Persuasion is as old as human relationships. Influencing beliefs and behavior has always been part of social life.
Aristotle already identified three fundamental dimensions of persuasion:
- Ethos – the credibility of the speaker
- Pathos – the emotional dimension of influence
- Logos – the rational structure of argumentation
Over time, technological developments — especially in advertising — have changed the scale, personalization, and automation of persuasive communication.
The rise of the internet further expanded scale and data collection, increasing the precision of personalization.
What is different today is not that persuasion exists.
It is how generative AI reshapes the way persuasion can be exercised — and how effective it can become.
What Changes with GenAI
1. From Selective Personalization to Generative Personalization
Personalization itself is not new. Online advertising has long relied on behavioral data to tailor content. Traditionally, however, personalization meant selecting from predefined variants.
With GenAI, personalization becomes generative.
It no longer simply selects a message — it constructs one.
GenAI systems can:
- adapt tone and language style
- reshape arguments
- reflect values or personality traits
- adjust framing dynamically
Research by Matz et al. (2024)1 shows that LLM-generated messages personalized on psychological traits can be significantly more persuasive than non-personalized ones.
The difference is not only quantitative.
It is qualitative.
2. Conversational and Adaptive Persuasion
A study published in Nature Human Behaviour (Salvi et al., 2025)2 found that GPT-4, when provided with personalized information about an interlocutor, could outperform humans in structured online debates.
This is not merely about producing polished text.
It is about:
- maintaining argumentative coherence
- adapting in real time
- modulating tone and emotional framing
This moves persuasion beyond static targeting.
It becomes automated dialogical influence.
3. The Opacity of Influence
GenAI can operate without making its persuasive intent visible.
It may not be clear:
- that content is AI-generated
- that the message has been personalized
- which data were used to shape it
When personalization becomes invisible, asymmetry increases between those who influence and those being influenced.
And it is within this asymmetry that persuasion may begin to approach manipulation.
Persuasion and Manipulation: A Fragile Boundary
Not every form of influence is manipulation.
In its classical sense, persuasion assumes certain conditions: a recognizable attempt to convince, the opportunity to evaluate arguments, and a decision that ultimately remains with the individual.
Manipulation, by contrast, operates less transparently. It does not simply persuade — it steers behavior by reducing or bypassing the possibility of an informed choice.
Susser, Roessler, and Nissenbaum (2019)3 describe online manipulation as a form of “covert” influence that exploits vulnerabilities and informational asymmetries to shape decisions without full awareness.
The boundary is subtle. With generative AI, it becomes even more complex.
When personalization is invisible, when messages adapt in real time to our reactions, and when it is unclear who — or what — is speaking, asymmetry increases.
The difference lies not only in our ability to notice what is happening, but also in the intention and responsibility of those who choose to use GenAI to persuade.
Understanding influence helps you stay aware — and stay in control.
Further Reading
If you’d like to explore some of the ideas discussed in this article:
Rhetoric – Aristotle
The classical foundation of persuasive theory, introducing ethos, pathos, and logos — concepts that still shape how we understand influence today.
Online Manipulation: Hidden Influences in a Digital World
A philosophical and normative analysis of online manipulation and its implications for autonomy and informed choice. https://georgetownlawtechreview.org/online-manipulation-hidden-influences-in-a-digitalworld/GLTR-01-2020/
The Potential of Generative AI for Personalized Persuasion
Experimental research showing that LLM-generated messages personalized on psychological traits can be more persuasive than non-personalized ones. https://www.nature.com/articles/s41598-024-53755-0
AI Models Surpass Humans in Online Persuasive Debates
A study published in Nature Human Behaviour demonstrating that advanced AI models can outperform humans in persuasive online debates when personalization is applied. https://www.nature.com/articles/s41562-025-02194-6
Influence: Science and Practice – Robert Cialdini
A foundational work on the psychology of influence and the principles that shape human decision-making.
- S. C. Matz, J. D. Teeny, S. S. Vaid, H. Peters, G. M. Harari e M. Cerf, The potential of generative AI for personalized persuasion at scale, in Scientific Reports, vol. 14, 2024, p. 4692, https://doi.org/10.1038/s41598-024-53755-0. ↩︎
- Salvi, F., Horta Ribeiro, M., Gallotti, R., & West, R. (2025). On the conversational persuasiveness of GPT-4. Nature Human Behaviour. https://doi.org/10.1038/s41562-025-02194-6 ↩︎
- Susser, D., Roessler, B., & Nissenbaum, H. (2019). Online Manipulation: Hidden Influences in a Digital World. Georgetown Law Technology Review, 4(1), 1-45.
https://georgetownlawtechreview.org/online-manipulation-hidden-influences-in-a-digital-world/GLTR-01-2020/ ↩︎