A B1 office worker can read If I had known, I would have called and understand the regret.
Producing the same pattern in an email is harder because the writer must choose both verb phrases under pressure, while a model shown first can look familiar without changing the next sentence they compose. A better sequence starts with a brief attempt, then supplies a model worth comparing. The failed or uncertain attempt gives the learner a precise reason to inspect the grammar. This article is for an adult learner who understands explanations in class, scores around 620 on the TOEIC, and still avoids complex conditionals in workplace writing. The claim is narrow: when the target form is partly known, one attempt before the model can make later input receive closer attention.
A gap the learner can feel
Consider this prompt: “Mina missed the client’s message yesterday, so describe the different result if she had seen it.” A learner might write, If she saw it, she replied earlier. The teacher does not correct it yet. The learner keeps the sentence beside a short model containing If she had seen it, she would have replied earlier.
Now the model answers a live problem.
The learner can compare saw with had seen, then replied with would have replied. Reading has acquired a purpose that “study these conditional forms” rarely supplies. The comparison also reveals which part was already secure. Perhaps the learner chose the right conjunction and conveyed the intended time relation, while the two verb phrases remained unstable. This is output used as a diagnostic event. Keep it brief. It can be one sentence. It need not become a graded performance, and the teacher need not collect 14 polished paragraphs.
What two experiments found
He, Godfroid, and Li studied this sequence with 79 L1 Chinese learners of English.
Their target was the English past counterfactual conditional. In the first experiment, 68 participants were randomly assigned to an output group or a comparison group, 34 in each. Both groups read the same 12-sentence model essay twice. Between readings, the output group reconstructed the essay for content and grammar; the comparison group answered comprehension questions. The output group improved more in grammar production. Grammar carried the load. Eye tracking also showed a difference during later processing of the target forms. After reconstruction, those learners returned to relevant parts of the model more often. The study recorded attention while it happened instead of relying only on a questionnaire after the lesson.
Limits remain.
The authors found no statistically significant mediation from the measured attention variables to learning, and they did not administer a delayed post-test. Only 11 of 25 participants in the follow-up stimulated-recall experiment remained after data-quality screening. The paper supports an immediate output-then-input sequence for this structure. It cannot promise durable gains for every grammar point. An earlier controlled study by Shinichi Izumi reached a compatible classroom conclusion. Adult ESL learners completed reconstruction and reading tasks involving English relative clauses. Learners in the output-plus-input condition made greater learning gains than learners who read for comprehension, while visual enhancement of the input alone did not produce measurable gains. Bold type can catch the eye. Try first. Producing a sentence asks the learner to resolve a form-meaning problem.
Build a six-minute comparison
Choose one grammatical decision that learners can almost make.
Past counterfactuals suit a B1 or B2 group that already knows ordinary past forms. Article choice may suit another class. A beginner who lacks the words needed for the prompt will spend the attempt searching for vocabulary, so the intended grammar gap may never become clear. Start with a compact situation and request one or two sentences. Give 90 seconds. Ask learners to underline the part where they felt least certain, without showing a rule or answer. That uncertainty mark matters more than a red cross because it preserves the learner’s own prediction.
Then show a model of 47 to 73 words that repeats the target form enough to permit comparison while still sounding like a plausible email, message, or account of an event a colleague might actually describe. Six nearly identical textbook sentences encourage surface matching. One coherent model lets tense, meaning, and discourse remain connected.
Comparison needs a task.
Learners can circle one difference, explain what changed in time or meaning, and revise their first sentence. For If I had seen the message, the question might be: “Which action happened first, missing the message or the imagined reply?” Grammar terminology can follow after that answer. I prefer one clean revision over a worksheet containing 18 isolated blanks. This sequence sits beside Choc Education’s earlier account of why a model answer works best after a first draft. The present version narrows the idea to a target grammar form and a much shorter cycle. A full essay is unnecessary when the decision fits inside one clause.
Keep the first attempt safe
An attempt loses its value when learners expect a score.
They may copy a familiar sentence, wait for a stronger partner, or leave the line blank. Label it “prediction” or “first try,” keep it ungraded, and allow a question mark beside any uncertain verb. The purpose is to expose a gap to the learner. Correction timing also needs restraint. If the teacher immediately supplies the full sentence aloud, the model has little work left to do. Collect two anonymous versions on the shared screen, then let everyone compare them with the model. After the revision, the teacher can explain the remaining pattern and check whether the intended meaning survived.
Some forms are poor candidates.
An A1 learner meeting the third conditional for the first time lacks enough prior knowledge to produce a useful hypothesis. A rare exception with no transparent meaning may also call for a direct explanation first. Output helps most when learners possess pieces of the system and need to discover where those pieces fail. Tomorrow’s smallest version needs one prompt, one sentence box, and one 58-word model. Leave the learner’s original sentence visible until the second version is finished.