Writing with AI, not by AI

Brainstorm, outline, get feedback and edit, and keep your voice. About 45 min, beginner level.

You'll be able to

Writing is thinking

When you write, you find out what you actually think: which claim you believe, which evidence holds, where the argument has a gap. If an AI drafts it, you get fluent text without doing that work, and in a course, that work is usually what's being assessed.

So the most useful role for AI in writing is editor and sparring partner, not author. A workflow that keeps you in charge:

The same pattern works for a seminar paper, a grant section or a staff report. What changes is the policy you're working under.

Ask for critique before rewrites

If you ask an assistant to “improve this paragraph”, it rewrites it, and you lose track of what was wrong and what's now yours. Asking for a diagnosis first keeps you learning and keeps the words yours.

Prompts that get useful critique:

Give context: the audience, the assignment and what the piece needs to do. Then revise yourself, and ask again to see if the problem is fixed.

Assistants tend to be agreeable and can praise weak work. Asking explicitly for the weakest point, or for three problems, counters that.

Keep your voice and your meaning

Editing tools make text smoother, and smoothness has costs.

A good habit: when a rewrite is better, work out why (a shorter subject, a stronger verb, one idea per sentence) and apply the lesson yourself next time. Then the tool is teaching you rather than replacing you.

Know the rules and keep a record

Before you start, read the AI policy for your course, journal or office. Policies differ: one instructor allows brainstorming but not drafting, another allows proofreading only, a third asks for AI use to be documented.

Keep a simple record as you go: which tool, what you asked, and what you kept. Save the key prompts. Grammarly's Authorship feature can document your process if an instructor asks for it.

A useful disclosure says three things: what you used, for what, and how you checked it. For example: “I used Claude to critique two drafts for unclear claims and counterarguments; all wording and revisions are mine.”

For researchers, publishers and publishing-ethics bodies such as COPE hold that AI tools can't be listed as authors, because they can't take responsibility for the work. Whatever you submit, you're accountable for every sentence in it.

A tough reviewer, then your revision

Your draft: “Remote work is better for everyone. Studies show people are more productive at home, and companies save money on offices, so all universities should let staff work remotely.”

Prompt: “Act as a skeptical reviewer for a staff policy memo. What's the weakest claim here, and why? Don't rewrite it.”

Critique (summarized): “Better for everyone” is unsupported; roles differ, and some depend on being on site. “Studies show” names no studies, and the evidence on productivity is mixed. The jump to “all universities” doesn't follow.

Your revision: “For roles that don't require being on campus, a hybrid option could help staff focus and reduce commuting. Before adopting it, we should review evidence on productivity in comparable roles and ask each department which tasks need people on site.”

What changed: a narrower claim, a plan for evidence instead of a vague appeal to it, and the wording is still yours. Ask the reviewer again: the next weakest point is a smaller one.

Try it

Your original paragraph, the AI's critique, your own revision, the second round of critique, and a one-sentence disclosure of how you used AI.