An ESL speaking assessment rubric scores spoken English on a few criteria, each described at set levels. Here is a free CEFR-aligned template, with how to score, calibrate and give feedback.
In short
- Use an analytic rubric for classroom speaking: five criteria (task achievement, fluency and coherence, range, accuracy, pronunciation), plus interaction for pair work.
- The free template below scores each criterion 0–4, with bands loosely aligned to CEFR A2–C1.
- Report the criterion profile as well as the total, and convert totals to grades with cut-scores your school agrees.
- Calibrate with colleagues on three to five shared recordings before each marking round.
- AI practice can give students instant feedback between lessons; the teacher reviews the results and decides grades.
What is an ESL speaking assessment rubric?
An ESL speaking assessment rubric is a scoring guide that names a small set of criteria and describes what spoken English sounds like at each level of each one, so that different teachers score the same performance the same way. It turns an impression ("she speaks well") into evidence you can explain to a student or a colleague.
Most speaking rubrics measure the same few qualities. The CEFR's grid for spoken performance, Table 3 of the 2001 framework, covers range, accuracy, fluency, interaction and coherence, and the CEFR Companion Volume (Council of Europe, 2020) adds phonology. IELTS Speaking scores fluency and coherence, lexical resource, grammatical range and accuracy, and pronunciation, weighted equally.
The template below uses five criteria:
- Task achievement and content: did the speaker do what the task asked?
- Fluency and coherence: can the listener follow a connected turn?
- Range: how much vocabulary and grammar is available?
- Accuracy: how well is that language controlled?
- Pronunciation and intelligibility: how much effort does the listener need?
Task achievement is included because a fluent answer to the wrong question should not score highly. Interaction is an optional sixth row for pair work and role-plays. The Companion Volume notes that assessment grids usually keep to four or five criteria (p. 43), so drop any row a task doesn't need.
Analytic or holistic: which speaking rubric should you use?
An analytic rubric gives a separate score for each criterion, while a holistic rubric gives one overall score for the whole performance. Use analytic when you need feedback, progress tracking or shared marking between teachers; use holistic for quick, low-stakes judgements such as sorting a new class into groups.
| Holistic rubric | Analytic rubric | |
|---|---|---|
| What you record | One band for the whole performance | One band per criterion, plus a total |
| Speed | Fast | Slower at first |
| Suited to | Quick grouping, participation marks | Feedback, progress, report grades, shared marking |
| Drawback | Hides why a student got the score | Criteria blur if descriptors overlap |
Research favours analytic rubrics where consistency matters. A 2007 review of 75 studies by Jonsson and Svingby found that rubrics can make scoring more reliable, especially when they are analytic and topic-specific and come with example performances or rater training. It also cautions that a rubric does not make a judgement valid by itself, and that its value for learning comes mainly from making expectations explicit. IELTS combines both approaches: four criteria scored separately, averaged into one band.
The free ESL speaking rubric template (4 bands × 5 criteria)
This free ESL speaking rubric scores five criteria from 0 to 4, with an optional interaction row for pair and group tasks. Brackets show an approximate CEFR level. The wording is our own, written so that each point can be heard in a classroom recording; it does not reproduce CEFR or IELTS descriptors. Score 0 for no rateable response: no answer, off task, or mostly in another language.
| Criterion | 1 · Emerging (≈A2) | 2 · Developing (≈B1) | 3 · Secure (≈B2) | 4 · Advanced (≈C1) |
|---|---|---|---|---|
| Task achievement and content | Gives basic information on familiar topics. Needs prompts; parts of the task are missing. | Covers the main points with simple reasons or examples. Ideas are relevant but undeveloped. | Covers every part of the task with relevant detail. Supports a view with reasons and examples. | Handles the whole task, including less familiar parts. Develops and qualifies ideas for the listener. |
| Fluency and coherence | Short bursts with frequent stops and restarts. Joins ideas mainly with "and", "but" or "so". | Can be followed, though pauses to plan or self-correct are obvious in longer answers. Simple sequencing ("first", "then"). | Keeps a steady rhythm in longer turns and rarely stalls. Links ideas in varied ways; a long turn may drift a little. | Speaks at length with ease; pauses are for ideas, not words. Signposts structure ("on the other hand"). |
| Range (vocabulary and grammar) | Memorised phrases and short, simple sentences for everyday needs. Often stuck for a word. | Enough words for familiar topics such as school, work and free time; talks around gaps. Common sentence patterns. | Can describe, explain and argue about common subjects and their own field. Mixes simple and complex sentences. | Chooses precise words and a suitable register on academic or work topics, without simplifying the message. |
| Accuracy | Some simple forms are right, but the same basic errors keep returning. Meaning is usually clear from context. | Common patterns mostly accurate in predictable situations. Errors in newer forms rarely block meaning. | Good control. Errors occur but do not mislead the listener. Often repairs own slips. | Slips are occasional and small, and usually caught at once. |
| Pronunciation and intelligibility | Understandable with effort. Strong influence from other languages; listeners may need repeats. | Generally easy to follow. Some words mispronounced, but listeners rarely need to ask again. | Clear throughout. An accent may be noticeable but rarely costs the listener effort. | Easy to follow throughout. Uses intonation and stress to signal emphasis and attitude. |
| Interaction (optional) | Answers simple questions and asks a few. Relies on the partner to keep the exchange going. | Gets a simple chat on a familiar topic going, asks questions to keep it alive and wraps it up. Checks understanding ("So you mean…?"). | Takes turns at natural moments, builds on what others say and asks for their views. | Steers the conversation with ease, ties own points to others' and repairs misunderstandings smoothly. |
- Accent is not an error. Score intelligibility (how much effort the listener needs), not closeness to a "native" accent. The Companion Volume rebuilt its phonology scale around intelligibility (p. 133), and the Council of Europe says the new scale removes native-speakerism.
- The ≈CEFR labels are indicative. They describe this performance on this task, not a certified level. For beginners, move each band down a level and simplify the wording; for advanced classes, move them up.
- For a single-level class, relabel the bands Not yet, Approaching, Meets target and Beyond target, with "Meets target" set to the class goal.
How do you turn speaking rubric scores into a mark or grade?
To turn speaking rubric scores into a grade, score each criterion 0–4, apply any weights, convert the total to a percentage and map it to cut-scores your school agreed in advance. Keep the criterion profile next to the grade, because the profile tells the student what to work on.
- Record a long enough sample. As a practical guide, aim for one to two minutes or more of the learner's own speech, and two different tasks before a report grade.
- Choose the closest fit. Read the whole row; if a performance fully meets band 3 and shows some band-4 features, give 3 (or 3.5 if you allow half points).
- Score each criterion on its own. Weak pronunciation should not pull down task achievement; avoiding that halo effect is the point of an analytic rubric.
- Add up and weight. Unweighted, the total is out of 20 (24 with interaction). Publish any weights with the task.
- Convert with agreed cut-scores. Don't turn an average into a CEFR level; one classroom task can't certify a level.
A worked example for a short presentation, where content counts most:
| Criterion | Score (0–4) | Weight | Points |
|---|---|---|---|
| Task achievement and content | 3 | 30% | 22.5 |
| Fluency and coherence | 2 | 20% | 10 |
| Range | 3 | 20% | 15 |
| Accuracy | 2 | 20% | 10 |
| Pronunciation and intelligibility | 4 | 10% | 10 |
| Total | 14 / 20 | 100% | 67.5 / 100 |
Points are the score divided by 4, multiplied by the weight. Unweighted, the same performance is 14 out of 20 (70%). The report shows 67.5% and the profile 3-2-3-2-4, which points to fluency and accuracy as next steps.
How do teachers calibrate speaking scores?
Calibration is a short session in which teachers score the same recordings independently, compare scores criterion by criterion and agree what each band looks like, so that a 3 means the same in every class. Run one before each marking round and whenever a new teacher joins.
- Choose three to five recordings across the range of your students, including at least one borderline performance.
- Each teacher scores every recording alone, using only the rubric.
- Compare scores criterion by criterion. Discuss any gap of more than one band by pointing to moments in the recording.
- Agree a score for each recording and note why in one line. These become anchor samples: recordings with agreed scores to check against while marking.
- Double-mark borderline and high-stakes cases, and recalibrate each term.
This is the rater training and use of examples that Jonsson and Svingby link to more reliable scoring. For an external reference, Cambridge ESOL's Examples of speaking performance at CEFR levels A2 to C2 (2009) documents recorded tests rated against the CEFR; its terms permit use for examiner and teacher training in non-commercial contexts. Your own agreed recordings are often more useful, because they show your tasks and your students.
How do you give speaking feedback students can act on?
Speaking feedback that students can act on names one specific strength and one specific next step in the rubric's own words, then gives the student a chance to try again soon. Hattie and Timperley's 2007 review of feedback research frames good feedback around three questions: where am I going, how am I going, and where to next?
- Share the rubric before the task. The Companion Volume suggests sharing assessment grids with learners (p. 43).
- Return the profile, not just the total. A 3-2-3-2-4 profile shows where the next point will come from.
- Choose one next step. Pick the criterion closest to the next band, which is not always the lowest.
- Repeat the task. A retake or parallel task within a week lets students use the feedback while it is fresh.
| Vague feedback | Feedback a student can act on |
|---|---|
| Work on your fluency. | You paused mid-sentence to find topic words. Before the retake, list five key words and say your opening two sentences aloud three times. |
| Use more linking words. | Your points came out as a list. Next time, contrast two of them with "however" or "on the other hand". |
Printable student self-check
A student self-check is a short version of the rubric, written as statements learners tick after listening back to their own recording. Ask students to complete it before they see your scores, then compare: the gaps between their view and yours make a useful feedback conversation.
| My speaking self-check | Yes | Not yet |
|---|---|---|
| I answered every part of the task. | ||
| I gave a reason or an example for my main point. | ||
| I kept talking without long stops. | ||
| I used linking words such as because, so and however. | ||
| I used at least three words or phrases from this unit. | ||
| I noticed and fixed at least one of my own mistakes. | ||
| A classmate could understand me without asking me to repeat. | ||
| Pair tasks: I asked my partner a question and responded to their ideas. | ||
| Next time, I will: |
For young learners or A1 classes, cut the list to three statements and read them aloud.
Where do AI speaking practice and automatic feedback fit?
AI speaking practice and automatic feedback work well for practice and formative feedback between lessons: students speak more often and get instant comments on your rubric's criteria, while the teacher reviews the results afterwards and keeps the final judgement on grades. Treat automated scores as a first read, not a verdict.
| Rubric criterion | What automatic feedback does reasonably well | Where to be careful |
|---|---|---|
| Task achievement | Checks whether the answer covers the points asked for | Relevance and appropriacy still need a teacher |
| Fluency and coherence | Flags long pauses, repetition and missing linking words | Some pauses are thinking time |
| Range and accuracy | Marks errors in the transcript and suggests corrections | A recognition slip can look like a learner error |
| Pronunciation | Scores words and sentences sound by sound | Weaker on stress and intonation |
| Interaction | Little: an AI partner is not a classmate | Assess interaction in class |
A 2024 meta-analysis by Ngo, Chen and Lai of 15 studies found that pronunciation practice with automatic speech recognition (ASR) had a medium overall effect (g = 0.69): large on individual sounds, small on stress and intonation, and stronger with explicit feedback.
ASR accuracy also varies across speakers. In a 2020 PNAS study by Koenecke and colleagues, five commercial ASR systems had an average word error rate of 0.35 for Black speakers and 0.19 for white speakers in US interviews. Systems have changed since, but check the transcript before treating a flagged word as the student's mistake. As Litman, Strik and Lim (2018) argue, whether speech technology is adequate for assessment depends on how you define the speaking skill.
A workable routine: set the task's criteria and weights to match your rubric; let students practise and get instant feedback before the deadline; then review the results, adjust any score that looks wrong, and pick one or two class-wide issues for the next lesson. Keep interaction and high-stakes grades with teachers.
In AvoLingo, an AI language-learning service by AvoLabs, teachers set their own criteria and weights for speaking homework. Conversation assignments start with Language Accuracy, Communication Effectiveness, Conversational Flow and Vocabulary Usage at 25% each, which teachers can edit, reweight or replace with this rubric's rows. Students get feedback straight away; teachers review the results afterwards and can re-grade against their own criteria. Word and sentence practice adds pronunciation scores for accuracy, fluency and completeness. See AvoLingo for schools, or our guide to AI speaking homework for task ideas.
FAQ
What criteria should an ESL speaking rubric include?
Most ESL speaking rubrics use four or five criteria: task achievement, fluency and coherence, range, accuracy, and pronunciation or intelligibility, plus interaction for pair work. These follow the CEFR's qualitative aspects of spoken language and overlap with the four IELTS Speaking criteria.
How many bands should a speaking rubric have?
Four to six bands is common. Four bands, as here, avoid an easy middle score and are quick to tell apart; add a 0 for no rateable response. More bands only help if teachers can reliably separate neighbouring bands, which calibration will show.
Can I use this rubric to give students a CEFR level?
No. The ≈CEFR labels describe one performance on one task. A claim about a learner's CEFR level needs several tasks, standardised conditions and trained raters, or an external exam. Use the labels to plan teaching, not to certify.
Can AI score speaking reliably?
Partly. AI gives useful provisional scores, especially on individual sounds and on grammar in the transcript. It is weaker on interaction, appropriacy and intonation, and speech recognition is less accurate for some speakers. Use it for practice and formative feedback, with a teacher reviewing results and deciding grades.
How do I adapt the rubric for ESP or workplace courses?
Keep the five criteria and rewrite task achievement for the field, for example "follows the steps of a hotel check-in" or "uses the correct technical terms". See our guides to ESP course design and corporate language training.
Sources
- Council of Europe (2001), CEFR Table 3: qualitative aspects of spoken language use
- Council of Europe (2020), CEFR Companion volume (PDF)
- Council of Europe, CEFR key concepts: phonological competence
- IELTS, Speaking key assessment criteria (PDF)
- IELTS, IELTS scoring in detail
- Jonsson and Svingby (2007), The use of scoring rubrics, Educational Research Review 2(2)
- Cambridge ESOL (2009), Examples of speaking performance at CEFR levels A2 to C2 (PDF)
- Hattie and Timperley (2007), The power of feedback, Review of Educational Research 77(1)
- Ngo, Chen and Lai (2024), ASR in ESL/EFL pronunciation: a meta-analysis, ReCALL 36(1)
- Koenecke et al. (2020), Racial disparities in automated speech recognition, PNAS 117(14)
- Litman, Strik and Lim (2018), Speech technologies and the assessment of second language speaking, Language Assessment Quarterly 15(3)
Written by the AvoLingo team. We used AI to help research and draft this article; sources are linked in the text.
Back to all articles