A 19-year-old university student in Taichung reads a polished internship email, understands every line and closes the tab. Ten minutes later, the blank screen returns. A model answer becomes more useful after the learner has already tried to express the same message. The first draft creates specific gaps. Comparison can then attach language to those gaps, and revision tests whether the learner can use it.
This article is for a B1 writer who can recognise good English and struggles to produce a 320-word application, report or GEPT response.
Recognition gives weak evidence
Models remove many writing decisions. The opening is chosen, the verbs agree, and each paragraph already has a job. A learner can follow the result without rehearsing any of the choices that produced it. Writing first changes the reading. Someone who spent 43 seconds searching for a way to say 「我能配合時間」will notice I can adjust my schedule in a model. Another learner may pass over that phrase and stop at the paragraph opening they could not organise. The same text supplies different feedback because each draft has exposed different problems.
This is the productive value of a failed attempt. The learner has a question before the answer appears.
What 37 writers noticed
The sequence fits Choc's argument that peer feedback needs a second draft because feedback earns class time when the learner makes another language decision with it, and revision tests how the learner uses a model as feedback. Osamu Hanaoka studied 37 Japanese university learners in a four-stage picture-story task. Participants wrote a story, compared their draft with two native-speaker models, revised immediately, and later revised again. They noticed lexical features far more often than other features. Items connected to problems they had noticed while writing entered revision at a higher rate and lasted longer than unrelated features. The 2007 Language Teaching Research paper supports a precise claim: producing a draft can direct attention during later comparison.
It leaves several questions open. The picture-story task differed from a workplace email or argumentative essay, and the 37 learners came from only two university classes. The study examined what participants noticed and reused within this task; it cannot promise broad improvement in writing proficiency. Qi and Lapkin followed two Mandarin-speaking adult English learners through a related process. Each person composed a text, received a reformulated version that preserved the original ideas, compared the two while thinking aloud, and wrote again. Better-quality noticing was related to better changes in the final text, while the two learners differed sharply in how deeply they explained the changes. The Journal of Second Language Writing case study is unusually detailed and unusually small.
Two cases reveal a process.
They cannot estimate an average effect for Taiwanese classes.
A model cannot diagnose everything
Comparison can fail in two directions. A learner may copy a polished phrase without understanding its grammar. They may also spend eight minutes collecting attractive vocabulary while missing the model's paragraph structure. Direct correction sometimes works better for accuracy. Rebecca Sachs and Charlene Polio ran one repeated-measures study with 15 adult English learners and a second study with 54 participants. Learners compared their writing with either error corrections or reformulations, then revised. In both experiments, the error-correction condition produced more accurate revisions than the reformulation condition. Think-aloud instructions also changed performance in the first experiment. The 2007 study warns teachers against treating model comparison as a superior replacement for marked corrections.
Corrections and models answer different questions. A correction points efficiently to a wrong verb ending or article, while a model can show how another writer stages a request, links evidence or limits a claim across several sentences. If today's target is third-person -s, mark the form. If the learner's email delivers six facts with no readable order, a model offers structure worth comparing.
I would rather ask a learner to explain three chosen differences than underline 27 differences in silence. Depth matters more than a crowded page here.
Give comparison one target
Use a model written for the same purpose, audience and approximate level. Comparing a B1 internship email with a lawyer's formal letter introduces prestige language that the learner has little reason to retrieve. A useful model solves the same communicative problem in a reachable way.
A 24-minute cycle can work with 180 to 260 words:
- Draft for 9 minutes without a dictionary. Put a small circle around any phrase that caused a pause longer than 10 seconds.
- Read the model once for meaning. On the second pass, find language that answers two circled gaps and one organisational problem.
- Close the model. Revise the original for 8 minutes, then write a margin note explaining each borrowed choice.
Explanations can be short: “available on Tuesday needs the preposition on,” or “the reason comes before the requested action.” If the note merely says “sounds better,” the teacher has found a place for a 52-second discussion. Keep the model out of view during the final revision. Copying while looking tests transcription. Closing it asks the learner to retrieve and adapt. A delayed prompt 3 days later can change the details while preserving the genre: replace an internship schedule email with a request to move a group presentation. Reusing the exact wording would reward memory for one message.
Compare choices, not decorations
For an internship email, a teacher might ask learners to track four features: the first sentence's purpose, the order of reason and request, the verb phrase used for availability, and the closing action. Four columns are enough. Sentence-by-sentence translation will bury those choices. Suppose the draft says, “Because Tuesday I have class, so I hope interview Friday.” The model says, “I have a required class on Tuesday. Would Friday afternoon be possible?” The comparison contains several lessons. The doubled causal marking can be repaired, on Tuesday supplies a preposition pattern, and the request becomes its own sentence. The learner should choose the feature connected to today's target, then produce a new version from memory.
Some learners need language before drafting because an A1 student with almost no relevant vocabulary may stare at the page, rehearse frustration and produce too little for useful comparison. Give that writer six usable phrases and a clear message plan, then ask for a short first attempt. Advanced writers may benefit from two contrasting models because genre allows more than one acceptable solution. The model also requires quality control. An anonymous “perfect answer” from a forum may contain awkward language, copied material or a register that clashes with the task. Teachers should know who wrote the model, why it suits the genre and which features they want learners to inspect.
Tomorrow, wait for a 9-minute draft, circle two stuck phrases, compare once, close the model, and begin the 8-minute revision from memory.