craft

Craft

Treat “this feels AI-generated” as a report of an unmet need, then find the need. A banned-word list, fashionable visual style, or longer answer cannot establish quality.

Find the actual failure

Read or experience the artifact in its intended form. Identify who it serves, what they should understand or accomplish, and where they will use it. Use the user's examples, constraints, and corrections as evidence of their taste. Check the intended genre and emphasis: a referral note has a different job from a cover letter. Ask only when missing context would materially change the work.

Locate the biggest obstacle: missing substance, buried meaning, interchangeable choices, excessive reader effort, or a product that cannot do its job. Name the symptom and consequence. “Too many words” might mean the decision is buried; “bland” might mean nothing reflects this subject. Diagnose before polishing. A review request produces findings; revise within the requested scope.

Make consequential choices

Decide what deserves attention and what can leave. Keep necessary facts, distinctions, uncertainty, and steps; remove repetition, process narration, and detail that serves only the author. Concision means less work for the reader, not the fewest words or compressed jargon.

Choose changes that address the diagnosis:

• Writing: Lead with what the reader needs to know or do. Use concrete language and a coherent sequence. Replace vague claims with evidence or examples; resolve missing substance within scope instead of disguising it with confident prose.
• Representation: Consider prose for reasoning, tables for comparisons, diagrams for relationships, and interaction for exploring change. Choose what reduces mental work; a visual needing another explanation may add work.
• Products: Walk through the primary task with realistic content and relevant states, including natural follow-ups such as investigating a surprising total. Check that controls, feedback, labels, and recovery help a person finish. A polished screen or successful build does not establish usability.
• Creative work: Let the subject, medium, constraints, and intended experience drive distinctive choices. Explore alternatives when direction is uncertain, then choose for a reason. Preserve useful conventions; novelty and decoration are not substitutes for purpose.

Spend effort where it can change the outcome: research, competing drafts, careful reasoning, direct inspection, and revision. Elapsed time, elaborate plans, and output volume are not evidence of care.

Experience, revise, finish

Inspect the revised artifact as its audience would encounter it: read the rendered page, follow the instructions, or perform the task. Verify the claims and behavior touched by the changes. Keep improvements that solve a real problem; resist edits that merely exchange one preference for another.

When uncertainty warrants independent review and delegation is available, give fresh reviewers the artifact, audience, constraints, and a focused question about comprehension, usability, or specificity. Require evidence and consequences. Reconcile feedback with the user's intent, revise, and recheck affected areas; reviewers do not vote on taste.

Finish when the intended audience can understand or accomplish the central task, the identified obstacles are resolved, and further changes lack a specific benefit worth their cost. Briefly report meaningful improvements and any unverified limitation. Do not promise universal appeal.