Say someone hands you a draft that recommends a new supplier. The sentences are clear. The headings look sensible. The conclusion sounds sure of itself. Then someone asks why this supplier is the best choice.
You can repeat the conclusion. But you can’t explain why the evidence supports it. The document is ready to read. Your own account of its reasoning is still missing.
You don’t have to turn down writing help to close that gap. You need to decide what the document claims, look hard at its reasons, and change what you can’t back up. You can do that work while you revise a draft someone else wrote.
This module is for writing you need to understand, learn from, or defend yourself.
What you’ll get from this module
In this module
By the end of this module, you’ll have a clearer way to split up writing work while keeping the reasoning yours. You’ll look at a theory of how writing works, research on writing and learning, and experiments on writing stories with AI help. Then you’ll connect what they found to everyday writing and choose one document for a practical exercise. You’ll finish with a rough explanation of its argument and a record of what needs changing or is still unanswered.
Ask Jing
Become a Jingius.
What happens between an idea and a finished document?
Writing-process theory: revision can reopen the argument.
Linda Flower and John R. Hayes put forward a theory of writing in 1981. It describes how planning, putting ideas into words and reviewing can overlap. When you judge a passage and revise it, you may go back to planning what you want to say.
That is a different picture from finishing your thinking, typing it out and fixing the spelling. In their account, rethinking a passage can be part of working out the thought itself.
So what you change during revision matters. Fixing an awkward phrase is one kind of work. Rethinking the reason for a conclusion is another. Both can happen after a first draft exists.
Why it matters: This theory helps explain why the stages of writing can overlap. It doesn’t measure what you lose when you hand one stage to someone else.
Read the theory of writing as a thinking process.
Educational meta-analysis: writing can help people learn content.
Steve Graham, Sharlene A. Kiuhara and Meade MacKay published a review in 2020. It pulled together 56 experiments on writing in science, social studies and mathematics lessons. The students were in Grades 1–12.
The review found a modest learning benefit from writing about the material. In these school settings, writing helped students learn the subject. It was more than a way to present an answer.
Why it matters: For an adult preparing a recommendation, this raises a useful question. Are you just producing words, or working through what they mean? The review supports writing’s role in learning at school. It doesn’t estimate the benefit of writing a document at work.
Read the review of writing and subject learning.
University quasi-experiment: learners can do important work on a supplied draft.
Chun Lai and colleagues published a study in 2026. It compared three ways of splitting up writing work among 101 university students learning English as a foreign language. In the first, AI wrote the draft, and the learners judged it and revised it.
In the second, the learners wrote the draft and revised it, and AI judged it. In the third, the learners did all three jobs.
The reported results for writing development favoured the first way, where AI supplied the draft and the learners judged and revised it. In all three, revising was the learner’s job.
What mattered was what the learners did with the text, not just who wrote its first sentences. A draft someone else supplies can still leave you demanding work: judging the argument and deciding what needs to change.
Why it matters: This was a quasi-experiment. The researchers compared existing classes instead of putting each student in a group at random. Other differences between the groups could have shaped the result. Its published abstract, the short summary, gives no number for the size of the benefit. The study supports treating active revision as real work. It doesn’t promise how much understanding your own way of working will keep.
Read the study abstract on dividing drafting and feedback tasks.
Randomized storytelling experiment: better-rated writing can still become more similar.
Anil R. Doshi and Oliver P. Hauser ran an online experiment in 2024. Writers were given no AI starting idea, one idea, or up to five ideas. Each wrote a story eight sentences long.
The analysis covered 293 writers and their stories, out of 500 people who started. A separate group of 600 people rated the stories.
Writers who could see up to five ideas got higher ratings for novelty and usefulness than writers with no AI help. At the same time, the stories written with AI help were more alike from one writer to the next.
Both findings can be true at once. One story can be rated higher while the whole set covers a narrower range of ideas. A useful suggestion and a varied set of starting points are different benefits.
Why it matters: The experiment measured the stories and their ratings, not what the writers understood. For this module, it shows why taking a useful idea and checking the direction it gives your document are separate decisions.
Read the experiment on story quality and collective diversity.
Modified storytelling experiment: the starting inputs matter.
Yun Wan and Yoram M. Kalman’s 2026 paper, in the authors’ accepted version, offers an important comparison. Some writers got ideas from a pool made by AI playing different assigned characters, called personas. Others wrote without that help.
The study recruited 100 writers and analysed 97 stories. Its main comparison did not find that the stories written with help were significantly more alike.
One analysis compared stories in pairs. It found more variety when writers had one supplied plot. A more cautious check gave each story its own similarity score. It did not find a significant difference.
The study was small for what it tested, so it could miss a real difference between the groups.
These mixed results don’t show that the variety was equal. They don’t promise that varied prompts will stop stories becoming more alike. They do challenge the idea that every way of supplying AI ideas gives the same result.
Why it matters: Taken together, the two storytelling studies say to watch the inputs and choices that shape the work. Whether help is there at all is only part of the picture.
Read the accepted manuscript on diverse starting ideas.
What this means for you
Keep the decisions that make the argument something you can explain: its main claim, its reasons, its evidence, the objections and the limits. Help can go into the document while you check and revise those decisions.
Decide what kind of writing you’re doing
A routine message about something you already understand doesn’t need to become a learning exercise. A document that explores an unfamiliar question, or recommends a choice with real consequences, deserves attention to its reasoning.
Ask what you will need to explain once the document leaves your hands. If someone challenges its conclusion, what reasons would you give? That shows you the part of the work you need to do yourself.
When the argument is new to you, start with a rough account before you accept a structure someone hands you. Write down or say out loud the claim, why you believe it, a strong objection, and what you still don’t know.
If you already have a draft written with help, you can rebuild its reasoning and revise it instead. The point is to make those decisions visible, not to retype every sentence yourself.
Treat suggestions as options you still have to choose
Treat ideas and outlines as material to look at closely. Ask what the suggested approach puts first, what it leaves out, and whether another starting point would suit your situation better.
In the made-up supplier example, a draft might build everything around price. That structure is useful if price answers the real question. It falls short if the recommendation also depends on needs the draft hasn’t considered.
Add other options that suit your situation, then choose the order of the reasons yourself. Keep the links between reasons and conclusions easy to see. An outline that looks familiar isn’t a reason to accept what it assumes.
The storytelling research makes this choice easier to see: helpful input and variety are separate concerns. You can welcome a suggestion and still ask what other direction is worth a look.
Revise the reasoning, not only the wording
Real revision changes what the document argues, or how it backs that argument up. Look for claims that go further than their evidence, sections that contradict each other, and missing reasons for the conclusion.
Say the supplier draft recommends the best supplier, but the only comparison behind it is the lowest price. Lowest price and best overall are different claims.
You could narrow the recommendation to the price comparison the evidence supports. Or you could get the missing information the broader choice needs. Until then, leave the broader conclusion open.
Changing best to excellent would only change the wording. The same unsupported leap would still be there. That is why the language-learning study matters here: judging and revising a draft can mean more than deciding that it reads well.
Check meaning after polish and formatting
You can ask for help with grammar, cutting words, turning speech into text, or layout. Then compare the result with the claims and source details you meant to keep.
Say a made-up sentence states that a change may reduce delays. A polished version says it will reduce delays. The second version sounds more certain, so it needs different support.
Put the supported wording back unless the evidence justifies the stronger claim. Check numbers, labels and citations after formatting too. Those details show which evidence supports which statement.
Judge the change itself. Calling a task polish doesn’t make every change it brings a surface one.
Explain the argument before you approve it
Close the polished text. Then explain its main claim, its reasons, its strongest objection and what is still uncertain, in your own words. You can write or say this rough thinking draft.
Open the document again and compare the two. If you can’t account for a reason or a qualification, go back to the evidence behind it and revise the passage. If information is missing, write down the question and leave the claim it affects open.
You have a good place to stop when the document matches the reasoning you can explain, and what is still uncertain is in plain view. The goal is a document you can take responsibility for. It is not a rule that you type every first sentence yourself.
How about your daily life?
Where do you use writing to understand a subject or explain a choice? You might be preparing a recommendation at work, working on an essay, or writing an explanation for other people.
List situations where you’d like a clearer account of the reasoning. One good idea is enough.
| A writing situation | What I need to understand or explain |
|---|---|
| ____________________ | ____________________ |
| ____________________ | ____________________ |
My chosen situation: ____________________
Choose one document whose reasoning you need to understand, learn from, or defend. Take it into the exercise below. You can change your choice if a different document would be more useful.
Check the reasoning before you approve
Start with the document you chose. If you came straight here, pick writing you need to understand, learn from, or defend yourself. A routine message about something you already understand doesn’t need to become this exercise.
- Make a rough thinking draft. Write down or say out loud the main claim, your reasons, a strong objection and what you still don’t know. For an argument that is new to you, do this before you accept a structure someone hands you. If you have a draft written with help, rebuild its reasoning and revise it yourself.
Claim, reasons, objection and uncertainty: ____________________ - Examine the argument and its evidence. Check whether each important reason supports the conclusion, whether the order makes sense, and whether anything contradicts it. Think about sensible alternatives that the starting ideas or the outline leave out.
Narrow any claim that goes further than its evidence, or name the information you would need to support it. Leave the conclusion it affects open while that information is missing.
Change needed or unanswered question: ____________________ - Compare the polished version with what you meant. After help with wording or formatting, check how certain it sounds, any claims about cause and effect, and the numbers, labels and citations. Correct any change that makes the argument say more than its support allows.
Meaning or detail to restore: ____________________ - Close, explain, then compare. Close the polished text and explain its main reasoning and its limits in your own words. Open it again and fix anything you can’t account for. Keep open questions in view instead of approving claims you can’t support.
What I can explain, and what still needs work: ____________________
Ask Jing
Become a Jingius.
Your next step
Finish with your rough explanation and a document that matches it, or a clear record of the gap that stops you approving it. You can keep the help that is useful and still hold on to the decisions that make the argument yours.
A note on the evidence
The gains from writing at school and the story ratings answer different questions. They don’t give a measured percentage of how much adults understand under each way of splitting up writing work. The study that used varied AI starting ideas also warns against assuming that stories written with help will always become more alike.
Use the findings to tell apart the kinds of work involved, while you judge the reasoning and evidence in your own document.
How we prepared this module
This module connects a theory of how writing works, research on writing and learning, and experiments on writing stories with AI help to practical decisions about drafting and revision. Made-up examples about a supplier and a change of wording show how to check a claim and a change in meaning.
Sources
- Linda Flower and John R. Hayes. (1 December 1981). A Cognitive Process Theory of Writing. College Composition and Communication, 32(4), 365–387. Four propositions, p. 366; “Reviewing,” p. 374.
- Steve Graham, Sharlene A. Kiuhara and Meade MacKay. (19 March 2020). The Effects of Writing on Learning in Science, Social Studies, and Mathematics: A Meta-Analysis. Review of Educational Research, 90(2), 179–226. Abstract.
- Chun Lai, Mengru Pan, Kai Guo and Yaqiong Cui. (October 2026). What task to offload to GenAI in the writing feedback process? – Effects of task-offloading approaches on EFL learners’ writing skill development. Computers & Education, 252, 105675. Institutional abstract, design and ANCOVA result sentences.
- Anil R. Doshi and Oliver P. Hauser. (12 July 2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), eadn5290. Introduction; Results, creativity and similarity; Materials and Methods, “Writer study and experimental conditions.”
- Yun Wan and Yoram M. Kalman. (2026). Diverse AI Personas Can Mitigate the Homogenization Effect in Human-AI Collaborative Ideation. Computers in Human Behavior: Artificial Humans, 100289. Accepted author manuscript v2, 16 March 2026; §§3.1–3.3, pp. 10–15.
