The 2026 TOEFL iBT four sections: what ETS has and has not disclosed

What does each section of the 2026 TOEFL iBT contain, and how many questions does it have?

The 2026 TOEFL iBT runs in the order Reading, Listening, Writing, Speaking, and the tasks, question counts, timing and scoring for each section are shown in the table below (Test Overview, page 4).

Sources: question counts and timing come from page 4 of the Test Overview and the ETS "TOEFL iBT Test Content" page (no publication date; checked 2026-10-30). Maximum points per question and module design come from pages 2 to 7, 10, 13, 15 and 18 of the ETS "TOEFL iBT Test: 2026 Update Test Blueprint and Specifications Document" (no publication date; checked 2026-10-30; hereafter the Specifications Document). The scoring column comes from the table on page 2 of the Specifications Document and pages 34, 35 and 39 of the ETS "TOEFL iBT Technical Manual" (ETS Research Report RR-25-12; October 2025; checked 2026-10-30; hereafter the Technical Manual). Page numbers in all three PDFs follow the PDF page order.

Who does the scoring is stated once, in this paragraph. Page 35 of the Technical Manual says Write an Email, Write for an Academic Discussion and the Speaking tasks are scored jointly by the ETS AI scoring engine and human raters. Page 39 of the same document says a human rater assigns the score when the AI scoring has low confidence in a response, and that randomly selected responses are also reviewed again by humans. Page 40 states that each test taker's responses are scored by two or more raters. Build a Sentence is scored against the answer key and uses neither human nor AI scoring (page 39, Table 7 note).

"Machine scored" in the table renders Machine Scored from note 4 on page 3 of the Specifications Document, meaning selected responses are scored by matching the answer key or by predefined scoring logic. Two-stage adaptive (the difficulty of the second module depends on performance in the first module) is the wording of note 5 on page 3 of the Specifications Document and page 28 of the Technical Manual, and page 3 of the Test Overview calls the same design multistage adaptive. The meaning of linear (every test taker who gets the same test form receives the same tasks) is given in note 5 on page 3 of the Specifications Document.

The four section scores and the total score are reported on the 1 to 6 bands in steps of 0.5, and the total is the average of the four section bands rounded to the nearest 0.5 (Test Overview, page 21). The ETS "TOEFL iBT Test Content" page (checked 2026-10-30) says a comparable 0 to 120 total score will also be provided for two years after January 2026. How the bands map to 0 to 120 is covered in the Unlock Prep (unlockprep.com) guide "How do TOEFL 1 to 6 scores compare with 0 to 120 scores?".

The "not disclosed" items below list only what we could not find after reading page 28 of the Test Overview, page 18 of the Specifications Document, the Technical Manual (68 pages, especially section II-7 on test design and section III on scoring), the ets.org Test Content page, the four section pages and the score page, and the "2026-2027 TOEFL iBT Information Bulletin" (hereafter the Bulletin).

What has ETS disclosed, and not disclosed, about Reading and Listening?

For Reading and Listening, ETS has disclosed the length of the text or audio for each task and the number of scored questions per module and path, but not how many questions go with a single Listen to an Announcement or how much time each task gets within a section.

Table 1 on page 29 of the Technical Manual gives the number of scored questions for each module and path. Lower and upper refer to the Lower and Upper module of the second module.

Whichever path a test taker takes, there are 35 scored questions (same table). Page 28 of the Technical Manual says Reading modules can contain 15 unscored trial questions and Listening modules may include 12. Adding 15 and 12 to 35 matches the 50 and 47 questions on page 4 of the Test Overview, but this arithmetic is our own check.

Reading texts stay on screen while you answer the questions (Specifications Document, page 9), and Listening audio plays only once (Specifications Document, page 12). The ETS Listening page (checked 2026-10-30) says you may take notes while listening. Page 12 of the Specifications Document states that the Listening audio uses AI-generated voices and sometimes human voice actors.

Here is what ETS has not disclosed for these two sections.

• A sentence that states directly how many texts or recordings each task has. The scored question counts in Table 1 above and the questions per text on pages 5 to 12 of the Test Overview are available, but no sentence gives the number of texts.
• How many questions go with a single Listen to an Announcement. Page 11 of the Test Overview and page 19 of the Technical Manual only say that you listen to the announcement and answer questions.
• A set time for each task within a section. Page 4 of the Test Overview and the Test Content page give only section times, and page 2 of the Specifications Document gives only module times.

What has ETS disclosed, and not disclosed, about Writing and Speaking?

For Writing and Speaking, ETS has disclosed the time for each task except Build a Sentence, the 0 to 5 point rubrics and who does the scoring, but not a separate time for Build a Sentence, the preparation time for Take an Interview or a length requirement for Write an Email.

The 0 to 5 points in the table are the score for one task (the raw score), a different scale from the 1 to 6 bands. Raw score ranges are 0 to 20 for Writing (0 to 10 for Build a Sentence) and 0 to 55 for Speaking (Technical Manual, page 30, Table 2). Who does the scoring is as described in the Technical Manual paragraph in the first section.

Here is what ETS has not disclosed for these two sections.

• A separate time for Build a Sentence. Pages 4 and 13 of the Test Overview and page 6 of the Specifications Document give only the 23 minutes for all of Writing. The article that covers this task in depth is "TOEFL Build a Sentence: which grammar spots cost points?".
• A length requirement for Write an Email. The sample prompt on page 14 of the Test Overview only says to write as much as you can in complete sentences, and the rubrics on pages 23 to 24 give no word count either.
• Preparation time before answering in Take an Interview. Page 19 of the Test Overview gives only the 45 seconds for answering.
• The weight the AI scoring engine gives each feature. Tables 3 to 6 on pages 37 to 38 of the Technical Manual only give examples of features the engine looks at (speech rate, pauses, pronunciation, words repeated correctly and so on) and do not state their weights.

Where is the section structure easy to misread?

The easy places to misread are where a question count is taken straight as a score, or a section name straight as a scoring method. The question counts in the tables count the questions that appear on screen, not only the questions that count toward the score. The four cases below are examples we wrote for illustration.

Example 1

• Misreading: "Reading has 50 questions, so the raw score can reach 50."
• Correct reading: "The 50 questions include unscored trial questions, and the Reading raw score range differs by document: 0 to 30 in the Test Overview and 0 to 35 in the Specifications Document and the Technical Manual."
• Why: Specifications Document page 3, note 1 and Technical Manual page 28 (unscored questions); Test Overview page 21, Table 1, Specifications Document page 2 and Technical Manual page 41, Table 8 (raw score ranges).

Example 2

• Misreading: "Writing and Speaking are all scored by AI alone."
• Correct reading: "The 10 Build a Sentence questions are scored against the answer key, while Write an Email, Write for an Academic Discussion and the 11 Speaking questions use an AI scoring engine together with human raters."
• Why: the Machine Scored and AI Scored columns in the tables on pages 2, 6 and 7 of the Specifications Document, and pages 35 and 39 of the Technical Manual.

Example 3

• Misreading: "Whether the second module is upper or lower, getting 30 questions right gives the same score."
• Correct reading: "The score reflects both the number correct and the difficulty of the module you received, and is set through equating (a statistical procedure that adjusts for difficulty differences between test forms)."
• Why: page 28 of the Technical Manual says the number correct across the two modules and the difficulty of the routed module are considered together, and page 34 says the same reported score means the same ability whichever second module was taken. The reported score is not the same as the number or percentage correct (same page).

Example 4

• Misreading: "Everyone answers 47 Listening questions in 29 minutes."
• Correct reading: "The Specifications Document gives the Listening second module as 7 minutes for lower and 11 minutes for upper, and the Test Content page says time and questions may vary as the test adapts."
• Why: page 2 of the Specifications Document and the sentence below the timing table on the Test Content page (checked 2026-10-30).

What has ETS left undisclosed across all four sections, and where do the documents give different numbers?

Across all four sections, ETS has disclosed how raw scores are converted to bands but has not published the conversion table built for each test form, and for Reading and Listening it does not state the score that decides the second module or which questions are unscored.

• Conversion tables. The method is disclosed; the tables are not published. Page 34 of the Technical Manual says Reading and Listening raw scores are converted to scale scores through IRT (item response theory) equating, with a conversion table built for each test form, and page 35 says the sum of Writing and Speaking raw scores is converted through weighted equipercentile linking. Table 1 on page 21 of the Test Overview and Table 8 on page 41 of the Technical Manual give only raw score ranges and band ranges.
• The score that decides the second module. Page 28 of the Technical Manual says that doing well on the first module leads to a harder second module, but does not say how many correct answers send you to the upper module.
• Where the unscored questions are. Page 28 of the Technical Manual gives the numbers (Reading 15, Listening 12) but not which questions they are.

There are five places where the ETS documents give different numbers or ranges.

The Test Overview and the Specifications Document have no publication date, and page 3 of the Specifications Document itself says minor revisions may be made before the full launch. So when the numbers differ, we do not pick one value as correct; we list each document's value side by side. Table 1 on page 29 of the Technical Manual gives 35 scored questions per path, which explains the 0 to 35 values, but we did not find a sentence explaining where the Test Overview's 0 to 30 comes from. Where the timing differs and what the test day schedule looks like are covered separately in "How long is the 2026 TOEFL iBT?".

Where should you check again before the test?

Before the test, reread three places in the Test Overview, the scoring sections of the Technical Manual and one ets.org page, and check the rules in the Bulletin.

• The table on page 4 of the Test Overview (tasks, question counts, timing), Table 1 on page 21 (raw score ranges) and the rubrics on pages 23 to 28 (0 to 5 points for each Writing and Speaking task)
• Pages 28 to 29 of the Technical Manual (adaptive design, scored question counts) and pages 34 to 35 and 39 to 40 (how scores are converted, who does the scoring)
• The timing table on the ETS "TOEFL iBT Test Content" page and the sentence below it (instruction time is separate; time and questions may vary because the test is adaptive)
• Rules such as score reviews and cancellations are in the Bulletin (in effect July 2026 to June 2027 per page 1; checked 2026-10-30)

The "not disclosed" items in this article are based on the documents we read on 2026-10-30. When ETS releases new documents, we will compare those items again first.