Optimized Reading

The world is increasingly oversaturated with text. This is an old complaint. As long as it has existed, has been accompanied by promises to solve this oversaturation. With the advent of commercial large language models, tools marketed to help users control and manage this influx of text are now ubiquitous. Much of the general discussion of—and consternation about—generative AI has focused on its ability to generate fluent and competent-sounding text. However, these tools have been shoved into workaday software such as Outlook, Word, Acrobat, and Google search with the hope that people will use them not just for writing, but also for reading. Everyday text-intensive activities such as navigating the internet, checking one's email, downloading a PDF, or opening a Word document now elicit automatic overviews or queries about summaries from various chatbots. Indeed, it can be difficult to use your computer or phone for almost any task associated with reading without being confronted with requests from one "AI assistant" or another to help you do it—or to do it for you.

One way to understand the current promotion of AI summarization tools, therefore, is as an attempt to optimize reading. Reading has often been treated with suspicion due to its perceived inefficiency: it takes up time and produces nothing tangible. As other pieces in this cluster emphasize, due to this persistent association of reading with leisure, or with laziness, those of us in literary studies have often found ourselves in the awkward position of having to defend the labor of reading as labor. Ironically, the companies marketing large language models as tools for optimizing reading recognize reading as labor, too; for them, reading is optimizable precisely because they see it as a form of work. Under this logic, to optimize reading is to put reading to use toward some profitable end, and so the more "reading" one can accomplish in the least amount of time, the better.

To view reading as optimizable positions it as a primarily managerial activity. For example, promotional materials describing Copilot, Microsoft's large language model, and its integration with Word claim that Copilot saves users "hours of manual reading...by condensing lengthy documents." The phrase "manual reading" is less a marker of embodiment here than it is of class. Copilot enables users to "extract actionable points" to "make every meeting more productive with instant summaries," replacing the manual (blue-collar) labor of reading with the managerial (white-collar) labor of extraction and time management.¹

Attempts to optimize reading long predate large language models.² Since the mid-20th century—about when the use of the term "optimize" to mean "to make the best or most effective use of" entered the written record in English—such attempts have often relied on computational technologies for their rhetorical, if not actual, efficacy.³ The mid-20th-century popularity of speed reading was one attempt to optimize reading in the shadow of computation. Speed reading during this time positioned itself as a more efficient, more modern, and ultimately more enjoyable form of reading that would allow people to learn to read, as Evelyn Wood, the founder of the most popular speed-reading company of the time, put it, "at a pace formerly reserved for electronic computers."⁴ I've argued elsewhere that, like close reading, which emerged around the same time, speed reading was paradoxically an attempt to promote literary reading, by humans, in a time of ascendant machine reading.⁵ Here, however, I want to focus on Wood's insistence that people could in fact learn to read at the speed of "electronic computers." If Wood's speed-reading company sought to promote literary reading, it did so by insisting that people could do it like a computer. What does this attempt to optimize reading along specifically computational lines reveal about our own moment of reading optimization, one also defined by computation?

* * *

The mid-20th-century United States witnessed the growing ubiquity of speed-reading books, courses, and devices. As Adrian Johns writes in The Science of Reading, "it is plausible to suppose that every literate American of the postwar generation must have been exposed to the hard-sell of [speed-reading] techniques by one institution or another."⁶ Training was available in various forms and by a variety of companies, ranging from standalone guidebooks to pacing machines to paid classes. This newfound efficacy could itself be speedily acquired: speed-reading companies promised to help people increase their reading speed in just a few weeks' time. Evelyn Wood's Reading Dynamics was the most widely known of these companies during the 1960s and early 70s. Wood, a high-school English, speech, and remedial reading teacher from Utah, started the company with her husband in 1959, and by the early 1960s the Woods had franchised institutes in many US cities. The wild claims Wood made about teaching people to read thousands of words per minute garnered enthusiastic coverage in local and national newspapers and on television programs such as the Tonight Show.

By the end of the 1960s, however, Reading Dynamics began to lose much of its credibility when studies appeared in mainstream outlets demonstrating that it is not physically possible to read thousands of words per minute. By the late 1970s, Reading Dynamics was widely understood—and often derided—as a scam. Nevertheless, when Wood died in 1995, an obituary in the New York Times proclaimed without irony that she "liberated students, professionals and business people from the habits that shackled them to the average American's reading rate of 250 to 300 words a minute."⁷

As mentioned above, during the company's heyday, Wood often repeated the claim that her program could teach people to read "at a pace formerly reserved for electronic computers." This was a convenient comparison during a time in which computing, inaccessible to most, was associated with technological innovation but not with everyday experience. Such comparisons implicitly tied Reading Dynamics to notions of science and progress while also taking advantage of the fact that most people would have been unfamiliar with how computing worked. But the comparison to computing was more than rhetoric for Wood: her imagination of how speed reading worked on a cognitive level was also curiously computational.

In one lecture at a 1965 Reading Dynamics convention for instructors and company personnel, Wood compared speed reading to a computational process. "We read very much like the computer.... Everything comes in...as it is read," she stated. Then, once "you let everything come in," "before you can send it out...we have put it in drawers, we have put it in cabinets, filed it, we have put it in cases and when you get that out you got it in one piece."⁸ In this garbled comparison of speed reading to computer processing, perceiving words on a page is imagined as input and the speed reader's recall of what they perceived is imagined as output. Along the way, the speed reader's brain organizes this input into drawers, cabinets, files, and cases for later reassembly and output as "the main ideas" of a text.⁹ This is of course not how computers read, and it wasn't how they read in 1965. Rather, "the computer" in this passage operates metonymically, standing in for an optimized office workflow; the analogy is more managerial than computational.

One aspect of this imagination of reading that aligns more closely with current ideas about computing is its instantaneous quality: "everything comes in...as it is read." In this passage, Wood imagines speed reading as an immediate grasping of the ideas on the page. Such immediacy of understanding was possible, Wood claimed, only after students learned to take in whole pages of text in one glance. Reading Dynamics instructor training materials describe this as learning to as "bypass the auditory system," a phrase Wood used to refer to the "conventional" way of reading one word at a time, as if one were reading them aloud. Instead, speed readers learn to access "a total concept" of a page in an instant.¹⁰ "This new concept is the difference between looking at the words on a page illuminated by a penlight moving across the print and a floodlight moving down the page," another instructor manual describes. "It is psychologically getting above the print and viewing it as though you were in an airplane viewing a city."¹¹ As we can see, Wood was fond of using a variety of metaphors to describe speed reading, but the basic idea remained consistent: like a computer, which does not need to take in textual input word by word but rather can process it all simultaneously, speed readers, through training, could learn to take in whole pages of text in a single glance.

Doing so involved learning how to physically train one's eyes and body to ingest text in the right way, and students in Reading Dynamics classes completed many exercises and drills aimed at supposedly improving their ability to see multiple lines on a page at once. But, as we have seen, it also involved learning to cognitively organize the large amounts of text one was ingesting for later recall. This organization happened, according to Reading Dynamics, via an associative process. Wood claimed that if readers could learn to remember a few "loaded words" from each reading session, they could then learn to "get from the loaded words the details," as "there is no limit as to the amount that the loaded words can carry."¹² In this imagination of reading, each loaded word stands in for a host of particular details about a text that are accessible to the reader via their memory of that word. The "main idea" of a text emerges out of this associational web.

This imagination of reading follows from Wood's claim that individual words in of themselves are not important to understanding the meaning of a text. Rather, she claimed that "a word is only important according to the context in which it is used."¹³ Accordingly, "rapid readers read for context."¹⁴ For Wood, teaching students to remember "loaded words" worked as a method of recall because such words helped students to access the context of a passage, scene, or even whole text. The thinking was that, although humans can remember only a relatively small number of discrete words at a time, those words themselves can be used to access knowledge of the entire text that the reader absorbed in a glance when reading, but that they may not consciously realize they understand without being asked to recall it.

Such claims were Wood's way of countering accusations that dynamic readers were not really reading because they could not recall specific words from what they had read (and often miserably failed standardized reading comprehension tests, which the company largely discouraged instructors from using). But she also, whether intentionally or not, repeats linguist John Firth's now-famous assertion that "you shall know a word by the company it keeps."¹⁵ Firth's insight about the context-dependent nature of linguistic meaning has been particularly important for natural language processing, where it is often referred to as the distributional hypothesis. This hypothesis suggests that words appearing in similar contexts tend to have similar meanings and it has been especially influential in relation to word embedding, a way of mathematically representing a word's meaning foundational to today's large language models.

And while Wood's idea that "loaded words" unlock the main ideas of a text is not the same thing as the distributional hypothesis, it is, like her description of reading as computational processing, analogous to it. For Wood, speed reading names a process not of reading individual words themselves more quickly - she didn't want her students to "word-read," as this slowed them down - but rather of learning to understand texts as something like distributions of words.¹⁶ In this imagination of reading, whole pages can be taken in in one glance because readers do not need to process the meaning of every individual word in order to understand what they are reading. Instead, readers can learn to perceive how meaning is distributed across a passage or a page as a totality, by allowing the words to "go in fast" and reading "from eye-to-mind directly."¹⁷ Like the understanding of language following from the distributional hypothesis, meaning in language for Wood arises not from the way a single word indexes something in the real world, but from the network of other signifiers in which it is suspended. In this sense, Wood's speed reader takes in something like a model of the text rather than reading "the text itself."

* * *

Of course, taking in words "from eye-to-mind directly" is not what happens when people encounter a page of text. People also knew at the time that this is not how human reading works, as attested to in the charges of fraudulence levied against speed-reading ventures by many academics, journalists, and pundits. It's therefore easy to dismiss Reading Dynamics as just one more grift. But even though the idea that people could be taught to read "computationally" may strike us today as ridiculous, it's worth reflecting on how much this imagination of optimized reading differs from dominant paradigms today. Today's attempts to optimize reading, described at the beginning of this essay, inch us ever closer to a world in which, instead of people reading like computers, all human reading is understood as replaceable by computation.

For Wood, the goal of teaching people to speed read was to encourage them to read more, and specifically to read more literature. "Most of us are interested in becoming conversant with great literature," she wrote in an early lesson plan draft, "but we just haven't had time. You are now at the threshold of a long-postponed opportunity. The time element no longer will force you to forego beneficial reading."¹⁸ In this statement and the many others she made like it, Wood posits that what keeps people from completing "beneficial reading" is their lack of time; speed reading fixes this problem, allowing for people to read "great literary masterpieces that [deserve] many re-readings."¹⁹

Sixty years later, Wood's confidence in the desire of her students to read literature perhaps strikes us as surprising. The case for learning to speed read to save time reading business reports, briefs, or required reading for school seems intuitive, and most speed-reading programs of the day focused their advertising on these use cases. But while Reading Dynamics mentioned such purposes in its advertising, Wood also continually insisted on the value of learning to speed read for the purpose of pleasure reading - specifically the reading of novels. The live demonstrations of speed reading the company designed for television featured novel reading and reading novels was an integral part of the program's curriculum. Whole lessons were devoted to understanding novels and students were encouraged to practice their techniques on novels first. Wood also claimed that learning to speed read would lead to greater enjoyment in reading, claiming that speed reading, once students mastered it, was a totally immersive experience. Though Wood may have been wrong about the technical possibility of speed reading, we might still learn something from the fantasy her approach encoded: optimizing reading meant a future of abundant pleasure reading.

Today's schemes to optimize reading have discarded this vision of abundance. Instead, companies attempt to sell the use of AI models on the promise that they will eliminate as much reading as possible. Adobe's claims about its "PDF AI Assistant," for instance, focus on its capacity to do away with reading. As with Microsoft's claims about Copilot, Adobe promises the model's "AI-Powered Summaries" will allow users to "quickly understand large or complex PDFs," "find a specific answer more quickly," and "extract key points."²⁰ These promotional materials do not ever mention human reading, one of the few things for which a pdf is designed. Instead, with AI assistants, people manage texts—they "understand," "find," "extract" information—but they don't read them. The endpoint of this understanding of optimized reading is the erasure of "manual reading" itself, and with it the understanding of reading as a labor of time, effort, and—yes—pleasure.

It's clear from Adobe's promotional materials that they do not think most of their customers are using Adobe for pleasure reading, however. The scenarios they envision for their PDF AI Assistant are all about work. But this is precisely the point: while Wood imagined that the horizon for speed reading stretched beyond the world of work into everyday life, for Adobe there is nothing beyond work. Wood tried to sell people on optimizing novel reading, on the idea that anyone could experience the joys of reading for fun, but faster. Adobe, on the other hand, simply assumes that its customers are only using its software to read for work (or school, which to Adobe is the same thing). Thus, the mainstreaming of AI tools for reading reminds us that, although people are reading (books, magazines, and newspapers) less than before, they are also reading (text) more than ever. This is because, today, all reading is imagined as work. Reading, for many white-collar workers, is what they do at work, as they manage the daily deluge of emails, Slack messages, slide decks, social media posts, meeting notes, reports, articles, vision statements, website content, etc. Like academics, these workers are paid to read; optimizing such reading is simply economical. While this vision of reading leaves pleasure out of the picture, companies like Adobe are betting their customers won't miss it. Who, in this economy, loves their job that much?

---

Lindsay Thomas (@lindsaythomas.net) is Associate Professor in the Department of Literatures in English at Cornell. She specializes in 20th- and 21st-century US literature, cultural studies, and the digital humanities. She is the author of Training for Catastrophe: Fictions of National Security after 9/11 (University of Minnesota Press, 2021) and winner of the 2022 ASAP Book Prize. She is currently working on a project about the history of "solutions" to the "problem" of novelistic length in the 20th- and 21st-century.

---

Banner image source: The Evelyn Nielsen Wood Photograph Collection, 1930-1980. MSS C 272, Box 1, Folder 41, "U Save Your Speed For Reading [billboard/2 copies]". Utah Historical Society, Salt Lake City, UT.

References

1. "AI summarizer with Copilot in Word," Microsoft, 2026, https://word.cloud.microsoft/create/en/ai-summarizer/. ↩︎
2. A (much) longer history of optimized reading might encompass phenomena as varied as the development of the codex format in the early Christian church, the use of early modern reference tools, or early-twentieth-century experiments in measuring eye movements. On these, see, for instance, Peter Stallybrass, "Books and Scrolls: Navigating the Bible," Books and Readers in Early Modern England: Material Studies, ed. Jennifer Andersen, Elizabeth Sauer, and Stephen Orgel (Philadelphia, PA: University of Pennsylvania Press, 2001), 43-79; Ann Blair, Too Much to Know: Managing Scholarly Information before the Modern Age (New Haven and London: Yale University Press, 2010); and Adrian Johns, The Science of Reading: Information, Media and Mind in Modern America (Chicago and London: The University of Chicago Press, 2023). ↩︎
3. "optimize, v." Oxford English Dictionary, Oxford UP, March 2025, https://doi.org/10.1093/OED/9386836696. ↩︎
4. "Instructor's Manual", Lesson six, n.d., 6, MSS B 272, box 8, folder 6, Evelyn Nielsen Wood Papers, ca. 1925-1979, Utah Historical Society, Salt Lake City, UT. ↩︎
5. Lindsay Thomas, "Speed Reading the Novel: Reading Dynamics and the Value of Reading at Midcentury," American Literary History vol. 37, no.2 (summer 2025): 370-399. ↩︎
6. Johns, The Science of Reading, 265. ↩︎
7. Lawrence Van Gelder, "Evelyn Wood, Who Promoted Speed Reading, Is Dead at 86," New York Times, August 30, 1995, Section D, 17. ↩︎
8. "Evelyn Wood Reading Dynamics Institute: Convention," June 12, 1965, 234-35, MSS B 272, box 18, folder 23, Evelyn Nielsen Wood Papers, ca. 1925-1979, Utah Historical Society, Salt Lake City, UT. ↩︎
9. "Evelyn Wood Reading Dynamics Institute: Convention," 238. ↩︎
10. "Evelyn Wood Reading Dynamics: Lesson IV Building Eye-Mind Reassurance," 1962, IJ-4-2, MSS B 272, box 8, folder 18, Evelyn Nielsen Wood Papers, ca. 1925-1979, Utah Historical Society, Salt Lake City, UT. ↩︎
11. "Instructor's Enrichment Material," 1962, 5, MSS B 272, box 8, folder 2, Evelyn Nielsen Wood Papers, ca. 1925-1979, Utah Historical Society, Salt Lake City, UT. ↩︎
12. "Evelyn Wood Reading Dynamics Institute: Convention," 236. ↩︎
13. "Evelyn Wood Reading Dynamics Institute: Convention," 249. ↩︎
14. "Addenda to Lesson Plans," n.d., 1, MSS B 272, box 8, folder 7, Evelyn Nielsen Wood Papers, ca. 1925-1979, Utah Historical Society, Salt Lake City, UT. ↩︎
15. J. R. Firth, "A Synopsis of Linguistic Theory 1930-1955," Studies in Linguistic Analysis, ed. J. R. Firth (Oxford: Basil Blackwell, 1957), 11. ↩︎
16. "Evelyn Wood Reading Dynamics Institute: Convention," 236. ↩︎
17. "Evelyn Wood Reading Dynamics: Lesson IV Building Eye-Mind Reassurance," IJ-4-2. ↩︎
18. "Instructor's Manual", Lesson six, 11. ↩︎
19. "Instructor's Manual", Lesson six, 12. ↩︎
20. "PDF AI," Adobe, 2026, https://www.adobe.com/acrobat/online/pdf-ai.html. ↩︎