A practical, level-by-level rubric for the three Writing tasks, so you can score your own emails, discussion posts and sentence arrangements the same way every time — and see exactly what would lift a response to the next level.
Test Ninjas Research · Updated October 1, 2026
These are Test Ninjas’ original practice rubrics, not official ETS scoring. They are written to match the published TOEFL task descriptions, and they use a 0–5 practice band so you can judge one response at a time. The real TOEFL reports each section on a 1–6 band scale (see TOEFL score range), and there is no conversion from a practice band to an official score.
The Writing section mixes a quick accuracy task with two extended responses, and they need different kinds of judgment. Build a Sentence has a right answer. The email and the discussion post do not: many different responses can succeed, so the rubric describes what a successful response does rather than what it says.
How each Writing task is judged in practice
Task
Practice scale
What is considered
Build a Sentence
0 or 1 per item
Complete accepted arrangement
Write an Email
0–5 holistic band
Purpose and development, Reader relationship and register, Organization, Language control
Academic Discussion
0–5 holistic band
Relevant contribution, Development, Language control
For the 0–5 tasks, the band is holistic: read the whole response, then choose the level whose description fits it best. The dimensions are a checklist for explaining that choice, not a formula that adds up to it.
Write an Email rubric
The email task gives you a recipient, a situation and three things to do. A strong email lets the reader understand the problem and act on it without guessing, in a tone that suits the relationship. Read the task guide at Write an Email, then compare your work with the email sample answers at every band.
What the rubric looks at
Purpose and development
Reader relationship and register
Organization
Language control
These are things to notice, not separate sub-scores. The band you choose describes the whole response; the dimensions help you explain why.
In our worked example at this level: The message explains the two decisions, proposes a workable distinction on the form, and makes a specific retrospective-review request. The language and respectful tone help the recipient understand both the problem and the next steps.
Band 4
Effective message with adequate detail and mostly controlled language.
In our worked example at this level: All requested actions are covered with adequate explanation and a suitable tone. The message is clear, although phrasing such as 'the right name' is less precise and 'your helps' is a noticeable word-form slip.
Band 3
Purpose partly achieved; gaps or language limit effectiveness.
In our worked example at this level: The three requested actions are recognizable, and the central distinction is understandable. Repeated agreement, verb-form, article, and complement errors reduce precision and ease of reading. The proposal and retrospective check are only briefly developed.
Band 2
Limited message; weak development or errors impede communication.
In our worked example at this level: The response attempts the relevant consent problem, but readers must reconstruct several relationships. The proposed choices and request about earlier interviews are unclear. Persistent sentence-level problems prevent the message from functioning effectively.
Band 1
Scant connected original language; purpose barely communicated.
In our worked example at this level: A few related fragments suggest an attempted message, but they do not explain the distinction or communicate a usable proposal and review request.
Band 0
Blank, wholly unrelated, copied-only, or no English response.
In our worked example at this level: The learner intentionally submitted no written response through a functioning attempt.
How to score a response with it
Read the recipient, relationship, situation, and three requested actions together.
For each requested action, record addressed, partly addressed, or absent, with response evidence. This checklist is not an automatic three-point grade.
Ask whether the recipient can understand the problem and act on the message without inventing missing information.
Judge tone against the relationship. A friendly note and a formal request need not use the same greeting.
Distinguish small editing slips from language that changes the intended request or makes it hard to follow.
Select the holistic band that best explains the whole response; do not impose a fixed deduction for one missed bullet.
Questions to ask before you choose a band
Which requested action needs a concrete detail?
Would this wording sound suitable to this recipient?
Which sentence could make the next step easier to understand?
What this rubric deliberately does not do
Avoid mandatory greeting or signature formulas.
Avoid unpublished minimum word counts.
Avoid requiring the model's exact proposed solution.
Avoid deducting one band for a single typo.
Academic Discussion rubric
The discussion task shows a professor’s question and two classmates’ posts. You are judged on whether your contribution is relevant, whether you develop it with a reason or example, and how well your language carries the idea. Any position can score at the top if it is well supported. See the Academic Discussion guide and the discussion sample answers.
What the rubric looks at
Relevant contribution
Development
Language control
These are things to notice, not separate sub-scores. The band you choose describes the whole response; the dimensions help you explain why.
The levels, one by one
Band 5
Relevant ideas thoroughly developed in precise, flexible language.
In our worked example at this level: The response gives a clear position, develops it through a specific bridge example, and explains how that example addresses the classmates' underlying concerns. The contribution is coherent and uses precise, varied language.
Band 4
Clear contribution with adequate support and controlled language.
In our worked example at this level: The position is relevant and adequately supported by a clear example. Language is controlled and easy to follow. The explanation is effective but less fully developed and precise than the stronger anchor.
Band 3
Understandable contribution; support or language has noticeable gaps.
In our worked example at this level: A relevant view and example are present, but the explanation is brief and relies on broad claims. Repeated agreement and word-form errors make the contribution less controlled while leaving most of the meaning understandable.
Band 2
Weakly supported ideas; language or relevance often limits understanding.
In our worked example at this level: The writer attempts a relevant position and a bridge example, but the relationships between ideas are poorly expressed. The support remains thin, and repeated structural problems make parts difficult to follow.
Band 1
Fragments convey little connected, original contribution.
In our worked example at this level: The response contains topic-related words and a rudimentary view but almost no connected development. It provides too little controlled language to communicate a meaningful contribution.
Band 0
Blank, wholly unrelated, copied-only, or no English response.
In our worked example at this level: A functioning attempt was deliberately submitted without a contribution.
How to score a response with it
Identify the professor's actual issue before judging the learner's position.
Locate the learner's contribution and the explanation, example, or detail that develops it.
Check the logical link between support and position; a list of examples is not automatically an explanation.
Accept a relevant position on either side, a qualified view, or another defensible approach.
Judge language by whether it expresses the intended relationships clearly; assess errors in context.
Use the 100-word recommendation as writing guidance, not an automatic grade boundary.
Questions to ask before you choose a band
What new reasoning does the response add?
What does the example show about the professor's question?
Where is a logical connection left for the reader to guess?
What this rubric deliberately does not do
Avoid requiring agreement with both students.
Avoid requiring classmates to be named.
Avoid requiring a compromise position.
Avoid grading personal opinions for correctness.
Avoid counting copied discussion language as original development.
Build a Sentence: right or wrong
In Build a Sentence you arrange word tiles around fixed text to make one grammatical sentence. Our practice checker gives each item one point or none.
Build a Sentence practice scoring
Score
Meaning
1
A complete arrangement matches a declared accepted order, including identical-tile equivalence.
0
A valid submitted arrangement is incomplete or does not match an accepted order.
Why no partial credit: Original local practice checker policy; the selected PDFs do not establish operational partial-credit rules. Counting correctly placed tiles would also reward sentences that are still ungrammatical as a whole.
How an arrangement is checked
Validate the submitted slot and tile identities before grading. Unknown IDs, repeated IDs, and a mismatched item revision are technical errors, not language scores.
Keep the supplied fixed words and punctuation in place, including interior fixed text.
Accept every declared order and explicitly equivalent identical tile. Do not require unused distractor tiles to be placed.
Give grammar feedback using this item's reviewed explanation; do not award invented partial credit by counting correct positions.
Route a plausible unlisted arrangement for editorial review rather than claiming the accepted-order list proves all other sentences are ungrammatical.
Questions that fix a wrong arrangement
Which supplied fixed phrase constrains this slot?
Which tile sequence expresses the intended response?
How does the reviewed grammar explanation justify the arrangement?
See the rubric applied
Six email answers from band 5 down to 0, each with the reasoning behind its level.
Self-scoring works best when you separate reading from judging. Write the response under time, put it away for a few minutes, then follow these steps.
Reread the prompt first, not your answer. List what the reader asked for.
For the email, mark each requested action as addressed, partly addressed or absent, and copy the sentence that proves it.
For the discussion, underline your position and the one reason or example that does the most work. Ask whether it actually supports the position.
Read the response once as the intended reader would. Note any place where you had to guess what was meant.
Choose the band whose description fits best. If you are torn between two, write down why — that note is your next practice target.
Compare with the worked examples at the same band on the sample answer pages.
When you write feedback for yourself or a study partner, keep it in this shape:
State the task outcome.
Cite a specific response excerpt or audible span.
Identify one useful improvement.
Show a possible revision without prescribing the model answer.
Scoring fairly
Do not grade the learner against the model's identity, personal experience, opinion, or exact choice of examples.
Do not count sophisticated words, clauses, or response length as a substitute for successful communication.
Record uncertainty and request review when the evidence supports more than one band.
Frequently Asked Questions
No. These are Test Ninjas’ original practice rubrics, written to match the published descriptions of the 2026 Writing tasks. ETS scores the real test, and it reports Writing on a 1–6 band scale. Use these rubrics to judge practice responses, not to predict an official score.
The 0–5 scale judges a single practice response, including a confirmed empty one (0). The official 1–6 band is a section-level result built from every Writing item. The two scales measure different things, so there is no conversion between them.
Not automatically. The rubric asks you to record each requested action as addressed, partly addressed or absent, then choose the band that best describes the whole message. A missing action usually matters a lot, because the reader cannot act on it, but there is no fixed deduction.
The practice rubric does not use a word count to set the band. Our discussion practice prompts ask you to aim for at least 100 words, and developing an idea usually takes at least that, but a long response with thin reasoning does not earn more than a shorter, well-developed one.
Each item is right or wrong: 1 if your complete arrangement matches an accepted order, 0 otherwise. We do not award partial credit for correct positions, because that rule is a local practice convention, not a claim about how ETS scores the task.