Executive Summary

This article presents the research-backed scientific foundation for American Digital Education's approach to image placement within its AI-powered, standards-based video lessons. Drawing on more than four decades of empirical research in cognitive psychology, instructional design, and multimedia learning, it demonstrates that the program's image usage strategy is not merely a stylistic choice but a deliberate application of well-established principles that maximize learning while minimizing cognitive waste.

The core finding is straightforward: American Digital Education's visual design is well-calibrated. Images appear only when students need to solve problems or connect concepts, not during definition reading, comprehension checks, or other phases where visual information would compete with essential verbal processing. This is precisely what the science prescribes.

This article explains the theories, principles, and experimental evidence that support this approach, organized into four major areas: Cognitive Load Theory and its three load types; Richard Mayer's Cognitive Theory of Multimedia Learning and its design principles; the research on image typologies (instructional versus decorative versus seductive); and the worked example effect and its relationship to visual scaffolding. Each section explains the relevant science, its implications for K–12 education, and how American Digital Education's design aligns with research-validated best practices.

1. Cognitive Load Theory: The Foundational Science

1.1 What Is Cognitive Load Theory?

Cognitive Load Theory (CLT) is one of the most extensively validated frameworks in instructional science. Developed by John Sweller beginning in the late 1980s, CLT is grounded in the architecture of human cognition, specifically the relationship between working memory and long-term memory during learning (Sweller, 1988; Sweller, van Merriënboer, & Paas, 1998).

The theory rests on a fundamental biological constraint: working memory, the mental workspace where we consciously process new information, is severely limited. George Miller's foundational 1956 research established that humans can hold approximately seven (plus or minus two) items in short-term memory at once. More recent research by Nelson Cowan (2001) suggests the effective limit may be closer to four chunks when processing novel material. This capacity limitation is not a flaw to be overcome; it is a feature of human cognitive architecture that instruction must respect.

Long-term memory, by contrast, is functionally unlimited. It stores organized knowledge structures called schemas, which are mental frameworks that allow us to recognize patterns, categorize problems, and retrieve procedures efficiently. The goal of instruction is to facilitate the construction and automation of schemas in long-term memory, but all new information must first pass through the bottleneck of working memory to get there.

1.2 The Three Types of Cognitive Load

Intrinsic Cognitive Load

Intrinsic load is determined by the inherent complexity of the material being learned and the learner's prior knowledge. A lesson on single-digit addition has low intrinsic load for most students; a lesson on solving systems of equations has high intrinsic load because many interacting elements must be held in working memory simultaneously (Sweller, 2010). Intrinsic load cannot be directly reduced without changing the content itself, but it can be managed through sequencing, scaffolding, and building on prior knowledge.

Extraneous Cognitive Load

Extraneous load is generated by instructional design choices that do not contribute to learning. It represents wasted cognitive capacity: mental effort spent on processing irrelevant information, searching for connections between poorly integrated materials, or decoding confusing layouts. Decorative images that have no instructional purpose, background music in educational software, and interesting-but-irrelevant anecdotes all generate extraneous load (Chandler & Sweller, 1991; Mayer, 2009). Reducing extraneous load is the primary goal of research-based instructional design.

Germane Cognitive Load

Germane load represents the mental effort devoted to constructing, elaborating, and automating schemas in long-term memory, which is the cognitive work of actually learning. When a student studies a worked example and actively maps each step to an underlying principle, that effortful processing is germane load (Paas, Renkl, & Sweller, 2003). Germane processing is productive and desirable. The ultimate goal of instructional design is to minimize extraneous load so that available working memory capacity can be devoted to germane processing.

1.3 The Additive Relationship

These three load types are generally understood to be additive: intrinsic load plus extraneous load plus germane load equals total cognitive load (Sweller, van Merriënboer, & Paas, 1998). Cognitive overload occurs when total load exceeds working memory capacity, and when that happens, learning is impaired or halted entirely. This means that even well-intentioned additions to instructional materials, such as a colorful illustration, an engaging anecdote, or a supplementary graphic, can push total load past the threshold and cause learning to fail.

1.4 Why This Matters for K–12 Students on Screens

The implications of CLT are especially acute for K–12 students learning from computer-based instruction. Children and adolescents have less developed working memory capacity than adults (Cowan et al., 2005), less prior knowledge to anchor new material (meaning higher intrinsic load for most academic content), and less developed self-regulation skills to manage their own attention. When they are working at a screen, where every pixel is competing for their limited cognitive resources, the design of what appears on that screen at each moment becomes a matter of cognitive science, not aesthetics.

Every image placed on a screen during instruction either supports learning (contributing to germane load) or detracts from it (adding extraneous load). There is no neutral option. An image that is irrelevant to the current instructional purpose does not simply "sit there" harmlessly; it consumes processing resources that could otherwise be allocated to learning. This is the fundamental insight that drives American Digital Education's image placement strategy.

2. The Cognitive Theory of Multimedia Learning

2.1 Mayer's Framework

Richard Mayer's Cognitive Theory of Multimedia Learning (CTML) extends CLT into the specific domain of instruction that combines words and pictures (Mayer, 2001, 2009, 2021). Building on Allan Paivio's Dual Coding Theory (1971, 1986), which established that humans maintain separate but interconnected systems for processing verbal and visual information, Mayer developed a comprehensive theory explaining when and how multimedia helps or hinders learning.

CTML rests on three empirically supported assumptions. First, the dual-channel assumption: humans possess separate channels for processing visual/pictorial and auditory/verbal information. Second, the limited-capacity assumption: each channel can process only a small amount of information at any given time. Third, the active-processing assumption: meaningful learning requires the learner to select relevant information, organize it into coherent mental models, and integrate it with prior knowledge from long-term memory (Mayer, 2009).

2.2 Key Principles and Their Evidence Base

Through decades of controlled experiments, Mayer and his colleagues identified twelve principles of multimedia design. Several are directly relevant to the question of image placement in educational technology:

The Multimedia Principle

People learn better from words and pictures together than from words alone, provided the pictures are relevant to the learning objective (Mayer, 2009). A 2025 meta-analysis of Mayer's body of research found large, consistent effect sizes for text combined with instructional diagrams across factual, inferential, and transfer outcomes. This principle establishes that images have genuine instructional power, but only when they serve the learning goal.

The Coherence Principle

People learn better when extraneous material is excluded rather than included (Mayer, 2009). This principle has three versions: learning improves when irrelevant words and pictures are removed; when irrelevant sounds and music are removed; and when unnecessary symbols and embellishments are eliminated. The coherence principle directly warns against adding images for aesthetic or engagement purposes when those images do not serve the instructional objective.

The research basis is substantial. Mayer, Heiser, and Lonn (2001) demonstrated that adding supplementary narration segments to a multimedia lesson, even when the additions were interesting and topically related, resulted in less learning, not more. A key finding from a 2012 meta-analysis by Rey found significant and large effect sizes for removing seductive details, confirming that interesting-but-irrelevant material reliably harms learning outcomes.

The Signaling Principle

Learning improves when cues are added to highlight the organization and essential content of the material (Mayer, 2009). This principle supports the practice of using images as signals: visual markers that direct student attention to what matters. When American Digital Education includes an "image-based depiction of the problem" during worked examples, it is deploying the image as a cognitive signal, telling the student, "This is where you apply what you're learning."

The Segmenting Principle

Complex material is better learned when presented in learner-paced segments rather than as a continuous flow (Mayer, 2009). American Digital Education's lesson structure, which moves through distinct phases from learning objectives to definitions to worked examples to practice, inherently applies the segmenting principle. The strategic inclusion of images only during certain segments reinforces the cognitive boundary between lesson phases, signaling to students when they are shifting from receiving information to applying it.

The Pre-Training Principle

Students learn more deeply from a multimedia presentation when they already know the names and characteristics of key concepts (Mayer, 2009). American Digital Education's decision to keep images absent during concept definition phases and present them only during worked examples aligns with this principle. Students first build verbal understanding of key concepts (pre-training), then encounter visual representations when they are cognitively ready to integrate them with that understanding.

3. Image Typologies: What the Research Tells Us About Different Types of Visual Content

3.1 Three Categories of Images in Instruction

Researchers distinguish three functional categories of images used in instructional materials, each with different effects on learning (Mayer, 2009; Sung & Mayer, 2012; Schneider, Nebel, & Rey, 2018):

Instructional Images

Images that are directly relevant to the learning objective and contribute to the learner's construction of a mental model. Examples include a diagram of a mathematical problem, a labeled illustration of a scientific process, or a visual representation of a word problem scenario. Instructional images consistently improve learning outcomes across multiple studies and age groups (Mayer, 2009; Mayer, 2020).

Decorative Images

Images that are aesthetically appealing but not directly relevant to the learning content. A picture of flowers on a mathematics worksheet, for instance, or a stock photo of a smiling child on a reading comprehension page. Research by Lenzner, Schnotz, and Müller (2013) using eye-tracking technology with 7th and 8th grade students found that decorative images receive only brief initial attention before being largely ignored. While decorative images were not directly harmful in isolation, they moderated the beneficial effect of instructional images, particularly for learners with lower prior knowledge. In essence, decorative images diluted the learning benefit of instructional images.

Seductive Images

Images that are highly interesting and attention-grabbing but not relevant to the instructional goal. These are the most dangerous category. Harp and Mayer (1998) demonstrated that when students read a lesson containing irrelevant but interesting illustrations, they performed significantly worse on both recall and transfer tests compared to students who studied the same material without those illustrations.

Research identifies three mechanisms through which seductive images harm learning. The distraction hypothesis holds that interesting images draw attention away from relevant content. The disruption hypothesis holds that they interrupt the coherent mental model construction process. The schema hypothesis, which Harp and Mayer found to be the most explanatory, holds that seductive details activate irrelevant prior knowledge, causing the learner to organize their mental representation around the wrong information (Harp & Mayer, 1998; Schneider et al., 2023).

3.2 The Critical Distinction for K–12

The distinction between image types is especially important for K–12 students. A 2012 study by Sung and Mayer found that students who received any kind of graphic in a lesson reported liking the lesson better, but only students who received instructional graphics actually learned more. Students were unable to distinguish between images that helped their learning and images that did not. This means educators and technology designers cannot rely on student preference to guide image selection; they must apply research-based criteria.

For younger students and those with lower prior knowledge, the risks are amplified. Research consistently shows that the negative effects of extraneous imagery are most pronounced for novice learners (Lenzner et al., 2013; Fiorella & Mayer, 2021; Mayer & Fiorella, 2022). K–12 students, who are by definition novices in the subjects they are studying, are the population most vulnerable to poorly designed visual materials.

4. The Worked Example Effect and Visual Scaffolding

4.1 What Is the Worked Example Effect?

The worked example effect is one of the most robust findings in educational psychology. Sweller described it as "the best known and most widely studied of the cognitive load effects" (Sweller, 2006). The effect demonstrates that novice learners who study worked-out solutions to problems learn more effectively than learners who attempt to solve equivalent problems through unguided practice (Sweller & Cooper, 1985; Renkl, 2014).

The mechanism is well understood through the lens of CLT. When a novice attempts to solve a problem without guidance, they typically resort to means-ends analysis, a general problem-solving strategy that requires holding the current state, goal state, and possible operations in working memory simultaneously. This imposes enormous extraneous cognitive load, leaving little capacity for the germane processing needed to construct schemas. Worked examples eliminate this burden by providing the solution pathway, freeing working memory resources for the essential work of understanding why each step works and how steps connect to form a general procedure.

4.2 Why Images Belong with Worked Examples

Research consistently shows that pairing relevant visuals with worked examples enhances learning in several ways. Visual representations of problems externalize complex information, reducing the demand on working memory by providing an external reference that students can inspect rather than having to hold in their minds (Mayer, 2009; Sweller, 2005). For mathematical and scientific problems, images help students map abstract notation to concrete situations, a process essential for building transferable schemas.

The combination is particularly powerful in K–12 contexts. When students encounter a word problem about dividing items equally among groups, an accompanying image depicting that scenario provides a concrete anchor for the abstract mathematical operation. The image serves a specific instructional function: it is an externalized representation of the problem that students can reference as they follow the solution steps. This is the definition of germane visual content.

4.3 Why Images Do Not Belong During Other Lesson Phases

The research is equally clear about when images should be absent. During concept definition phases, students are engaged in verbal encoding, processing and storing new terminology and its meanings. Adding visual content during this phase creates split attention: the learner must divide cognitive resources between reading the definition and processing the image, even if the image is relevant (Chandler & Sweller, 1991, 1992). If the image is redundant with the verbal content (depicting exactly what the words describe), it violates the redundancy principle and wastes processing capacity (Mayer, 2009).

During checking for understanding phases, students are retrieving information from memory and applying it to evaluate answer choices. This retrieval process requires focused verbal processing. Images at this point would either serve as unwanted cues (undermining the assessment function by providing hints), generate extraneous load (if irrelevant), or create split attention (if students look at the image instead of engaging in retrieval).

During closure and relevance sections, students are consolidating what they have learned and connecting it to broader contexts. This is a metacognitive process that benefits from focused reflection, not visual stimulation. Research on the segmenting principle supports this: clear boundaries between instructional phases help students organize their learning (Mayer, 2009).

5. Dual Coding Theory: When Two Channels Help and When They Hurt

5.1 Paivio's Foundational Work

Allan Paivio's Dual Coding Theory (1971, 1986) established that human cognition operates through two distinct but interconnected representational systems: a verbal system that processes linguistic information and a nonverbal system that handles imagery, spatial relations, and sensory impressions. When information is encoded through both systems, verbal and visual simultaneously, it creates multiple retrieval pathways in memory, enhancing recall and comprehension (Clark & Paivio, 1991).

This theory provides the foundational rationale for using images in instruction at all. However, Paivio's theory also contains an important constraint: dual coding only enhances learning when the verbal and visual inputs reinforce the same concept and can be integrated into a single coherent mental model. When integration fails, whether because the image is irrelevant, because it contradicts the verbal content, or because the learner lacks the capacity to process both simultaneously, the cognitive benefit evaporates and can reverse into a deficit (Mayer & Moreno, 1998; Sweller, 2005).

5.2 The Split Attention Effect

When visual and verbal materials that refer to each other are presented separately, or when the visual is not meaningfully connected to the current verbal content, learners must split their attention between the two sources and mentally integrate them. This consumes working memory resources for integration work that could be eliminated through better design (Chandler & Sweller, 1992; Kalyuga, Chandler, & Sweller, 1999).

For computer-based instruction, the split attention effect is a constant risk. Every element on the screen competes for the visual channel. When images appear alongside text during phases where the text is carrying the full instructional burden, such as definitions or comprehension questions, split attention is almost inevitable, even if the image seems related to the broader topic. The image is not serving the immediate instructional purpose, so the learner's visual system is processing information that does not integrate with what the verbal system is working on.

5.3 Implications for Screen-Based K–12 Instruction

In screen-based learning environments, students' visual channel is already processing text (which enters through the visual channel before being recoded into verbal representations in working memory). Adding images means the visual channel is processing both text and pictures simultaneously. This is manageable when the image directly supports the text's instructional goal, as when a problem illustration accompanies a worked example. It becomes counterproductive when the image is unrelated to the immediate learning task.

American Digital Education's approach of reserving images for worked examples and practice problems ensures that the visual channel is never asked to process competing information. During text-heavy phases (definitions, objectives, comprehension checks), the visual channel is dedicated to processing the text itself. During problem-solving phases, the visual channel receives both the text and a relevant image that directly supports comprehension and schema construction. This is dual coding applied correctly.

6. The Science of Strategic Absence

6.1 Why What You Leave Out Matters as Much as What You Include

One of the most counterintuitive findings in educational research is that more content frequently produces less learning. Mayer and colleagues have demonstrated this repeatedly: students who study concise summaries with targeted illustrations consistently outperform students who study comprehensive versions with additional details and images (Mayer et al., 1996). The research title captures the paradox: "When Less Is More."

This finding arises directly from the limited-capacity assumption. Working memory cannot expand to accommodate additional material. When more information is added, something must be displaced. If the displaced processing was germane (contributing to schema construction), learning suffers. The instructional designer's job is not to fill every available space with content but to curate ruthlessly, ensuring that every element on the screen earns its place by contributing to the learning objective.

6.2 Phase-by-Phase Analysis of American Digital Education's Image Strategy

Learning Objective Phase: No Images

Research basis: During learning objective reading, students are establishing a cognitive framework for the upcoming lesson, essentially creating a mental "folder" into which new information will be organized. This is a verbal encoding process that requires focused processing in the verbal channel. Adding images at this phase would introduce visual processing demands before students have any conceptual framework for interpreting them, generating extraneous load.

Concept Definition Phase: No Images

Research basis: Definitions require precise verbal encoding. Students are learning what terms mean, a fundamentally linguistic task. Images during this phase would either be redundant with the definition (violating the redundancy principle), partially related (creating split attention), or unrelated (generating extraneous load). The verbal system needs uncontested access to working memory during this phase.

Examples and Non-Examples Phase: No Images (Verbal Descriptions Only)

Research basis: This phase uses verbal examples and non-examples to delineate concept boundaries. By relying on verbal descriptions rather than images, students are forced to generate their own mental imagery, a desirable difficulty that strengthens schema construction. Research on the generation effect demonstrates that information students generate themselves is better retained than information simply presented to them (Fiorella & Mayer, 2016). Using verbal-only descriptions during this phase leverages the generation effect for concept learning.

Worked Examples Phase: Images Present ✓

Research basis: This is the optimal phase for visual content. Students have already learned the relevant concepts (pre-training effect) and are now applying them to specific problems. The image serves as an external representation of the problem scenario, reducing the intrinsic load of holding the problem context in working memory while simultaneously working through solution steps. The image is instructional, directly tied to the problem, and supports dual coding by providing a visual anchor for the verbal explanation (Mayer, 2009; Sweller, 2005).

Practice Problems Phase: Images Present ✓

Research basis: During practice, students are transferring what they learned from worked examples to new but structurally similar problems. An image depicting the new problem scenario helps students map the general schema they developed during the worked example to the specific conditions of the practice problem. This mapping process is germane load, the cognitive work of schema refinement and transfer (Paas, Renkl, & Sweller, 2003).

Checking for Understanding Phase: No Images

Research basis: Assessment phases require students to retrieve knowledge from memory. Providing images during retrieval would undermine the testing effect, the well-established finding that effortful retrieval strengthens memory more than passive review (Roediger & Karpicke, 2006). Images during comprehension checks could serve as retrieval cues that reduce the retrieval effort, or they could create extraneous load if irrelevant to the specific question.

Closure and Relevance Phase: No Images

Research basis: The closure phase involves metacognitive reflection and connection-making. Students are thinking about what they learned and why it matters. This is abstract, reflective processing that does not benefit from concrete visual content. Keeping the screen visually simple during closure allows working memory to focus on consolidation and transfer.

7. The Value of Consistent Visual Patterns

7.1 Predictability Reduces Extraneous Load

American Digital Education's image placement is not only scientifically justified in terms of when images appear, but also in the consistency of that placement across lessons. Research on schema-based learning demonstrates that consistent patterns in instructional design reduce the extraneous cognitive load associated with orienting to new material (Sweller et al., 2019).

When students encounter lesson after lesson in which images reliably appear during worked examples and practice problems, and reliably do not appear during other phases, they develop an automated expectation about the structure of instruction. This automation means students no longer need to devote working memory resources to figuring out "what does this image mean?" or "why is this here?" Instead, they can immediately begin integrating the image with the instructional content, because they already know the image is there to support problem comprehension.

7.2 The Expertise Reversal Effect and Adaptive Design

CLT research has identified the expertise reversal effect: instructional methods that benefit novice learners can become counterproductive for more advanced learners, and vice versa (Kalyuga et al., 2003). For novice learners, which describes the majority of K–12 students encountering new content, explicit guidance, worked examples, and supporting images are optimal. American Digital Education's consistent inclusion of problem-depicting images during worked examples and practice is calibrated for the novice learner, which is the appropriate default for a K–12 platform serving students across 2 languages (English and Spanish) who are building foundational literacy and numeracy skills.

8. Synthesis: A Research-Validated Design

The convergence of evidence from Cognitive Load Theory, the Cognitive Theory of Multimedia Learning, research on image typologies, the worked example effect, dual coding theory, and the split attention effect all point to the same set of design principles:

American Digital Education's image strategy is not an approximation of research-based principles; it is a direct implementation of them. The program's specification that images should provide an "image-based depiction of the problem" during worked examples and practice, and should be absent during all other lesson phases, precisely mirrors what four decades of experimental research in cognitive psychology and instructional design prescribes for optimizing learning in computer-based K–12 instruction.

References