Data Voids

• Bias: Data Voids are gaps in search coverage and available data where missing or low‑quality information is systematically exploited to spread disinformation (S011).
• What it breaks: Critical thinking, the ability to assess information credibility, trust in search engines and AI assistants, and the informational security of marginalized communities.
• Evidence level: L1 — high level of academic consensus with multiple empirical studies (84 citations of key works), validation from leading institutions (Microsoft Research, Stanford FSI, Harvard Misinformation Review).
• How to spot in 30 seconds: The search query returns a limited number of low‑quality results; an unusual consensus among sources on a contentious issue; warning banners from search engines about insufficient data.

When Information Gaps Become a Weapon

Data Voids constitute a critical threat to modern information ecosystems. The concept, first systematized by researchers Golebiewski and Boyd in 2019, describes information spaces where missing, limited, or low‑quality data create opportunities to manipulate search results (S011, S013). These are not merely empty spots on the internet—they are active security vulnerabilities that require systematic management.

The phenomenon of Data Voids has attracted considerable academic attention, with key papers receiving between 28 and 84 citations (S002, S010). Manipulators actively exploit Data Voids to expose users to problematic content via search engine results. Particularly concerning is that users seeking information online to fact‑check disinformation are at risk of landing precisely in those information spaces where high‑quality content is absent (S003).

Three Types of Data Voids

1. Low-quality result voids — the available search results are considered inadequate or unreliable.
2. Low-relevance voids — search results do not match the user’s intent.
3. Coverage gaps — topics with insufficient authoritative content.

Google Search and other platforms attempt to help users navigate these void types, but interventions often rely on heuristic handling rather than systematic remediation (S005, S014).

Artificial Intelligence Inherits the Problem

The Data Void problem is amplified by the rise of artificial intelligence. Large language models (LLMs) and other AI systems inherit vulnerabilities from Data Voids in their training data, leading to gaps, bias, and hallucinations (S004, S008). The data used to train LLMs suffer from limitations such as gaps, bias reflecting social inequality, and systemic distortions. This creates a new threat category—“LLM Grooming,” a cognitive threat to generative AI systems that exploits Data Voids in training data (S006).

Marginalized Communities Are Disproportionately Affected

Data Voids disproportionately affect marginalized and under‑represented communities, creating political information voids that reflect dynamics of exclusion and structural inequality (S009, S007). The study identified Data Void patterns in Google search queries that reflect exploitation by far‑right actors. Without the creation of new verified content, certain Data Voids cannot be quickly and easily filled (S001), making the issue especially hard to resolve and linking it to broader concerns such as confirmation bias and availability heuristic.

Cognitive Architecture of Information Voids: How Algorithms and the Brain Create the Illusion of Knowledge

Information voids arise at the intersection of three systems: algorithmic constraints of search engines, users’ cognitive biases, and the strategic behavior of malicious actors. When a user queries a topic with limited high‑quality content, the algorithms are forced to return whatever is available—even if it consists of low‑quality, biased, or manipulative sources (S010). This situation creates an illusion of information completeness where, in reality, there is none.

Availability Heuristic and Trust in Platforms

Availability heuristic is the first cognitive mechanism that makes information voids dangerous. We overestimate the importance and credibility of information that readily comes to mind, and the first‑page Google results are exactly that. Research shows that biased rankings of search results affect users’ attitudes toward controversial topics through information‑processing mechanisms (S010).

The second mechanism is the halo effect of authoritative platforms. We trust Google, Bing, and other search engines as reliable information intermediaries, automatically interpreting search results as “this is what is known about the topic” rather than “this is what is available in the index.” This intuitive error is amplified because platform interventions—such as warning banners—often rely on heuristic processing instead of systematic scrutiny (S005). Even when users are warned about a potential issue, they may not process the information deeply enough to change their behavior.

Confirmation of Beliefs Under Uncertainty

Confirmation bias amplifies the effect of information voids. If a user already leans toward a particular viewpoint, an information void filled with content supporting that view is perceived as confirmation of the correctness of the belief. This creates a vicious cycle: the more users rely on search engines to verify information, the more vulnerable they become to the exploitation of voids.

A study by Draws et al. (2021) at Delft University of Technology demonstrated that when search results are systematically skewed to one side of a debate, users are significantly more likely to adopt that position even if they initially held the opposite view (S010). This confirms that information voids are not merely passive gaps but active shapers of public opinion.

Scale of the Problem: From Search to Artificial Intelligence

Research by Stanford FSI showed that deep‑learning models can identify 29–58 times more low‑quality information voids than traditional methods, highlighting a scale of the problem invisible to ordinary users. Lewandowsky et al. (2023) examined the “human‑algorithm entanglement” issue in the context of information voids, demonstrating that gaps in search coverage can be exploited by malicious actors and that users often fail to recognize when they are interacting with manipulative content (S002).

Especially concerning are findings from studies of political information voids. Flores‑Saviaga et al. (2022) developed an AI system to detect political information voids in under‑represented communities, showing that marginalized groups systematically encounter more severe voids, exacerbating inequality in access to quality information. Norocel (2023) demonstrated how far‑right actors strategically exploit Google’s information voids to advance exclusionary policies.

Information Voids in the Era of Large Language Models

Recent studies from 2024–2025 have deepened our understanding of the issue. Shao et al. (2025) on AI hallucinations showed that LLM training data contain gaps, systemic bias, and reflect social inequality (S004). Information voids are embedded in the new generation of AI tools, potentially amplifying the problem.

Cognitive Mechanism	How it works in information voids	Result for the user
Availability heuristic	Top search results are perceived as the most relevant and trustworthy	Overestimation of the significance of manipulative content
Halo effect	Trust in the platform’s authority is transferred to the quality of the results	Blind acceptance of biased sources as objective
Confirmation bias	Voids are filled with content that supports existing beliefs	Reinforcement of erroneous views instead of correcting them
Bias blind spot	Users are unaware that their opinion is shaped by the algorithm	False sense of independent judgment

A study found that the Perplexity AI system generally outperformed Google in handling information voids, yet it still occasionally amplified belief in false information. This indicates that the information‑void problem is not merely technical but fundamental to any system tasked with delivering information under conditions of scarcity.

Information Security, Search Engines, Artificial Intelligence

Examples of Information Gaps in Real-World Situations

Scenario 1: Seeking Health Information During a Pandemic

At the onset of a new disease outbreak, a user looks for information about symptoms and treatment methods. Because the disease is new, authoritative medical sources have not yet published detailed guidelines, and scientific studies are in early stages. This creates a classic information gap—a coverage void on a critically important topic (S001).

In this situation, search engines return what is available: low‑quality blogs, social‑media posts, unverified “home remedies,” and deliberately crafted disinformation. Bad actors, aware of the information gap, strategically create SEO‑optimized content that promotes ineffective or dangerous treatments.

A user who trusts the search results and does not realize they are in an information gap may make health decisions based on unreliable information. Without the creation of new verified content, such gaps cannot be filled quickly, making the early stages of health crises especially vulnerable periods.

What could help: The user could check the publication dates of sources, consult official health websites (WHO, national ministries), and wait for peer‑reviewed studies rather than relying on the first search results.

Scenario 2: Political Information Gaps and Marginalized Communities

A member of an under‑represented ethnic or social community looks for information about local election candidates or political initiatives affecting their group. Authoritative sources may simply not cover issues important to these communities, creating an information gap.

This gap is filled either by low‑quality sources or by deliberately crafted content aimed at manipulating voters. Research shows that far‑right groups strategically exploit information gaps in search engines to promote exclusionary policies. A user from a marginalized community seeking information about their political rights may encounter search results that either ignore their interests or actively push an agenda contrary to their well‑being.

What could help: Turning to local community organizations, directly contacting candidate representatives, and cross‑checking information through multiple independent sources would help avoid manipulation and obtain a fuller picture.

Scenario 3: AI Assistants and Hallucinations Within Information Gaps

Modern users increasingly turn to AI assistants powered by large language models instead of traditional search engines. However, these systems inherit and potentially amplify the problem of information gaps, as their training data contain omissions, systemic bias, and reflect social inequality.

A student uses an AI assistant to write an essay on a little‑studied historical topic. If the model’s training data contain an information gap on that subject, the AI may “hallucinate”—generate plausibly‑sounding but factually incorrect information. New sources of error in AI systems are often directly tied to data omissions and information gaps.

Especially concerning is the phenomenon of “LLM grooming”—a new cognitive threat to generative AI that exploits information gaps in training data. Bad actors can deliberately create content intended to fill those gaps in future training sets, effectively “training” AI systems to spread disinformation.

What could help: The student should cross‑verify the AI‑generated information in academic databases, use multiple sources, and critically assess the plausibility of the content, especially on obscure topics.

Scenario 4: Manipulating Public Opinion Through Strategic Creation of Information Gaps

The most sophisticated form of exploiting information gaps involves deliberately creating them and then filling them with manipulative content. Ahead of a major political event or election, bad actors identify topics that receive limited coverage in authoritative sources.

They then build a network of interlinked websites, blogs, and social‑media posts optimized for search engines that promote a specific narrative line. When ordinary users search for information on these topics, they encounter a carefully constructed information ecosystem that appears diverse but actually coordinates to push a single point of view.

Manipulators leverage the fact that biased search rankings heavily shape users’ attitudes toward contentious issues. They strategically fill information gaps with content that steers public opinion in their desired direction. Deep‑learning models can identify 29–58 times more low‑quality information gaps than traditional methods, indicating the scale of potential exploitation hidden from ordinary users.

What could help: The user should verify information sources, watch for recurring narratives across different outlets, consult independent experts, and recognize that confirmation bias can lead them to believe content that aligns with their beliefs even when it is manipulative.