Recommendation algorithms are designed to make enormous streams of online content easier to navigate. Yet a feed that becomes highly accurate at predicting what a person already likes can also feel repetitive, crowded or difficult to escape. New research suggests that the way users perceive recommendation systems matters for information fatigue, and that novelty may play a particularly important role.
A peer-reviewed study published in Scientific Reports on 3 October 2026 surveyed 425 adult social media users and examined how three perceived features of algorithmic recommendations, accuracy, transparency and novelty, were associated with information fatigue. Rather than treating fatigue as a simple consequence of seeing too much content, the researchers tested whether perceived information overload and perceived information narrowing helped explain how recommendation features related to fatigue.
The clearest pattern concerned novelty. Users who perceived their recommendations as more novel tended to report less information overload, less information narrowing and less information fatigue. In a complementary importance analysis, recommendation novelty had the largest absolute total association with fatigue among the factors examined. Accuracy and transparency showed a more complicated picture: features normally regarded as desirable were also associated with cognitive appraisals that, in turn, were related to greater fatigue.
Looking beyond whether an algorithm is accurate
Much of the discussion around recommender systems focuses on relevance. If an algorithm can correctly identify what a user is likely to watch, read or click, it is usually considered to be performing well. The new study argues that this view is incomplete because users experience recommendations as an ongoing information environment rather than as isolated predictions.
The researchers distinguished three perceived characteristics. Recommendation accuracy captured whether users felt the system supplied content aligned with their interests. Algorithmic transparency concerned whether the recommendation process felt understandable or visible. Recommendation novelty reflected whether the system exposed users to content that felt new rather than simply repeating familiar material.
They then examined two appraisals of the resulting information environment. Perceived information overload refers to the sense that the quantity or complexity of information exceeds a person’s processing capacity. Perceived information narrowing describes the feeling that recommendations restrict the range of information encountered, potentially creating a repetitive or constricted feed. Both were expected to contribute to information fatigue.
The study combined three analytical approaches
The analysis used survey responses from 425 adult social media users. The researchers employed partial least squares structural equation modelling, or PLS-SEM, to estimate relationships among the perceived recommendation features, the two cognitive appraisals and information fatigue. This allowed them to examine both direct associations and indirect pathways through overload and narrowing.
They complemented the structural model with cIPMA, an importance-performance analysis designed to compare the relative total associations of the different factors with information fatigue. A third method, fuzzy-set qualitative comparative analysis, or fsQCA, was used to examine combinations of conditions rather than assuming that fatigue must arise through one single linear pathway.
This combination matters because recommendation experiences may be configurational. A user could experience high fatigue because several unfavourable conditions occur together, while another user may reach a similar outcome through a different combination. The fsQCA component was intended to identify those alternative patterns.
Novelty showed the most consistently favourable pattern
Perceived recommendation novelty was negatively associated with both information overload and information narrowing. It was also negatively associated with information fatigue itself. In practical terms, participants who felt that their feeds introduced genuinely new material tended to feel less overwhelmed, less confined by the recommendation system and less fatigued by the information environment.
The cIPMA analysis reinforced this result. Recommendation novelty had the largest absolute total association with information fatigue among the variables examined. Perceived information overload ranked next, followed by perceived information narrowing. This does not establish that increasing novelty will necessarily reduce fatigue, but it identifies novelty as a prominent feature within the study’s observational model.
The finding also highlights a potential tension in recommendation design. Systems optimised heavily for past preferences may become increasingly good at serving familiar material. From a prediction perspective, that can look successful. From the user’s perspective, however, a feed with too little novelty may become repetitive or psychologically tiring.
Accuracy was not simply protective
Perceived recommendation accuracy was positively associated with both information overload and information narrowing. Those two appraisals were themselves positively associated with information fatigue. Accuracy therefore showed indirect positive associations with fatigue through both pathways.
This does not mean that accurate recommendations are inherently harmful. An accurate system may deliver a large volume of highly relevant material, increasing the amount a user feels compelled to process. It may also repeatedly infer a narrow set of interests from past behaviour, making the resulting feed feel increasingly concentrated around the same themes. The study measures users’ perceptions and associations among those perceptions, so these mechanisms remain interpretations rather than experimentally demonstrated causal effects.
The result is nevertheless important because it challenges a one-dimensional definition of recommendation quality. Accuracy can coexist with overload. Relevance can coexist with narrowing. A system can therefore succeed at predicting preferences while still producing an information experience that some users find exhausting.
Transparency also had an unexpected association
Perceived algorithmic transparency was positively associated with perceived information overload and was indirectly associated with information fatigue through that appraisal. Transparency was not presented as uniformly detrimental, but the pattern suggests that making recommendation processes more perceptible does not automatically eliminate cognitive strain.
One possibility is that awareness of how content is being selected adds another layer of information for users to evaluate. Another is that people who pay closer attention to recommendation processes become more conscious of the volume of algorithmically selected material around them. The current design cannot determine which explanation, if either, is correct.
The broader lesson is that transparency should not be treated as a single design switch whose psychological effects are guaranteed to be positive. The form, timing and cognitive demands of explanations may matter. Future experimental work would be needed to test how different transparency designs influence fatigue while preserving informed user control.
Fatigue emerged through more than one configuration
The fsQCA analysis found that no individual antecedent was a necessary condition for high information fatigue. In other words, there was no single factor that had to be present whenever fatigue was high.
Instead, the researchers identified two configurations associated with high information fatigue and four configurations associated with non-high fatigue. This supports the idea that users can arrive at similar experiences through different combinations of recommendation features and cognitive appraisals.
That result is useful for understanding why simple platform rules may not work equally well for everyone. Reducing one source of strain may help some users but have little effect when another combination of conditions is driving their fatigue. Conversely, multiple different recommendation environments may allow users to avoid high fatigue.
Why recommendation diversity may matter
Algorithmic feeds must balance several objectives. Users generally expect relevant content, but relevance is not identical to repetition. A recommendation system can potentially remain useful while also introducing material outside a person’s most predictable interests.
The study suggests that novelty and diversity deserve greater attention alongside conventional accuracy metrics. For platform designers, this could mean evaluating whether optimisation systems repeatedly exploit known preferences at the expense of exploration. It could also mean giving users more control over whether they want familiar, diverse or exploratory recommendations at a particular moment.
These possibilities should be treated as design hypotheses rather than proven interventions. The research did not randomly alter users’ feeds and then measure subsequent fatigue. It examined how adults perceived the recommendation systems they were already using.
The findings are associations, not proof of causation
The study’s cross-sectional survey design is its most important limitation. Because recommendation perceptions and fatigue were measured rather than experimentally manipulated, the direction of influence cannot be established. People who are already fatigued may perceive their feeds differently, and unmeasured characteristics could influence both recommendation perceptions and fatigue.
The sample of 425 adults also limits how confidently the results can be generalised across countries, age groups, platforms and patterns of social media use. Recommendation systems differ substantially between short-video services, social networks, news feeds and other platforms. The psychological meaning of novelty or transparency may likewise differ across those settings.
PLS-SEM can estimate complex relationships among latent constructs, but a statistically supported path remains an association. The fsQCA configurations similarly identify combinations linked to outcomes within the dataset rather than deterministic recipes for producing or preventing fatigue.
Even with these constraints, the research adds an important dimension to debates about personalised feeds. Recommendation quality is not only about whether a system correctly predicts the next click. Users must also live inside the stream of information that those predictions create. In this study, novelty stood out as the recommendation feature most consistently associated with a less fatiguing experience, while accuracy and transparency showed pathways that were more complicated than a simple assumption that more is always better.
Source Information
Study: Yao, P., Xia, S., Zhang, X. et al. Perceived algorithmic recommendation features and information fatigue among social media users.
Journal: Scientific Reports
Published: 3 October 2026
DOI: 10.1038/s41598-026-72901-4
Study type: Cross-sectional survey of 425 adult social media users using PLS-SEM, cIPMA and fsQCA.
