Two Speaking tasks, two very different jobs: reproduce a sentence exactly, then answer open questions with developed ideas. This practice rubric explains every level of each, and how to judge your own recordings without fooling yourself.
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.
Relevant development, Intelligibility and pacing, Language control
Both are holistic: listen to the whole response, then pick the level that describes it best. Neither task rewards a particular accent, and neither can be judged fairly from a transcript alone.
Listen and Repeat rubric
In Listen and Repeat you hear a sentence once and say it back. The rubric asks two questions: did you reproduce the words and their meaning, and could a listener understand you? A small slip that leaves the meaning intact costs far less than a change that reverses it. For worked examples and practice sentences, see Listen and Repeat scoring.
What the rubric looks at
Reproduction and meaning
Intelligibility
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 response reproduces the complete target in its original order.
Band 4
Small departures; original meaning remains understandable.
In our worked example at this level: The initial article is absent, but the camera's orientation and the identified surface are preserved. The small function-word omission does not materially change this message.
Band 3
Mostly complete sentence; meaning or intelligibility is affected.
In our worked example at this level: Most target words remain, but adding 'away from' reverses the direction. The speaker produces a complete sentence without accurately preserving the original meaning.
Band 2
Substantial missing or inaccurate material leaves a fragmentary message.
In our worked example at this level: Some target content survives, but the orientation and connecting sentence structure are missing. The result is a fragmentary message rather than a full repetition.
Band 1
Very little target reproduced or largely unintelligible attempt.
In our worked example at this level: Very little of the prompt is reproduced, leaving almost the entire message absent.
Band 0
No response, no intelligible English, or wholly unrelated content.
In our worked example at this level: No target material is reproduced.
How to score a response with it
Listen to the prompt and response; verify that the saved target belongs to this item.
Mark audible omissions, additions, substitutions, changed endings, and word-order differences.
Explain the effect of a difference in this sentence. Losing a negation can matter more than losing a repeated function word.
Judge how much of the message an unfamiliar listener can understand, not how much an evaluator can reconstruct from knowing the target.
Treat self-correction in its actual context, including whether the speaker finishes intelligibly.
Use recognition output as an aid only; neither exact text matching nor word-error rate supplies the holistic band.
Questions to ask before you choose a band
Which exact word or ending was not audible?
Did that change the instruction or meaning?
Which short phrase would be useful to repeat again?
What this rubric deliberately does not do
Avoid equating a familiar accent with correctness.
Avoid automatically scoring a paraphrase as exact repetition.
Avoid automatic one-band deductions per word.
Avoid grading a silent or damaged upload before checking the recording state.
Take an Interview rubric
In Take an Interview you answer four recorded questions about experiences and opinions. Each answer is judged on whether it responds to the question, how well it develops the idea, and how clearly and accurately it is spoken. See six full interviews with model answers on the interview sample answers page.
What the rubric looks at
Relevant development
Intelligibility and pacing
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
Fully developed answer with clear, fluent, precise speech.
In our worked example at this level: The answer describes a concrete situation and explains the precise information gap. The proposed label follows from that explanation. Varied, accurate language and the assumed clear delivery support the strongest practice band.
Band 4
Developed answer; delivery and language generally communicate effectively.
In our worked example at this level: Both parts of the question are addressed with adequate detail. The narrative and reason are clear, while the assumed delivery and less developed explanation fit the generally effective range.
Band 3
Relevant answer with limited support, clarity, or language control.
In our worked example at this level: The situation and cause of confusion are present, but development is limited. Repeated language problems and the assumed choppy delivery reduce clarity without destroying the main message.
Band 2
Thin relevant attempt; meaning often difficult to follow.
In our worked example at this level: There is a relevant attempt, but little developed explanation. Limited language and the assumed delivery make the intended message difficult to follow.
Band 1
Vaguely related fragments or mostly unintelligible speech.
In our worked example at this level: The attempt has a vague connection to the task but does not describe the situation or explain it in connected speech.
Band 0
No response, no intelligible English, or wholly unrelated content.
In our worked example at this level: There is no spoken attempt to evaluate.
How to score a response with it
Play the exact interviewer question and identify every requested part.
Follow the learner's own scenario or position; a previous answer may provide context but must not replace the current answer.
Find support that explains the answer instead of merely restating the question.
Listen for whether pauses, pronunciation, rhythm, or stress actually make the message difficult to understand.
Distinguish a natural pause to think from repeated breakdowns that interrupt communication.
Judge the whole recorded response. A polished transcript does not establish fluent delivery, and recognition mistakes do not establish poor grammar.
Questions to ask before you choose a band
Which part of the interviewer question still needs an answer?
What specific reason or example makes the answer convincing?
At what audible moment did the listener lose the thread?
What this rubric deliberately does not do
Avoid requiring the model speaker's life story.
Avoid requiring a particular viewpoint.
Avoid scoring transcript punctuation as spoken grammar.
Avoid inventing pronunciation faults from text.
Avoid treating every hesitation as an error.
About the worked examples quoted above. Spoken scores assume the explicitly described hypothetical delivery. No learner recording exists for these anchors; the text cannot demonstrate pronunciation, fluency, or intelligibility.
Hear the model answers
Six practice interviews with questions, model answer text and audio.
Record every attempt. Memory is not a reliable judge of what you actually said.
Wait a few minutes, then listen once without the target text or your notes in front of you. Write down what you can understand.
For Listen and Repeat, compare that note with the target sentence. Mark omissions, additions, changed endings and changed order, and ask whether each one changes the meaning.
For the interview, check that you answered every part of the question, then find the reason or example that carries the answer.
Choose the band that fits the whole response. Note one specific improvement and record the answer again.
If the recording failed, do not score it. A silent file, a broken upload or a speech-recognition error is a technical problem, not a band 0 — record the item again. A 0 is only for an attempt that genuinely contains no relevant response.
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.
For speech, confirm perceived omissions against the audio before giving correction feedback.
Do not require an imitation of the prompt speaker's accent or voice.
Record uncertainty and request review when the evidence supports more than one band.
Frequently Asked Questions
No. It is Test Ninjas’ original practice rubric, aligned to the published descriptions of the two 2026 Speaking tasks. The official test is scored by ETS and reports Speaking on a 1–6 band scale. A practice band cannot be converted into an official score.
Under this practice rubric, no. What matters is whether an unfamiliar listener can understand you. A different accent from the recording is fine; sounds, stress or pauses that make words hard to recognize are what reduce intelligibility.
It depends on what the mistake does. Dropping a short word like “the” may leave the meaning intact, while dropping “not” or adding “away from” can reverse it. The rubric asks you to judge the effect of each difference, not to subtract a fixed amount per word.
No. Model answers show one possible response. Any relevant, well-developed answer to the question can reach the top band, whatever opinion or personal example it uses.
Only partly. A transcript shows content and grammar, but pronunciation, pace and intelligibility can only be judged from the recording. Record yourself, then listen back before you choose a band.