Harms arising from targeted online 'nudging'

Exploring what can be done to ensure that the influences of online ‘nudges’ do not become manipulative or harmful to vulnerable groups


About a decade ago, the concept of ‘nudging’ was popularised by behavioural scientists and economists Richard Thaler and Cass Sunstein and since then much research has been done into the modalities and ethics of ‘nudging’ in the real world. But nudging is increasingly being used in the online sphere as well and there are reasons to suspect that online nudging is importantly different from what has previously taken place offline. For example, while offline nudges have typically been ‘one-size-fits-all’ changes to choice architectures, the use of AI and data-driven techniques have allowed online nudges to be increasingly personalised. The key objective of this research is to determine the potentially harmful effects of current practices of targeted online nudging on the decisional privacy and autonomy of individuals, with a particular focus on vulnerable populations, for example, those suffering from mental health conditions.

Explaining the science

Nudge theory proposes ways in which behaviour can be indirectly influenced by altering the environment, or ‘choice architecture’, in different ways, usually to trigger some kind of desired behavioural outcome by exploiting our natural cognitive biases (Kahneman & Tversky, 1986). The idea is, in a sense, nothing new - advertisers have long known that by hijacking our attention through pictures and words they can influence our decision-making. What is new, however, are the various ways in which AI and data-driven methods are being used to nudge people online.

In particular, the effectiveness of online nudging has increased with three key technological advancements:

  • The type and amount of personal data that can be collected
  • Who holds and has access to that data; and, perhaps most importantly
  • How that data can be used by those who can access it

With these changes it is now possible for governments and businesses to collect information about small groups and even individuals, and to model and predict their beliefs, preferences, and behaviour, allowing them to target interventions (Hu et al. 2007, Kosinksi et al. 2012). There is also mounting evidence that shows these personalised interventions are effective at influencing consumer behaviour (e.g. Matz et al. 2017; Goldfarb 2014; Noar et al. 2007).

Project aims

The technological advancements in AI and data science that have facilitated current practices of personalisation in online nudging present new opportunities to make interventions more effective, but at the same time their use of more fine-grained information about individuals also raises serious risks. In particular, there seem to be new risks to both our decisional privacy - the right not to be accessed or interfered with in our decisions and actions by third parties, unless this influence was otherwise consented to (Lanzig 2018) - and our autonomy - the right and ability to freely reflect on and decide on one’s values, actions, and behaviour, and to act on those choices. The key objective of this research is to delineate these new and potentially harmful effects of personalised online nudging, especially for individuals with known psychological vulnerabilities.

To achieve this objective, this project will proceed in two steps:

  1. Understand the different methods and modalities that are being used in online nudging, as compared with those used in traditional ‘offline’ nudging (e.g. object properties and placement), and the evidence of their effectiveness
  2. Identify and analyse the new risks to decisional privacy and autonomy that are brought about by their use.

This project will focus on the potential effects of online nudges on those with mental health conditions that, by definition, affect one’s capacity to reason, one’s ability to make decisions, and other cognitive capacities that are core to one’s ability to self-govern. Because of their conditions these individuals are especially vulnerable to external forms of influence, and there is some evidence that these vulnerabilities are already being targeted. For example, Sax, Helberger, and Bol (2018) discuss the increasing number of mental health applications that merge health and commercial content - they attract users interested in improving their health and then, having captured their attention, target advertisements intended to serve commercial interests.

Online nudging in mental health applications also presents positive opportunities, for example, to help encourage people to make better choices that align with their own values and goals, but the countervailing risks to one’s decisional privacy and autonomy need to be made clear.


The outcomes of this research will be a philosophical paper outlining the relevant differences between traditional nudging and targeted online nudging, including their distinct methods and modalities, as well as an analysis of the significance of these differences for the decisional privacy and autonomy of those being targeted with nudges. To make the potential harms clear, the paper will consider the risks to psychologically vulnerable populations in particular.


In addition to an academic research publication, this project also aims to support informed and evidence-based debate on the most effective ways to mitigate harmful effects of these emerging techniques. It will do this by informing UK policy on online harms of targeting through submissions of evidence and feeding into ongoing governmental reviews.


Contact info

[email protected]