Within Recommenders
Who Gets Seen When Rankings Control Attention?
The most influential recommendation decision is often where an item appears in a ranked list rather than whether it appears at all.
On this page
- Why ranking matters more than selection
- Position effects on clicks and viewing
- Consequences for creators and audiences
Page outline Jump by section
Introduction
Recommendation systems are often described as tools that choose what to show users. In practice, their most powerful decision is usually not selection but ranking. A platform may identify thousands of potentially relevant videos, posts, songs or products, yet only a handful fit on the first screen. The order in which those items appear determines who receives attention and who remains invisible. As a result, ranking systems function as digital gatekeepers: they decide which content reaches human awareness at scale. [Knight First Amendment Institute]knightcolumbia.orgunderstanding social media recommendation algorithmsThese algorithms are the engine that makes Facebook and YouTube what they are.Read more…
This ranking power matters because human attention is limited. Users rarely inspect every available option. Instead, they focus disproportionately on the first items they encounter, giving top-ranked content a major advantage regardless of whether lower-ranked alternatives might be equally valuable. Research across recommender systems, search engines and online marketplaces consistently finds that position alone has a strong influence on clicks and engagement. [arXiv+2arXiv]arxiv.orgA Study of Position Bias in Digital Library Recommender…February 19, 2018 — by A Collins · 2018 · Cited by 65 — “Position bias” d…
Why Ranking Matters More Than Selection
A recommendation system typically operates in two stages. First, it identifies a pool of candidates that might interest a user. Second, it ranks those candidates into a specific order. The ranking stage is where attention is allocated.
Imagine a video platform that identifies 1,000 potentially relevant videos. If only six appear on the user’s screen, then 994 recommendations effectively disappear from view. Whether an item is ranked first, fifth or fiftieth can matter far more than whether it was included in the candidate pool at all. This is why modern recommendation research places enormous emphasis on “learning to rank” methods that optimise ordering rather than simple inclusion. [Shaped]shaped.ailearning to rank for recommender systemsLearning to Rank for Recommender Systems: A Practical Guide25 Sept 2024 — In this practical guide, we dive deep into the world of l…
The logic is straightforward: users cannot evaluate everything. Ranking acts as a filter between the abundance of available information and the narrow slice that enters conscious attention. In large digital environments, that filter becomes one of the most influential forms of AI-mediated decision-making. [Knight First Amendment Institute]knightcolumbia.orgunderstanding social media recommendation algorithmsThese algorithms are the engine that makes Facebook and YouTube what they are.Read more…
Why Top Positions Receive Disproportionate Attention
The power of rankings comes from a well-documented phenomenon known as position bias. People tend to interact more with items placed near the top of a list simply because they encounter them first. Higher-ranked items receive more visual attention, more clicks and more opportunities to prove their relevance. [arXiv+2eugeneyan.com]arxiv.orgA Study of Position Bias in Digital Library Recommender…February 19, 2018 — by A Collins · 2018 · Cited by 65 — “Position bias” d…
Studies of recommendation systems have repeatedly demonstrated this effect. In one large-scale experiment involving millions of article recommendations, items placed in the highest positions received substantially more clicks than would be expected if users evaluated all options equally. The researchers concluded that rank strongly influenced user behaviour independently of actual relevance. [arXiv]arxiv.orgA Study of Position Bias in Digital Library Recommender…February 19, 2018 — by A Collins · 2018 · Cited by 65 — “Position bias” d…
Several behavioural factors contribute to this pattern:
- Limited attention: Most users review only the first few options.
- Cognitive shortcuts: People often assume higher-ranked items are better choices.
- Time pressure: Evaluating every option requires effort that many users are unwilling to spend.
- Screen constraints: Mobile devices make lower-ranked content even less visible.
Together, these factors create a steep visibility gradient in which small ranking changes can produce large differences in engagement. [arXiv+2eugeneyan.com]arxiv.orgA Study of Position Bias in Digital Library Recommender…February 19, 2018 — by A Collins · 2018 · Cited by 65 — “Position bias” d…
How Ranking Creates Winners and Losers
Because attention is concentrated at the top of ranked lists, recommendation systems do not distribute exposure evenly. A minor improvement in rank can dramatically increase visibility, while a small decline can effectively remove content from public view.
This creates a form of digital competition in which creators, publishers and businesses are not merely competing to be included but competing for placement. Being recommended in position three may generate substantial traffic; being recommended in position thirty may generate almost none. [Shaped]shaped.ailearning to rank for recommender systemsLearning to Rank for Recommender Systems: A Practical Guide25 Sept 2024 — In this practical guide, we dive deep into the world of l…
The consequences extend beyond individual pieces of content. Exposure itself generates additional engagement data. A highly ranked item receives more clicks, views and interactions, producing signals that can justify future promotion. Lower-ranked items receive fewer opportunities to demonstrate their value. Researchers describe this as a feedback loop in which visibility creates engagement and engagement creates further visibility. [CEUR Workshop Proceedings+2Machine Learning Frontiers]ceur-ws.orgCEUR Workshop ProceedingsReducing Popularity Influence by Addressing Position BiasJanuary 12, 2025 — by A Dzhoha · 2022 · Cited by 9 — As…
As a result, ranking systems can amplify early advantages. Content that gains initial exposure may continue to rise, while equally relevant alternatives struggle to attract attention because they never receive comparable opportunities. [MDPI]mdpi.comPopularity Bias in Recommender Systems: The Search for…by F Carnovalini · 2025 · Cited by 22 — Recommender systems are known to be…
What Ranking Power Means for Creators
For creators, ranking systems often determine whether work reaches an audience at all. A musician, writer, video producer or seller may technically have access to a platform’s entire user base, yet actual visibility depends heavily on recommendation placement.
This dependence creates uncertainty. Creators may observe large swings in reach without corresponding changes in quality. A modification to ranking algorithms can increase or reduce exposure across millions of users, affecting traffic, revenue and audience growth. Because rankings operate continuously and often invisibly, creators frequently adapt their behaviour to align with signals that appear to be rewarded by the system. [Lawfaremedia]Lawfareexploring tradeoffs in ranking and recommendation algorithmsResearch shows…Read more…
The result is a competitive environment shaped not only by audience preferences but also by the design choices embedded within ranking models. Decisions about what signals to prioritise—watch time, clicks, shares, purchases or other measures—can influence which types of content receive attention. [Shaped]shaped.ailearning to rank for recommender systemsLearning to Rank for Recommender Systems: A Practical Guide25 Sept 2024 — In this practical guide, we dive deep into the world of l…
What Ranking Power Means for Audiences
For audiences, rankings help solve a genuine problem: information overload. Without ranking systems, navigating enormous catalogues of content would be difficult and time-consuming. Effective rankings can surface relevant material quickly and improve discovery. [Websites]site.warrington.ufl.eduWebsitesHow Do Product Recommendations Help Consumers Search…by XS Wan — Product recommendations can benefit consumers' online product…
However, rankings also shape perceptions of what is available. Users often encounter only a tiny fraction of the content that exists on a platform. The apparent popularity, importance or relevance of information may therefore reflect ranking decisions as much as underlying quality.
This does not mean rankings are inherently manipulative. Rather, it highlights that attention is always being allocated. Every decision to place one item above another influences what users are likely to see, read, watch or purchase. Because most users never explore far beyond top recommendations, ranking systems effectively define much of their digital environment. [Knight First Amendment Institute+2Springer]knightcolumbia.orgunderstanding social media recommendation algorithmsThese algorithms are the engine that makes Facebook and YouTube what they are.Read more…
Why Platforms Try to Correct Ranking Bias
The influence of ranking has become so significant that researchers and technology companies actively study methods to measure and reduce position bias. If clicks are driven partly by placement rather than relevance, then using click data alone can reinforce distorted rankings. [Google Research+2Anne Schuth]research.google.comThe extracted bias can improve the learning-to…
To address this problem, platforms experiment with techniques such as randomised testing, bias correction methods and fairness-aware ranking approaches. The goal is not to eliminate ranking—an impossible task in information-rich environments—but to ensure that visibility reflects meaningful relevance rather than merely previous exposure. [arXiv+2arXiv]arxiv.orgOpen source on arxiv.org.
The need for these corrections reveals a central reality of modern AI-driven recommendation systems: rankings do not simply measure attention. They help create it.
The Core Takeaway
When recommendation systems control rankings, they control access to attention. The critical question is often not whether an item is recommended but where it appears. Because users concentrate on top positions, ranking systems become powerful gatekeepers that determine which creators gain audiences, which ideas circulate widely and which options remain unseen. In an online world overflowing with information, the order of a list can be as influential as the content within it. [arXiv+2Knight First Amendment Institute]arxiv.orgA Study of Position Bias in Digital Library Recommender…February 19, 2018 — by A Collins · 2018 · Cited by 65 — “Position bias” d…
eBay marketplace picks
Marketplace Samples
Example marketplace items related to this page. Use the search link to explore similar finds on eBay.
Endnotes
-
Source: shaped.ai
Title: learning to rank for recommender systems
Link: https://www.shaped.ai/blog/learning-to-rank-for-recommender-systemsSource snippet
Learning to Rank for Recommender Systems: A Practical Guide25 Sept 2024 — In this practical guide, we dive deep into the world of l...
-
Source: arxiv.org
Link: https://arxiv.org/pdf/1802.06565Source snippet
A Study of Position Bias in Digital Library Recommender...February 19, 2018 — by A Collins · 2018 · Cited by 65 — “Position bias” d...
Published: February 19, 2018
-
Source: arxiv.org
Title: arXiv A Study of Position Bias in Digital Library Recommender Systems
Link: https://arxiv.org/abs/1802.06565 -
Source: eugeneyan.com
Link: https://eugeneyan.com/writing/position-bias/Source snippet
How to Measure and Mitigate Position BiasPosition bias happens when higher positioned items are more likely to be seen and thus clicked r...
-
Source: research.google.com
Link: https://research.google.com/pubs/archive/46485.pdfSource snippet
The extracted bias can improve the learning-to...
-
Source: arxiv.org
Link: https://arxiv.org/abs/2205.06363 -
Source: mdpi.com
Link: https://www.mdpi.com/2078-2489/16/2/151Source snippet
Popularity Bias in Recommender Systems: The Search for...by F Carnovalini · 2025 · Cited by 22 — Recommender systems are known to be...
-
Source: link.springer.com
Link: https://link.springer.com/article/10.1007/s11257-023-09364-zSource snippet
Fairness in recommender systems: research landscape and...by Y Deldjoo · 2024 · Cited by 296 — Research on fairness in recommend...
-
Source: arxiv.org
Link: https://arxiv.org/html/2412.08780v1Source snippet
Reducing Popularity Influence by Addressing Position Bias11 Dec 2024 — Position bias poses a persistent challenge in recommender systems...
-
Source: knightcolumbia.org
Title: [understanding]({{ ‘understanding/’ | relative_url }}) social media recommendation algorithms
Link: https://knightcolumbia.org/content/understanding-social-media-recommendation-algorithmsSource snippet
These algorithms are the engine that makes Facebook and YouTube what they are.Read more...
-
Source: ceur-ws.org
Link: https://ceur-ws.org/Vol-3924/short4.pdfSource snippet
CEUR Workshop ProceedingsReducing Popularity Influence by Addressing Position BiasJanuary 12, 2025 — by A Dzhoha · 2022 · Cited by 9 — As...
Published: January 12, 2025
-
Source: mlfrontiers.substack.com
Link: https://mlfrontiers.substack.com/p/some-biases-in-recommender-systemsSource snippet
Machine Learning Frontiers(Some) Biases in Recommender Systems You Should KnowIf we train a ranking model using clicks as positives, natu...
-
Source: Lawfare
Title: exploring tradeoffs in ranking and recommendation algorithms
Link: https://www.lawfaremedia.org/article/exploring-tradeoffs-in-ranking-and-recommendation-algorithmsSource snippet
Research shows...Read more...
-
Source: site.warrington.ufl.edu
Link: https://site.warrington.ufl.edu/kumar/files/2023/07/How-do-recomemndations-help-consumer-search-products.pdfSource snippet
WebsitesHow Do Product Recommendations Help Consumers Search...by XS Wan — Product recommendations can benefit consumers' online product...
-
Source: anneschuth.nl
Link: https://anneschuth.nl/assets/hofmann-effects-2014.pdfSource snippet
Anne SchuthEffects of Position Bias on Click-Based Recommender...by K Hofmann · Cited by 68 — We investigate this effect using bias mode...
Additional References
-
Source: recombee.com
Title: making recommendations fairer a new way to guarantee exposure for all
Link: https://www.recombee.com/blog/making-recommendations-fairer-a-new-way-to-guarantee-exposure-for-allSource snippet
Making Recommendations Fairer: A New Way to...29 Apr 2025 — As recommender systems become more widespread across digital platforms, conc...
-
Source: researchgate.net
Link: https://www.researchgate.net/publication/323753668_Position_Bias_in_Recommender_Systems_for_Digital_LibrariesSource snippet
Position Bias in Recommender Systems for Digital LibrariesFor example, position bias in [search rankings]({{ 'search-ranking/' | relative_url }}) strongly influences how many clic...
-
Source: academiccommons.columbia.edu
Link: https://academiccommons.columbia.edu/doi/10.7916/1h2v-pn50/downloadSource snippet
Academic CommonsUnderstanding Social Media Recommendation Algorithmsby A Narayanan · 2023 · Cited by 273 — In computer science, the algor...
-
Source: innovation.ebayinc.com
Title: e Bay Inc.A Personalized User-Based Ranking Model
Link: https://innovation.ebayinc.com/stories/evolving-recommendations-a-personalized-user-based-ranking-model/Source snippet
Personalized User-Based Ranking Model - Innovation StoriesWe developed a ranking model to generate personalized recommendations that opti...
-
Source: researchgate.net
Link: https://www.researchgate.net/publication/387721184_Algorithmic_Bias_in_Recommendation_Systems_and_Its_Social_Impact_on_User_Behavior_Algorithmic_Bias_in_Recommendation_SystemsSource snippet
n strategies of algorithmic bias in recommendation systems.Read more...
-
Source: dl.acm.org
Title: Digital Library First Things First?
Link: https://dl.acm.org/doi/10.1145/3557886Source snippet
Things First? Order Effects in Online Product...Research on recommender systems has noted that the ranking of recommended items may play...
-
Source: youtube.com
Link: https://www.youtube.com/watch?v=VJOtr47V0eoSource snippet
Reform part 2 (Day 2, Optimizing for What? Algorithmic Amplification and Society)...
-
Source: youtube.com
Title: Tutorial 3B Improving Recommender Systems with Human in the Loop
Link: https://www.youtube.com/watch?v=WfKIfWGuO_kSource snippet
How YouTube's Algorithms Can Fool You How YouTube's Algorithms Can Fool You...
-
Source: pmc.ncbi.nlm.nih.gov
Link: https://pmc.ncbi.nlm.nih.gov/articles/PMC8821643/Source snippet
NIHby L Espín-Noboa · 2022 · Cited by 61 — Here we study two well-known algorithms, namely PageRank and Who-to-Follow (WTF), and sh...
-
Source: youtube.com
Link: https://www.youtube.com/watch?v=PqbYdDiwKBYSource snippet
ion and presentation biases in search and recommender systems...
Topic Tree