In this experiment, drafts—human-written or AI-generated—anchored the final essay, with AI-generated drafts revised somewhat more but underlying phrases largely retained across conditions.¶

Claim · All claims
Evidence · 1 study · causal r2 · q3
Currency · no dated study recorded · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? Draft-to-final 3-gram retention was ≥93% in every condition and 5-gram retention dipped to ~88% only in ai draft, indicating AI support shifted how verbatim carryover was rather than whether content flowed through. → ngram-retention

Evidence¶

ngram-retention¶

Katy Ilonka Gero, Tao Long, Carly Schnitzler, and Paramveer S. Dhillon. 2026. From Planning to Revision: How AI Writing Support at Different Stages Alters Ownership. In Designing Interactive Systems Conference (DIS '26), June 13–17, 2026, Singapore, Singapore. ACM, New York, NY, USA. https://doi.org/10.1145/3800645.3813003

q3 · i? · causal · r2

N-gram retention analysis (Table 2) across the n=253 experiment's four conditions: "people did not revise too much in any condition; when AI provides the draft, people do revise a bit more (lower 5-gram), but the underlying phrases remain largely intact (still high 3-gram)." Reported as descriptive telemetry, not model covariates.

"In other words, people did not revise too much in any condition; when AI provides the draft, people do revise a bit more (lower 5-gram), but the underlying phrases remain largely intact (still high 3-gram)."

Discussion¶

Related Claims¶

• Essay quality was highest for AI-generated drafts and lowest with no AI, negatively correlated with ownership, suggesting an ownership-quality tradeoff. — related
• AI support shifts perceived attribution of ideas and text toward the AI, most strongly for AI-generated drafts, and an AI draft based on the writer's own outline contributed more ideas than expected (27%). — related
• AI support during drafting lowers ownership substantially more than support during planning or revision, and planning support preserves more ownership than revision support. — related
• Engage-to-Unlock participants incorporated less verbatim AI-response text into final essays than Standard Chatbot participants — related
• Essays within each tool group showed consistent homogeneity across named entities, n-grams, and topic ontology — related
• Essays written under Standard Chatbot sat closest to AI-only essays in semantic embedding space, and Engage-to-Unlock differed significantly from Standard Chatbot along PC1 — related