The labels introduced on social networks to warn users that a piece of content is potentially false or misleading have a limited effectiveness. This method’s shortcomings could lie in the way our memory works: it progressively dissociates the memory of the incriminated content from that of the warning label attached to it – a phenomenon called the “sleeper effect”. Once this dissociation has occurred, if the content in question reappears on social networks without a warning, it regains its strength of conviction because the exposed users would no longer associate it with the label they had seen at the first exposure.
Some research has shown that a warning label placed before misleading content would be more robustly remembered by users and would thus be more effective. But here again, this method has limitations: the effectiveness of the warning could be weakened if the exposed users are inclined to believe the content in question. In general, it has indeed been observed that individuals can be driven to believe false information if it is consistent with their political leanings.
In their study, Grady and her colleagues tested the effectiveness of warning labels by implementing them in various ways to identify their limitations. The results of the study confirm that a warning shown before the appearance of false content is more effective in the short term than one shown during or after its appearance. However, this comparative advantage disappears after only two weeks. Thus, the “sleeper effect” impacts all types of warnings: over time, the individuals tested are no longer able to associate the content in question with the warning that was attached to it, whether the latter is placed before, during, or after the appearance of the content.
This study further shows that participants tended to deem more credible false content that supports their political leanings, even if this content was accompanied by a warning label informing them of its falsehood at the time it was presented. This effect was observed among participants regardless of their political positioning – i.e., conservative or liberal, in the American context of this study.
These results confirm the need to multiply the devices and methods to fight against misinformation on social networks, as warning labels alone are not sufficient to discredit false or misleading content.