The Palimpsest (Manuscript Evidence)
In McGann's framework, the manuscript palimpsest is not merely a record of how a text came to be; it is itself a text, carrying meanings that the published version systematically erases. The crossed-out word tells the reader something the chosen word alone cannot convey. The marginal insertion illuminates the path not taken. The change in pen or ink records the passage of time and the shift in perspective that time produces. The physical evidence of a writer's hand pushing language toward clarity is an archive of the creative process that scholarly editing has long valued as a source of interpretive insight. AI-assisted writing produces a different kind of record — a conversational archive that is complete in ways traditional manuscripts are not, but clean in ways that lose the material traces of embodied struggle.
The Infrastructure of Erasure — Contrarian ^ Opus
There is a parallel reading that begins from the material conditions of AI systems themselves. The conversational record between human and AI may appear complete, but it exists within proprietary platforms that can alter, delete, or make inaccessible their archives at any moment. Where manuscript palimpsests survived centuries in libraries and private collections, becoming public patrimony that scholars could access freely, AI conversation histories live on corporate servers subject to terms of service, data retention policies, and the economic viability of their host companies. The scholar studying Keats's manuscripts need only gain access to a physical archive; the scholar studying AI-assisted composition must hope the platform still exists, the account remains accessible, and the company hasn't pivoted its business model.
More fundamentally, the AI conversation record is already a palimpsest of erasure at the model level. Each response emerges from training data that has been deduplicated, cleaned, aligned, and filtered through multiple stages of processing that leave no trace in the final output. The model itself is a black box whose internal states during generation remain opaque. Where the manuscript palimpsest preserves even what the author tried to erase, the AI system operates through systematic forgetting — discarding the vast majority of its computational work, preserving only the polished surface. The future scholar will inherit not a complete record but a curated performance, missing both the material struggle of human composition and the computational struggle of machine generation. The cleanliness McGann identifies is not just an absence of handwriting but an active suppression of process at every level, from training data to inference, leaving only the illusion of transparent dialogue.
In the AI Story
The traditional palimpsest preserves specific kinds of information that the published text erases. The heaviness of a cancellation mark communicates the decisiveness of a rejection. The tentative quality of a marginal insertion — written smaller, in a different ink, squeezed into the margin — communicates uncertainty. The shift in handwriting pressure across a page records the writer's emotional state over time. These are material features, not linguistic ones, and they carry meaning that the cleaned-up published version cannot.
McGann's scholarship on Keats, Byron, and Rossetti has consistently treated manuscript evidence as primary interpretive material. The manuscripts show poets in constant negotiation with their own language — trying alternatives, returning to earlier choices, layering revisions until the surface becomes a dense record of compositional struggle. Reading this struggle is reading the poem more fully than the published text alone permits.
The conversational record of an AI collaboration is, in some respects, richer than a traditional manuscript revision history. The sequence of prompts and responses is captured completely, with timestamps, in a format that shows the iterative back-and-forth in real time. Future scholars studying AI-assisted texts will have access to process documentation that Keats scholars can only approximate.
But the conversational record is also clean. It lacks the material residue that gives the traditional palimpsest its depth. There is no handwriting to analyze, no pressure variations revealing emotional state, no shifts in pen or ink marking the passage of time. The record preserves the decisions but not the embodied process of deciding. The hesitation, the false start, the physical evidence of a hand hovering over a choice before making it — these are absent from the AI collaboration's archive.
The loss is not catastrophic and does not require a nostalgic return to longhand composition. But it is worth recognizing. The material dimension of textual production, which McGann's scholarship has treated as essential interpretive material, is reduced in AI collaboration to a different kind of record — one that captures sequence but not embodiment, decision but not struggle, outcome but not the physical labor of arriving.
Origin
McGann's treatment of manuscript evidence developed through his editorial work on Byron and Rossetti, where he argued against the intentionalist tradition's practice of using manuscripts only to reconstruct a final authorial version. His alternative treated manuscripts as texts in their own right, worth reading for what they reveal about the compositional process.
Key Ideas
Manuscripts as texts. The palimpsest is not merely a source for reconstructing the published version; it is a text in its own right with its own interpretive value.
Material residue as meaning. Handwriting, ink changes, pressure variations, and physical layout carry interpretive information that the published text loses.
Temporal depth. Traditional manuscripts record the writer's changing relationship with the text over time; AI conversations do not capture temporal depth in the same way.
AI record richer in sequence. The conversational archive preserves the explicit dialogue of collaboration more completely than manuscript evidence ever could.
AI record cleaner in material terms. The absence of embodied material residue eliminates a dimension of process evidence that scholars have long found interpretively valuable.
Appears in the Orange Pill Cycle
Scales of Compositional Memory — Arbitrator ^ Opus
The tension between manuscript palimpsest and AI conversation record resolves differently depending on what we're trying to preserve. For capturing deliberate compositional choices — the explicit alternatives considered, rejected, revised — the AI record is unquestionably superior (90% AI advantage). Every prompt variation is logged, every response preserved, creating a sequential archive that manuscripts only approximate through their layered revisions. McGann's focus on reading the manuscript as text remains valid, but the AI conversation literally is a text of negotiation, not merely material evidence of one.
For preserving embodied process and temporal depth, however, the manuscript palimpsest remains irreplaceable (80% manuscript advantage). The contrarian view correctly identifies that AI systems operate through systematic erasure at multiple levels — from training data processing to model inference to platform control. A shaking hand, a wine stain, a different ink after lunch — these record not just what was thought but how thinking felt in the body over time. The manuscript preserves involuntary evidence; the AI conversation preserves only what was deliberately typed.
The synthetic frame might be "scales of compositional memory." Manuscripts excel at preserving what we might call intimate memory — the personal, embodied, unintentional traces that reveal the writer as a physical being struggling with language. AI conversations excel at preserving dialogical memory — the explicit back-and-forth of collaborative thinking. Neither is complete; both are palimpsests, but they preserve different layers. The question isn't which archive is richer but which questions each archive can answer. Future scholars will need both: the manuscript tradition's attention to material evidence and new methods for reading the computational archive's different kind of depth.
Further reading
- Jerome McGann, The Textual Condition (Princeton, 1991)
- Robin G. Schulze, The Degenerate Muse: American Nature, Modernist Poetry, and the Problem of Cultural Hygiene (Oxford, 2013)
- Hans Walter Gabler, ed., Contemporary Collaborative Writing (Cambridge, 2015)
- Matthew G. Kirschenbaum, Track Changes: A Literary History of Word Processing (Harvard, 2016)