TOEFL iBT Writing:
methods and practice by task

What has the Writing section included since January 2026?

ETS names three tasks: “Build a Sentence,” “Write an Email,” and “Write for an Academic Discussion.” Its content page lists 12 questions for the section and gives an “Approx. Base Time” of 23, without specifying the unit. The 2026 update specifications break this down into 10 Build a Sentence questions, one Write an Email, and one Write for an Academic Discussion, for an estimated 23 minutes. They add that Reading and Listening are adaptive in two stages, while Writing and Speaking are linear: all test takers taking the same version receive the same tasks.

These are estimates. Beneath its table, ETS says the test adapts and its duration and number of questions may vary; the specifications document warns that it may be revised slightly before the test officially launches. The official Practice Test 1, aligned with the test since January 21, 2026 but adapted for paper, specifies 7 minutes for the email and 10 minutes for the discussion. For Build a Sentence, it refers to the on-screen timer without giving a specific duration. Always follow the instructions and timing shown on screen.

The same document says the section requires written responses using standard input methods, such as a keyboard or an adapted device; the content page specifies that writing assistance software is not allowed during the test. The 2026–2027 Bulletin adds a condition for receiving scores: answer at least one question in Reading and Listening, and at least one task in Speaking and Writing.

Who scores your responses, and how does the section become a score from 1 to 6?

Each Build a Sentence question is scored right or wrong, for 1 point or 0. ETS’s technical manual, dated October 2025, specifies that this task is scored using an answer key, with no human or AI scoring.

The two written responses are scored from 0 to 5 using a rubric specific to each task. The manual says they are evaluated by ETS’s proprietary AI engines and human raters to improve score accuracy and consistency. It also describes how scoring is divided: when automated scoring has low confidence or encounters a difficulty, human raters take over, and a random sample of responses is reread by certified raters. In the same chapter, ETS adds that each test taker’s responses are scored by more than one rater. These passages do not say what proportion of responses a human reads: write for a reader, since the engines have learned to reproduce the human scoring rubrics.

For a sample of 1,914 written responses, ETS reports a correlation of 0.86 between automated scores and the average human scores, and 0.85 between human raters. These correlations describe a statistical association between scores, not a guarantee of agreement for any particular response.

Raw points for the section range from 0 to 20: 10 for Build a Sentence, 5 for the email, and 5 for the discussion. ETS converts them to a score from 1 to 6 using a weighted equipercentile equating method that accounts for difficulty differences between test versions, and says the Writing score is based on overall performance across all tasks. The cited documents do not include a conversion table from raw points to score bands, so do not calculate a band from your practice points.

If you dispute a score, the Bulletin allows you to request and pay for a Writing or Speaking rescore through your ETS account, up to 30 days after the test date. A rescore is no longer available if you have already requested that your scores be sent to an institution. Your responses are then reviewed by scoring specialists; if the score is confirmed, you are notified by email within 3 weeks. If it changes, up or down, the revised score becomes your official score for that test date.

What does a Writing score mean? The April 2026 report describes each CEFR level for writing. We summarize those descriptions, which indicate a level rather than a scoring rubric.

What do ETS rubrics reward in the email and discussion?

Both rubrics range from 0 to 5. At level 5, ETS describes a fully successful response that is clear and consistently fluent. The table presents the features ETS expects in a typical response; the discussion rubric does not list a social-conventions criterion.

At level 3, the response is partially successful: its development only partly supports the purpose (email) or leaves an explanation missing, unclear, or off topic (discussion); the language has an average range or some variety, and errors are noticeable. At level 1, ETS describes serious and frequent errors in both tasks and coherent language drawn mostly from the prompt; for email, it also describes telegraphic language. A blank response, one that is entirely off topic or refuses the topic, is not in English, is copied entirely from the prompt, or consists of random keystrokes receives a 0.

For email, the manual gives examples of features the automated engine measures: sentence count, discourse coherence, similarity to the prompt, sentence variety, word frequency, collocation accuracy, politeness markers such as modal verbs and hedging phrases, and grammatical, word-choice, or form errors (spelling and punctuation). ETS says the model learns to connect these features to human-assigned scores. This is not a checklist to manufacture; these are signs of what the rubric describes: developed, coherent, polite, and accurate writing.

How should you approach Build a Sentence?

ETS says the task measures command of sentence structures, a skill needed for all writing: you see words or groups of words in scrambled order and rearrange them to form a grammatical sentence or question. The specifications document targets all levels from A1 to C2 for this task, so questions range from very easy to difficult. Each question is worth 1 or 0 points, with no partial credit: an almost-correct sentence earns nothing.

The following tips are ours, not ETS rules.

• First, identify the core clause: the subject, auxiliary or modal, and main verb. The rest of the sentence connects to it.
• Check whether the result is a statement or a question; ETS says either is possible.
• After verbs such as “know,” “ask,” or “wonder,” an indirect question follows statement word order: do not invert the subject and auxiliary.
• Keep related words together: a preposition and its complement, a determiner and its noun, a verb and its particle.
• Reread the whole sentence before submitting: check agreement, tense, and the placement of linking words.

How should you approach the email and academic discussion?

For email, ETS says it measures the ability to produce a multi-sentence text that achieves its communicative purpose while following basic social conventions, with enough development, clarity, and coherence, varied and accurate structures and vocabulary, and correct spelling, punctuation, and capitalization. For discussion, it measures the ability to develop an argument for a position by responding to other posts or using their information, support it with one’s own reasons, experiences, or knowledge, and remain clear and coherent. In its April 2026 report, ETS says the Write for an Academic Discussion task has remained unchanged.

Here is our method. In its 2026-aligned Practice Test 1, ETS says an effective discussion contribution contains at least 100 words. For email, it asks test takers to write as much as possible, in complete sentences, without specifying a word count in this example. This length guideline does not replace the development expected by the rubrics.

• Email situation first: who is writing to whom, for what purpose, and what the situation asks you to do (recommending, inviting, and proposing a solution are ETS’s examples).
• Email structure: an opening, the purpose in the first sentence, relevant facts with a concrete detail for each, the request or proposal, and a polite closing.
• Email tone: phrase a request with a modal (“could,” “would”), offer an alternative instead of making demands, and use a register appropriate for the recipient.
• Discussion position: quickly choose the position you can develop, not the one that seems most true to you; scoring is based on the quality of the development, not the opinion.
• Discussion connection: quote or restate an idea from a student or the professor that you agree or disagree with, then give your reason and an example.
• In both tasks, save the final minutes to proofread: common typos are tolerated at level 5, while ETS describes a buildup of errors at levels 2 and 1.