November 2025 – Learning Aloud

Newsela – A Follow-Up Look

From time to time, I take a look at a topic I was interested in, say 5 or so years ago, and ask what has happened since. Have classroom strategies that seemed to be enjoyable and productive survived and how have they matured? Here is an example of what I mean.

A decade ago, I was interested in the potential of technology for allowing greater individualization of instruction. My primary interest was in technology that allowed ideas from the 1970s-80s called mastery learning to become practical. Mastery learning proposed that group-based instruction largely ignored differences in aptitude and background knowledge, leading to frustration and learning challenges because the group advanced whether individuals were ready or not. To relate this to a widely recognized alternative based in technology consider the approach now allowed by the Kahn Academy.

A different approach, maintaining more of a group-based strategy, was proposed by Newsela. This company argued that reading content (individual articles) could be presented at different reading levels allowing a class to read versions of the same material maintaining the opportunity for social opportunities such as class discussions. This approach made sense to me especially when applied to reading tasks that might be described as reading to learn – e.g., assignments in science, social studies, etc. The focus on informative content rather than fiction had obvious implications for present student learning and for the future. The following two images contrast the same content presented at different reading levels.

I wrote multiple posts describing Newsela and how it might be implemented. Others were offering similar observations.

Individualizing literacy instruction with Newsela (2015)

Layering Newsela (2017)

Not all good ideas work or are practical so I decided to follow up what is now a decade later and see how the company and the product seem to be doing.

Adoption Level

It is difficult to get accurate information about student use. Newsela currently reports that it is used by 3.3 million teachers and 40 million students, the exact total it reported in 2016. Newsela has a lite and pro level and the lite level has attracted a lot of attention and occasional use. Occasional is a guess as I could not find stats on the level of activity. For some classrooms and individuals reading an occasional story would be a productive activity. I am assuming that the combination of those using the lite and the paid levels accounts for the differences in usage statistics that are reported.

The paid version is better suited to using the tool as part of the curriculum. The startup’s paid product is between $6 to $14 per student. Newsela is sold at a rate of $6000 per school or $1000 per grade. Newsela estimates that gross bookings have grown 115% over the years of the pandemic, and that revenue grew 81%. More than 11 million students were using Newsela under a licensing agreement by the end of 2021.

The version of Newsela I described in my late 2010s posts has changed substantially. Newsela has significantly evolved in recent years to become more AI-driven, expanding both its suite of educational products and the ways users interact with its content and assessment tools. A secondary emphasis on writing has emerged. Usage trends reflect a shift toward greater integration of artificial intelligence and differentiated instruction, as well as changes in accessibility and assessment features for teachers and students.

Efficacy Studies

My tendency when advocating, or at least describing, an instructional strategy implemented through a commercially available tool or product is to search for published research that evaluates the approach I want to describe. The following are descriptions of two studies I located.

WestEd (2018) Newsela efficacy study: Building comprehension through leveled nonfiction content.

Classes of fifth-grade students from two districts were randomly assigned to a Newsela or a Control condition. Reading instruction in the Newsela classes was modified to include at least two Newsela articles per week – one in class and at least one at home. Students in the Control condition relied on their normal reading curriculum. The study ran for 14 weeks and used the difference in STAR pre and post-performance scores as the dependent variable. Student compliance with the Newsela homework expectation varied widely, with 55% meeting the one-per-week expectation. When those meeting the expected level of engagement were compared with the control group, their achievement gains were significantly greater.

Literacy gains from weekly Newsela ELA use

This year-long study made use of differences in pre and post-MAP reading assessments as the dependent variable. The classes of third and fourth grade educators participated as Newsela and control conditions. The Newsela classes were asked to read at least two stories and take one multiple-choice test per week. Teachers in the control condition relied on their own selection of reading material with the largest source being content they had found through Google sources. Fourth-grade students, but not third-grade students, achieved at a significantly higher level in the Newsela condition.

Why can’t I find peer-reviewed published studies

Often, I am frustrated when I cannot find studies that directly support the strategy I want to describe. This is the case with Newsela and I have been thinking about why this is the case.

Newsela has engaged outside agencies (e.g., WestEd) to conduct research using their products, but these studies are available as what I would describe as technical reports and don’t seem to appear in scholarly journals. After reading these reports I can see that if I had been asked to review the research for publication, I would also identify issues that would cause me to suggest that the study not be published. In the studies I will describe here, I see flaws in the research design that allow alternate explanations for the positive results.

Applied research is often very difficult because those implementing the research have their own issues and priorities. Sometimes a methodology does involve tight controls from the beginning and sometimes it seems that original design is allowed to slip as unanticipated issues come up.

For example, in the first study I describe, the plan was to have a control group and a Newsela group with one in-class and one homework reading assignment a week. It turned out that the homework assignment was ignored in many cases and to generate significant evidence that Newsela was productive the researchers compared those who did the homework against the control group. This may not be important, but it could also mean that the Newsela group now consists of more motivated readers than the control group, and this interest in reading, rather than the Newsela content and approach, was what created the difference in the development of reading skill. It is unclear to me from reading the description why expectations for completing the homework such as including completion as part of the grading scheme was not implemented. I can imagine a different controversy if what I propose was implemented as you would then extra reading required in the Newsela group and not in the control group. Perhaps the most ideal approach would be to maintain control of all of the reading assignments within the classroom setting so that the time allocated could be matched.

The second study I have described is limited by what I would describe as clear identification of what is the intended independent variable. What has always attracted my interest in Newsela was the group-based, but individualized approach the content allows. Each Newsela document is available at multiple level (5) of complexity. This allows those readers at different levels of aptitude and skill development to read a variant of the same content so that discussion and a social element of instruction can be maintained. My personal interest in technology-supported learning has always been based on the potential of individualization. One argument some make about many technology applications that allow for differences in rate of learning is that students are isolated and miss out on the social benefits of a classroom setting. Newsela offer an alternative approach that maintains the social setting.

This study creates a different or at least an added difference when comparing the Newsela group and the control group. The authors report that when teachers select the reading content for the control condition this material differs in category from the Newsela treatment. Teachers in the control condition were described as relying on Google searches to find content fitting with the topics that they covered and this content contained significantly less “nonfiction” content. A cleaner approach more consistent with what I think is the unique Newsela content would be to compare the impact of a single version of articles versus multiple versions of the same articles.

Summary

As a commercial venture Newsela seems to be doing well. It has a solid base of schools that have committed to purchasing the program. My criticism of the weak methodologies used in evaluation efforts is mostly a function of my interest in the impact of the individualization efforts the resources provide. Having current and nonfiction content is important, but the strategy on which the company originally made its name has not been rigorously evaluated.

It now seems educators could use any of several AI tools to create similar content. Prompts such as rewrite this content at a level appropriate to fifth grade students could be applied to any content a teacher could upload. Given this option, the value to a district would depend on the time savings to teachers and the constant access to new content would be the advantages Newsela provides.

Where is the thinking in the AI classroom?

The concept of generative activities has consistently shaped my thinking and teaching on learning. I admit that such activities are not ultimately necessary. Still, they represent ways for learners and those who try to help them grow to understand and imagine how skills and knowledge might be applied. Focusing on generative activities was particularly useful in my interest in studying – the work an individual does to make experiences personally informative and useful.

Basic Definitions:

Studying – the mental and external activities a learner engages in after exposure to potentially useful experiences that are intended to store a representation of these and create meaning.

Generative activities – external tasks intended to encourage productive cognitive (mental) behaviors

Why is this perspective important at this time? My concern is that certain uses of AI are frequently being substituted for generative activities allowing individuals to accomplish tasks without achieving the cognitive benefits (i.e., retention, understanding) engagement with the generative tasks make more likely.

Why would learners substitute AI for generative activities? It seems likely they see AI as producing an equal or even superior product without the effort required to create such products on their own. This reflects both a focus on short-term benefits over long-term benefits and probably a lack of understanding of how personal knowledge and skills are developed, or perhaps even a disinterest in developing these personal attributes.

Some background:

When I explain the concept of generative activities I like to start with Rothkopf’s concept of a mathemagenic task because this researcher’s focus tends to make intuitive sense to most people. Rothkopf was interested in questions and variations in how questions might be associated with written material.

Questions presented before you read.

Questions presented after you read.

Inserted questions – questions added within text.

Different types of questions – application questions, factual questions.

Mathemagenic tasks

The made-up word mathemagenic translates roughly as giving birth to knowledge, implying that in attempting to answer questions, you might accomplish something else – a better likelihood of future retention, greater likelihood that you would recognize possible applications – that would not have occurred without exposure to the questions. My favorite example relates to the challenge educators often face in encouraging students to see the relevance of general concepts they have been taught. This translates as connecting new ideas with what you already know. The examples and the relevance are potentially there if you can make connections. So, what not ask students directly – provide an example of XXX? If personal examples exist, but learners have not made the effort to make the connections, perhaps the request will encourage that specific cognitive effort.

There is a huge body of research on all aspects of questioning. Questions are an everyday classroom activity, but the insight is just why do we spend the time, and could a more careful use of questions result in improved results? My favorite example here is what is called wait time – the average delay after asking a question (silence to allow thinking) is a little over a second. If we want students to think, typical behavior in classroom discussions is not particularly rationale. There is reason to examine and challenge typical behavior.

Anyway, questions are an external task that can be used to manipulate – change the odds of – productive cognitive behaviors. I suggest adding one important final point: a learner can ask herself questions, e.g., using flashcards. So various ways in which questions can be generated and used are an aspect of what those interested in study behavior investigate.

Generative Tasks

The concept of generative activities is simply an expansion of this same idea and asking questions would be one of many generative strategies. The idea of generative activities is not new (Wittrock, 1974, 2010) and to many educators may seem obvious and a reflection of common classroom practices. While true, researchers have attempted to understand the underlying mechanisms and to consider just how efficient different activities were especially in the comparison of one to others ( Fiorella & Mayer, 2016). A personal interest and one clearly relevant to the topic of how AI is applied in classrooms is writing to learn. I have always felt through self awareness that requires careful examination of existing ideas and integration of ideas from a variety of experiences to produce a product. There is a substantial body of research to support such perceptions (e.g., Graham et al., 2020). To be clear, researchers consider a variety of writing activities under the umbrella of writing to learn. The product need not be a massive, semester summarizing paper, but perhaps also notes and short, five-minute end of class descriptions related to the content just presented.

Caveat

One issue I think is important that may not be apparent in the notions that generative activities are intended to encourage productive cognitive skills is that such skills may occur without this external requirement and guidance and there is always the possibility that for some motivated and capable of thinking deeply, without such tasks, the task represents a form of “busy work”. In other words, the task adds little beyond annoyance. Of course, the reality is that educators in actual classrooms typically do not feel that they can arbitrarily assign tasks to some students and not others, so they must always deal with reactions to assignments, both legitimate and resulting from laziness.

AI and Generative Tasks

AI discussions related to education always seem to generate a good news / bad news situation. There seem to be several examples that apply to this general topic.

AI can be applied to render the potential benefits of a generative strategy useless. For example, if AI is used to respond wholly to a writing-to-learn assignment, the learner completes the assignment without engaging in much cognitive work. The educator is then in a position of assigning a task that takes valuable learning time and adds a commitment to the effort to provide feedback, but has little impact.

In contrast, AI can be used to formulate questions (both objective and open-ended) related to assigned material and to respond to a learner’s responses to such questions. Learners can even generate such activities on their own. It seems to me that the use of what might be described as short essay questions offer a unique advantage that would be difficult or at least very time consuming for the educator to administer. AI tools are very flexible and can ask and react to the answers for different types of questions. Short answer questions are a form of writing to learn and involve greater “retrieval practice” benefits than formats such as multiple choice that are useful, but less demanding of retrieval.

Summary

My effort here was intended as a way educations might frame their way of thinking about AI in classrooms using examples I assume are familiar. I hope this approach can be generalized. Of course, the challenge is in manipulating AI-based and any assigned activities so that productive thinking activities are encouraged and also that students gain insight into the importance of the mental work that is required of certain task. I understand this may seem obvious, but the work of adjusting to the advantages and disadvantages of AI tools will take some time and careful study. For example, I wonder if writing and organizing notes may accomplish much the same benefits as creating a writing to learn product. Learning to write is somewhat different than writing to learn although writing across the curriculum offers a secondary benefit of practicing writing skills. There are plenty of options to consider. We presently do little to teach advanced note making skills and note using skills even though these topics have received a great deal of attention as benefits to out of school functioning.

Citations

Fiorella, L., & Mayer, R. E. (2016). Eight ways to promote generative learning. Educational Psychology Review, 28(4), 717-741.

Graham, S., Kiuhara, S. A., & MacKay, M. (2020). The effects of writing on learning in science, social studies, and mathematics: A meta-analysis. Review of Educational Research, 90(2), 179-226.

Rothkopf, E. Z. (1970). The concept of mathemagenic activities. Review of educational research. 40(3), 325-336.

Wittrock, M. C. (1974). Learning as a generative process . Educational Psychologist, 11(2), 87–95. https://doi.org/10.1080/00461527409529129

Wittrock, M. C. (2010). Learning as a generative process. Educational Psychologist, 45(1), 40-45.