Most advice about AI writing tools is about choosing one. This article is about what happens after you have chosen: how to work with an assistant so that the result still sounds like you, says things that are true and respects whoever will read it. The workflow below applies to chat assistants and to dedicated editors alike, and it is built around one idea. The assistant is good at producing options and at spotting problems. You are the one who knows what you mean.

If you are still deciding on a product, start with our buyer’s checklist for AI writing assistants. The steps here work with any of them.

Step 1: Write the brief before you prompt

Good drafts begin with a clear brief. Before you ask for a word, write down four things in plain notes: who the reader is, what you want them to do or understand after reading, the single main point, and the facts you must include. A reader-first brief is the difference between an assistant that returns generic prose and one that gives you something to build on. Model makers’ own guides, such as Anthropic’s prompt engineering overview and OpenAI’s prompt engineering guide, agree on this: be specific about the task, supply context and say what the output should look like.

Step 2: Draft the skeleton yourself

Resist the urge to ask for a full article. Write your own outline in a few bullet points, with the claims you want to make in your own words. This preserves your structure and argument, which is the part readers experience as your voice. Then ask the assistant to expand one section at a time, giving it your bullet and a sample of your own writing to match.

A useful prompt: “Here is a paragraph I wrote last month so you can see my style. Using it as a guide for tone and sentence length, expand the following three bullets into a 150-word section. Do not add facts that are not in the bullets.”

Step 3: Ask for options, not answers

Assistants vary their wording between runs, which is a feature when you need alternatives. Ask for three versions of an opening line, or five ways to say a technical point more simply, and pick the one that sounds like you. Our explainer on temperature and sampling shows why the same prompt returns different text. If you see the same stock phrases, ask the model to avoid them and name the ones you dislike.

Step 4: Edit in passes

Layer your edits rather than fixing everything at once.

  1. Structure pass. Does each paragraph make one point? Is the order logical? Ask the assistant to summarize each paragraph in one line and check that the summaries tell your story.
  2. Voice pass. Read aloud and replace anything you would never say. Cut hedges and filler.
  3. Clarity pass. Ask for the three sentences that are hardest to follow and rewrite them yourself.
  4. Line pass. Use a grammar tool or assistant for spelling, punctuation and consistency. Accept suggestions one by one, and decline those that flatten your tone.

Step 5: Check every fact yourself

Assistants can state wrong things confidently, a failure known as hallucination. Never accept a number, date, quote, citation or name without checking it against a primary source. If the assistant gave you a reference, open it and confirm that the page exists and says what the text claims. Delete anything you cannot verify. When you quote someone, quote their words exactly and link the original.

Step 6: Decide what to disclose and what to share

Check the rules that apply to you. Schools, employers, publishers and some platforms set policies on AI-assisted writing, and they differ. Follow them, and when in doubt, ask. Also think about what you paste in. Many tools keep prompts, and settings for training use vary by product and plan; check the vendor’s privacy page before you paste a client contract, a draft under embargo or personal data. Our guide to writing system prompts shows how to set standing instructions so you do not repeat sensitive background in every chat.

Common mistakes

  • Asking for the whole piece first. You end up editing someone else’s structure.
  • Accepting every suggestion. Grammar tools optimize for a generic standard English; your idioms may be the point.
  • Skipping the read-aloud. Rhythm problems are easiest to hear.
  • Trusting invented sources. A convincing citation is not evidence.
  • Padding. Longer is not better; use the assistant to cut as often as to expand.

A sample prompt kit

  • Tone match: “Rewrite this paragraph to sound like the sample, keeping every fact unchanged.”
  • Shortener: “Cut this by 30% without removing any claim. Show what you removed.”
  • Critic: “Act as a skeptical editor. List the three weakest claims and what evidence each needs.”
  • Reader test: “Explain this section back to me as a busy new employee would understand it. Where did you get confused?”

What to watch next

  • Editors that learn your style from past documents and apply it automatically.
  • Clearer labeling and disclosure norms in schools and publishing.
  • Better citation checking built into writing tools.
  • Wider enterprise controls over what staff may paste into assistants.

Key takeaways

  • Write the brief and the outline yourself so the argument stays yours.
  • Use the assistant for options, critique and line edits, not for facts.
  • Edit in passes, and read aloud.
  • Verify every fact, quote and citation at the source.
  • Follow disclosure rules and protect confidential material.

Frequently asked questions

Will an assistant make my writing sound generic?

It can if you accept its first draft. Giving it a sample of your own writing and editing in passes keeps your voice.

Can I trust the sources an assistant gives me?

Not without checking. Open each link and confirm it supports the claim, and remove anything you cannot verify.

Should I tell readers I used an assistant?

Follow the policy of your school, employer or publisher. If none exists, consider what your readers would reasonably expect.