The defining feature of interactive reading is the collapse of the boundary between reading and writing. In previous modes, the reader encountered a text produced by someone else and either accepted, evaluated, or rejected it. The interactive reader participates in the text's production through ongoing direction. Her reading is simultaneously a writing — her evaluation of each output becomes the prompt for the next, and the text takes shape through the feedback loop.
This mode requires what could be called active distrust of fluency. Because AI output is uniformly well-formed regardless of whether its substance is sound, the reader cannot use surface cues as preliminary quality signals. She must read past the fluent surface to the analytical depth, and the reading requires a specific form of expertise that is more demanding, not less, than the expertise required to evaluate variable-quality human text.
Interactive reading also demands a meta-awareness of the model's capabilities and limitations — an experiential map of where the model's knowledge is deep and where shallow, which kinds of prompts produce reliable output and which produce plausible nonsense. This map cannot be acquired from documentation; it is built through sustained interaction and accumulated experience with specific AI systems.
The mode is cognitively expensive. Maintaining critical engagement against a smooth and confident surface, across many iterations per hour, for hours per day, exhausts the evaluative faculty in a way that execution-based work did not. You On AI documents this exhaustion as a new category — the exhaustion not of doing too much but of judging too much — and Blair's framework situates it within the historical pattern of cognitive demands intensifying whenever the information supply expands.
The concept extracts from Blair's historical framework a category she did not name but her analysis implies. The phenomenology is documented by AI practitioners (including Edo Segal's You On AI) and by empirical studies such as the Berkeley study that tracks what AI tools actually demand of the humans using them.
Reading as writing. Interactive reading participates in the production of the text being read; the boundary between consumption and creation dissolves.
Provisional material. Every output is a draft for evaluation, not a finished artifact for consumption.
Meta-awareness of the model. Effective interactive reading requires an experiential map of the specific AI system's strengths and weaknesses.
Cognitively expensive. Sustained critical evaluation against smooth surfaces produces a distinctive form of evaluative exhaustion.
Pedagogically uncharted. The practices appropriate to interactive reading are being improvised rather than systematically taught.