Keep responsibility for claims, context, examples, reader needs and publication decisions. AI assistance does not remove the need for accountable editing.
Rather than predict which capabilities AI can never acquire, focus on the responsibilities you retain when you publish something under your name.
Verify consequential claims
Decide which statements need evidence and check them against sources. An assistant can help organize a claim list, but a plausible citation is not a completed fact-check.
Supply the missing context
Explain who the reader is, what they already know and what constraints matter. A technically correct instruction can still be unhelpful if it assumes software or permissions the reader does not have.
Contribute honest observations
Use tests, interviews or experiences you actually conducted and have permission to share. Label illustrative examples. Do not manufacture a personal story merely to make a draft feel original.
A bounded observation
“This sample contains a tab before the name” describes something inspectable. “This tool saves every writer hours” requires evidence that a single sample cannot provide.
Judge usefulness and trade-offs
Ask whether a proposed shortcut could remove meaningful information. A whitespace cleanup can improve prose while damaging code indentation. A shorter translation can accidentally remove a qualification.
Own publication and corrections
Decide when a draft is ready, explain its limitations and provide a way to report mistakes. Google's guidance on generative AI content makes accuracy and review relevant even when generation is assisted.
These are practical responsibilities, not a claim that humans always outperform models or that a particular writing style proves authorship.
For a concrete review exercise, inspect the formatting methodology, then run a harmless sample through the spacing cleaner. Compare the actual output with the claimed behavior before recommending the tool to someone else.