educational personalization, directly contributing to the PR[AI]RIE-PSAI mission of developing AI for societal benefit. The thesis will be conducted at Institut Jean Nicod (DEC, ENS-PSL), benefiting from its experimental facilities and its interdisciplinary environment in cognitive science, and in close collaboration with Institut Curie (PSL).

References

Altay, S., Hacquin, A.-S., Chevallier, C., & Mercier, H. (2023). Information delivered by a chatbot has a positive impact on COVID-19 vaccines attitudes and intentions. Journal of Experimental Psychology: Applied, 29, 52–62.

Avenel, J., Bidant, E., Pasquereau, A., Guignard, R., Nguyen-Thanh, V., Chevallier, C., Nettle, D., & Dubourg, E. (2026b). Motivational content features improve tailoring of anti-smoking advertisements. Manuscript under review.

Avenel, J., Chevallier, C., & Dubourg, E. (2026a). Beyond one-size-fits-all: Personalising health communication to drive real behaviour change. Quarterly Medical Review (accepted manuscript).

Bandura, A., Ross, D., & Ross, S. A. (1961). Transmission of aggression through imitation of aggressive models. Journal of Abnormal and Social Psychology, 63, 575–582.

Bongini, P., Becattini, F., Bagdanov, A. D., & Del Bimbo, A. (2023). Is GPT-3 a good data annotator? Proceedings of the ACM Conference on Human Factors in Computing Systems.

Bruner, J. (1991). The narrative construction of reality. Critical Inquiry, 18, 1–21.

Bullock, O. M., Shulman, H. C., & Huskey, R. (2021). Narratives are persuasive because they are easier to understand. Frontiers in Communication, 6, 719615.

Costello, T. H., Pennycook, G., & Rand, D. G. (2024). Durably reducing conspiracy beliefs through dialogues with AI. Science, 385, eadq1814.

Del Giudice, M. (2024). The motivational architecture of emotions. In The Oxford Handbook of Evolution and the Emotions. Oxford University Press.

DellaVigna, S., & La Ferrara, E. (2015). Economic and social impacts of the media. Handbook of Economic Growth, 2, 639–699.

Dore, R. A. (2020). The cognitive foundations of learning from stories. Trends in Cognitive Sciences, 24, 521–523.

Dore, R. A. (2022). The effect of character similarity on children’s learning from fictional stories. Journal of Experimental Child Psychology, 214, 105310.

Dubourg, E., & Baumard, N. (2024). Does fiction impact belief? A critical view. In F. Lavocat, A. James, & A. Kubo (Eds.), The Routledge Handbook of Fiction and Belief. Routledge.

Dubourg, E., Thouzeau, V., & Baumard, N. (2024). A step-by-step method for cultural annotation by LLMs. Frontiers in Artificial Intelligence, 7, 1365508.

Dubourg, E., Thouzeau, V., Borredon, Q., & Baumard, N. (2025). Quantifying and explaining the rise of fiction. Evolutionary Human Sciences, 7, e20.

Gilardi, F., Alizadeh, M., & Kubli, M. (2023). ChatGPT outperforms crowd-workers for text-annotation tasks. Proceedings of the National Academy of Sciences, 120, e2305016120.

Hirsh, J. B., Kang, S. K., & Bodenhausen, G. V. (2012). Personalized persuasion: Tailoring appeals to recipients’ personality traits. Psychological Science, 23, 578–581.

Joyal-Desmarais, K., Scharmer, A. K., Madzelan, M. K., See, J. V., Rothman, A. J., & Snyder, M. (2022). Appealing to motivation to change attitudes, intentions, and behavior: A systematic review and meta-analysis of 702 experimental tests of the effects of motivational message matching on persuasion. Psychological Bulletin, 148, 465–517.

Kreuter, M. W., Strecher, V. J., & Glassman, B. (1999). One size does not fit all: The case for tailoring print materials. Annals of Behavioral Medicine, 21, 276–283.

Maccoby, E. E., & Wilson, W. C. (1957). Identification and observational learning from films. Journal of Abnormal and Social Psychology, 55, 76–87.

Mar, R. A., Li, J., Nguyen, A. T. P., & Ta, C. P. (2021). Memory and comprehension of narrative versus expository texts: A meta-analysis. Psychonomic Bulletin & Review, 28, 732–749.

Marengo, D., Davis, K. L., Gradwohl, G. Ö., & Montag, C. (2021). A meta-analysis on individual differences in primary emotional systems and Big Five personality traits. Scientific Reports, 11, 7453.

Matz, S. C., Kosinski, M., Nave, G., & Stillwell, D. J. (2017). Psychological targeting as an effective approach to digital mass persuasion. Proceedings of the National Academy of Sciences, 114, 12714–12719.

Maunoir, P., Chambon, V., Baumard, N., Thouzeau, V., & Dubourg, E. (2026). Motivational alignment, not demographic similarity, predicts narrative interest. Manuscript under review.

Meltzoff, A. N., Waismeyer, A., & Gopnik, A. (2012). Learning about causes from people. Developmental Psychology, 48, 1215–1228.

Murphy, S. T., Frank, L. B., Chatterjee, J. S., & Baezconde-Garbanati, L. (2011). Involvement, transportation, and persuasion in an entertainment-education narrative. Journal of Communication, 61, 407–431.

Noar, S. M., Benac, C. N., & Harris, M. S. (2007). Does tailoring matter? Meta-analytic review of tailored print health behavior change interventions. Psychological Bulletin, 133, 673–693.

Olsson, A., Knapska, E., & Lindström, B. (2020). The neural and computational systems of social learning. Nature Reviews Neuroscience, 21, 197–212.

Olsson, A., & Phelps, E. A. (2007). Social learning of fear. Nature Neuroscience, 10, 1095–1102.

Riley, E., Soko, D., & Terrell, C. (2017). The impact of role models on academic performance. Journal of Human Resources, 52, 919–949.

Singhal, A. (Ed.). (2004). Entertainment-education and social change: History, research, and practice. Lawrence Erlbaum Associates.

Sperber, D., & Hirschfeld, L. A. (2004). The cognitive foundations of cultural stability and diversity. Trends in Cognitive Sciences, 8, 40–46.

Thorndyke, P. W. (1977). Cognitive structures in comprehension and memory of narrative discourse. Cognitive Psychology, 9, 77–110.