AI Is the Historiographical Wave of the Future

Used properly, LLMs will enable historians to focus more than ever on the truly important stuff: finding the archival nuggets and placing them in historiographical context

A unfortunately common view appears to be developing within some quarters of the historical profession — a collective side-eye aimed at the arrival of Large Language Models. Bubbling up on academic social media and in the common rooms of university departments, this sentiment ranges from weary skepticism to outright hostility. LLMs, the thinking goes, are glorified plagiarists, serial hallucinators, and tools for intellectual laziness, antithetical to the historian’s core commitments to archival integrity, nuanced interpretation, and the slow, deliberate craft of reconstructing the past. As one recent study put it, “The main takeaway from this study is that LLMs, while impressive, still lack the depth of understanding required for advanced history. They’re great for basic facts, but when it comes to more nuanced, PhD-level historical inquiry, they’re not yet up to the task.” In sum, the prevailing fear is that LLMs represent a fatal shortcut, a technological temptation that will lure the next generation away from the virtues of deep reading and archival grit. (Not all historians feel this way of course. The Verge recently had a good article on the positive uses of LLMs in historical research, and there are various other examples.)

This chorus of rejection is understandable. It is also, I believe, profoundly mistaken. It mistakes the tool for the user, confuses a nascent technology’s current limitations with its ultimate potential, and, most critically, misidentifies the true threats to our profession. The hand-wringing over LLMs is a distraction from the real crises: the collapse of funding for research, the systemic evisceration of the research university for ideological reasons, and a broader cultural disinterest in the past that threatens to render our work irrelevant. Far from being a threat, the thoughtful integration of tools like LLMs isn’t only essential for the future of historical practice but will, in fact, make us better historians.

We’ve been here before

If I am myself deeply skeptical of these rejections, it is because I have lived through a cognate if much smaller-scale technological revolution in historiographical methods. The anxieties of today rhyme uncannily with the ones I remember from the dawn of my own career. In 1995, I was a graduate student embarking on my dissertation, an intellectual history of modernization theory that would eventually become my first book, Mandarins of the Future. My research plan was, by today’s standards, pathetically analog. As endorsed by my advisors, this plan entailed spending the better part of a year, likely more, in the library stacks, physically pulling volumes of academic journals from the shelves. My target list included a couple of dozen flagship publications in economics, political science, and sociology. I intended to read through more than fifty years of their back catalogs — from the 1930s to the 1980s — page by page, to manually trace the flow of ideas about “modernization.”

It was a daunting, antiquarian task. Yet, it was the only way to do it — and I was psyched for it! How else could one get a feel for the intellectual currents of an era, for the shifting salience of a concept, except by immersing oneself in the entire conversation as it unfolded chronologically?

But 1995 was also the year that a new technology began to reshape the landscape of academic research: JSTOR. It started small, digitizing the back catalogs of a handful of the most important journals in economics and history — the American Economic Review, the Journal of Political Economy, the American Historical Review, the Journal of Modern History — and making them searchable.

For a budding historian, it was a miracle. Suddenly, my year-long pilgrimage through the stacks was rendered obsolete. Instead of manually flipping through half a century of the American Political Science Review, I could simply type “modernization” into a search box, and read just the relevant articles. Moreover, I could quantitatively track the rate of use of the term, and its diffusion in time across different journals and authors. The results were instantaneous and transformative.

Now, an antiquarian purist might argue, with some justification, that this shortcut cost me something. By not reading these journals cover to cover, I undoubtedly missed the broader context, the serendipitous discoveries, the panoramic view of how these disciplines evolved. This is no doubt true.

But what I gained was immeasurably greater. The search results immediately revealed a crucial pattern that would have taken me months, if not a year, to discern the old way. Despite the fact that the most famous avatar of modernization theory was an economist, Walt Rostow, the quantitative data from JSTOR showed that the real intellectual energy around the concept wasn’t in economics at all. The keyword “modernization” appeared with far greater frequency and urgency in the flagship journals of sociology and political science. The core of the debate was taking place among sociologists in the orbit of Talcott Parsons and Edward Shils, and political scientists surrounding Gabriel Almond and Lucian Pye. This realization, which took mere days to crystallize with JSTOR, allowed me to redirect my research efforts with surgical precision. It soon pointed me toward more specialized publications, like Comparative Studies in Society and History, which proved to be a far richer vein of material.

This efficiency wasn’t about laziness; it was about the reallocation of intellectual and temporal resources. The months saved from the drudgery of combing through irrelevant economics journals were months I could now spend on more vital tasks. I had time to read much more deeply in the broader literature of postwar American liberalism, a context that proved decisive for my eventual argument. The central thesis of Mandarins of the Future — that the tidy division between the “good” domestic liberalism of the New Deal and the Great Society and the “bad” foreign policy of Cold War liberalism was a self-serving fiction — was an insight born directly from this technologically-enabled research path. JSTOR didn’t do the thinking for me, but it cleared the intellectual underbrush, allowing me to see the forest for the trees. It facilitated a more focused, more profound, and ultimately more original argument. Was this a bad thing? I don’t think so.

The parallel to our present moment with LLMs is obvious. The fear that they will make us lazy, that they will supplant “true” intellectual labor, is the same fear that greeted the searchable database, albeit greatly amplified. Of course there will be people who use these new technologies poorly. There will be students who prompt an AI to write a passable, soulless essay. There will be scholars who use them to generate superficial analyses of complex topics. But this is a critique of the user, not the tool. Those of us avocationally committed to the craft of historical research should consider LLMs not a threat but a thrilling new frontier. They will enable entirely new approaches to the archive and new methods of synthesizing data, leading to fresh and surprising understandings of the past. The fundamental work of the historian — the curiosity that drives us to ask new questions, the intuition that guides us toward a hidden story — isn’t going away. It is simply being augmented.

The true challenges facing our profession lie not in the code of AI, but in the spreadsheets of university administrators and the priorities of a culture that is losing its historical memory. The real problems are the collapse of research funding that makes ambitious archival work an unaffordable luxury for many; the broader evisceration of the research university, which is being if not destroyed then repurposed to serve exclusively corporate ends; and, perhaps most chillingly, the general barbarization implied by a growing societal disinterest in the past. In this environment, to fixate on the supposed dangers of LLMs is to fiddle while Rome burns. We should instead be asking how these tools can help us fight the fire.

The historiographical methods of the future

So what does the future of the historian’s craft look like in the age of LLMs? I believe it will be defined by a strategic re-focusing of our efforts, guided by three core principles.

First, the real, irreplaceable value of the historian will increasingly be found in the time spent trolling in the physical archive. This may seem paradoxical in a digital age, but it is precisely the explosion of digital tools that will make our time in the dusty boxes of manuscript collections more important, not less. The vast, overwhelming majority of historical data — letters, diaries, ledgers, photographs, memos, ephemera — is not, and likely never will be, digitized. It exists on paper, on film, on tape, in a thousand scattered repositories. The only way this material will ever enter the digital conversation is if an historian, guided by expertise and intuition, bothers to spend time in those archives, recognizes a document’s potential significance, and makes the effort to bring it into the light. The historian of the future will be, more than ever, a treasure hunter, the essential human interface between the analog past and the digital present, the one who selects the raw material to feed into the machines. This also implies new possibilities for team-based historical research of a sort that has long been lamentably rare among historians, who have traditionally been solo foragers.

Second, once this material is digitized, LLMs will become indispensable tools for both analysis and composition. Their true power for historians will lie, first, in pattern recognition across vast documentary corpora, and second, in the generation of first drafts of hypotheses. Imagine feeding an LLM the complete digitized correspondence of a dozen key figures from the Harlem Renaissance, thousands of letters in total. A human could spend a decade reading this material. By contrast, if fed the corpus and given a good prompt, an LLM could, in minutes, map the social network, identify shifts in language and sentiment over time, track the dissemination of specific ideas, and flag previously unnoticed connections between individuals. It could perform a quantitative analysis of rhetoric that would be near-impossible at human scale, revealing the hidden structures of an intellectual movement.

Having identified these patterns, the historian could then use the LLM for a second task: drafting a preliminary narrative. One could prompt it: “Based on the attached correspondence, write a 1500-word summary of the evolving debate about Marcus Garvey’s influence between 1922 and 1926, highlighting the shifting positions of W.E.B. Du Bois and Alain Locke.” The output of course wouldn’t be the final word, but rather a structured starting point, a first draft of a hypothesis based on the data. It would be a scaffold upon which the historian could build a more nuanced and sophisticated argument. Not a final product, but a feisty provocation.

This leads to the third and most crucial point: the human historian in this equation doesn’t disappear. On the contrary, her effort is elevated and re-focused on the tasks where human intelligence is paramount. The historian’s craft will shift. More time will be spent in the archives, performing the irreplaceable work of discovery. With LLMs handling the more mechanical aspects of data processing and initial synthesis, more time will be freed up for deep, contemplative reading in the historiography, and thus the essential work of understanding how new archival findings can be made meaningful in conversation with existing scholarship.

The historian’s new skillset will include the art of crafting careful prompts — the ability to ask the machine the right questions, to guide its analysis, to challenge its interpolations. This isn’t just a technical skill but also an intellectual one, rooted in deep subject matter expertise. And finally, the historian’s most important role will be as editor and author, taking the raw output of the LLM and rewriting it, infusing it with her distinctive voice, interpretive insight, and the profound, intuitive understanding of meaning and import that comes only from years of study. Among other things, it will mean a taste for the juiciest quotes from the archive. The machine can find the pattern; the human task will be to pick the best notes and make it sing.

In short, independent environmental and military historian Megan Kate Nelson is right, but for the wrong reason.

Nelson is right that those are the reasons to become a historian, but she’s wrong that LLMs will compromise that. Used properly, they should make historical research and writing more fun and more of a challenge!

In sum, the specter of the lazy LLM-reliant historian, content to let the machine do the work, is a phantom. The real future of our discipline will belong to the ambitious historian, the one who sees in this new technology not an excuse for indolence, but a powerful lever. By outsourcing the drudgery amd distractions, we free ourselves for the deeper work. By embracing new ways of seeing the past, we stand a better chance of making it matter to the present. The choice isn’t between the antiquarian’s craft and the machine’s logic. It is between a fearful retreat into a nostalgic vision of our profession and a courageous embrace of a future where our oldest skills are augmented, sharpened, and made newly powerful by the tools we are only now beginning to understand.