Output Interrogation
Output interrogation is the disciplined practice of reading AI-generated output with active critical engagement — treating each output not as a finished artifact to be consumed but as a provisional hypothesis to be tested. The practice integrates the Renaissance ars critica with the specific demands of AI-era content: distrust of fluency, attention to invisible omissions, independent verification of connections, and deliberate resistance to the smooth surface that induces premature acceptance. Ann Blair's framework treats output interrogation as one of the four core operations of AI curatorial practice (alongside prompting, selecting, and integrating), and as a specific skill that can be taught, refined, and institutionalized.
In the AI Story
The practice begins with the recognition that AI output is provisional by nature. Every output is a draft for evaluation, not a finished product for acceptance. This disposition must be maintained across many iterations per hour, for hours per day — a sustained effort that constitutes the specific labor AI collaboration demands.
The operations of interrogation include fact-checking (claims against sources), logical analysis (does the argument hold?), identification of omissions (what is missing?), verification of connections (is this association real or merely statistical?), and stylistic evaluation (does the voice serve the work or impose an alien register?). Each operation draws on capacities that the Renaissance ars critica cultivated, adapted to the distinctive features of AI-generated content.
The practice also includes meta-level operations that the ars critica did not need. Because AI output is shaped by prompts, the interrogator must ask whether the output's deficiencies reflect genuine limitations in the AI or inadequate prompt design. Because AI systems vary in capability and specialization, the interrogator must calibrate her scrutiny to the specific system. Because AI output is produced on demand, the interrogator can often request alternative formulations — an option that creates both opportunity and distraction.
The affective dimension is crucial. Output interrogation must be sustained against a surface designed to reduce the sense that sustained attention is necessary. Fluent, organized, confident output induces the reader to accept it; interrogation resists the induction. The resistance is cognitively expensive and cannot be automated by the tools that produced the output. It is the specific form of ascending friction that AI collaboration produces for the evaluator.
Origin
The practice synthesizes Blair's historical work on critical reading with contemporary experience of AI collaboration. It has antecedents in every previous tradition of disciplined reading, adapted to the specific features of the AI medium.
Key Ideas
Output as hypothesis. Every AI output is provisional; acceptance requires verification, not the reverse.
Integrated operations. Fact-checking, logical analysis, omission-finding, connection verification, and stylistic evaluation are components of a single disciplined practice.
Meta-level questions. The interrogator asks about the prompt, the system, and the possibility of alternative formulations.
Sustained against surface induction. The practice resists the smooth surface's pull toward premature acceptance.
Irreducible human labor. Output interrogation is the specific labor AI tools create and cannot themselves perform.
Appears in the Orange Pill Cycle
Further reading
- Ann Blair, Too Much to Know (Yale, 2010).
- Anthony Grafton, The Footnote (Harvard, 1997).
- Edo Segal with Claude Opus 4.6, The Orange Pill (2026).