1. Framing capture
The model gives the first plausible thesis. The writer begins editing it rather than discovering an independent position.
Countermeasure: state your own provisional claim before asking for alternatives.
2. Fluency laundering
Weak inference becomes persuasive because the model supplies smooth causal language.
Countermeasure: separate every claim into evidence, mechanism, and uncertainty before prose editing.
3. Framework laundering
The model maps a messy situation to a familiar concept, such as Goodhart's law, the sunk-cost fallacy, or the principal-agent problem. The label feels like explanation.
Countermeasure: remove the framework name and state the physical or economic mechanism.
4. Source drift
A summary preserves the gist but drops a condition, an exception, or a speaker attribution.
Countermeasure: keep source claim, evidence, and your interpretation separate, and audit against the originals.
5. Synthetic voice
AI supplies polished transitions, aphorisms, and symmetry until the document sounds coherent but less like the person who thought it.
Countermeasure: protect original phrases, examples, confidence, and point of view. Make local edits.
6. Premature structure
A neat outline arrives before the material has been interrogated. Discovery starts filling boxes.
Countermeasure: do not create a Reader Architecture during the private inquiry stage.
7. False completeness
AI rarely feels stuck. It fills missing evidence, motives, causal links, and examples with plausible material.
Countermeasure: force explicit "unknown", "missing", or "cannot infer" states.
8. Version abundance
Generating ten alternatives feels like progress, but it creates selection work and weakens commitment.
Countermeasure: diagnose the exact problem before generating any alternative. Ask for one or two materially different approaches only when a choice is real.
9. Polishing sunk cost
AI makes a weak paragraph attractive. The writer becomes less willing to delete it.
Countermeasure: run structural and reader-response passes before prose work.
10. Outsourced taste
Repeatedly asking the model what "good writing" is can replace the slow formation of personal taste.
Countermeasure: read strong writers, compare passages, state why one choice works, imitate techniques in exercises, and decide before asking AI.
The blank-page compromise depends on the task
For some tasks, AI-generated starting material is useful: routine correspondence, generic documentation, standard summaries, or formats where authorship is not the learning objective.
For personal essays and serious thinking, the cost is higher. The blank page is part of the exercise.