How BAND9AI Scores IELTS Speaking Practice

FC · LR · GRA · Pron · May 2026

Direct answer

BAND9AI scores Speaking practice recordings against the four IELTS criteria: Fluency and Coherence (FC), Lexical Resource (LR), Grammatical Range and Accuracy (GRA), and Pronunciation (Pron). Each criterion gets a band-style sub-score plus leak notes tied to descriptors—not a single “sounds good” label. Audio is transcribed and analyzed for pace, development, vocabulary precision, error density, and intelligibility. Use sub-scores to pick one fix per week; the overall band is the average logic examiners use, not a chatbot guess.

From recording to four criterion scores

Capture You submit a practice response (Part 1–3 style prompts supported in product flows)

Transcript + audio Speech-to-text plus delivery features (pauses, rate, clarity)

Rubric map Feedback mapped to FC, LR, GRA, Pron descriptors

Unlike generic tutors, the goal is examiner-shaped diagnostics—see AI Speaking limits for what still needs human mocks.

Fluency and Coherence (FC)

FC scoring weighs how directly you answer, how ideas develop, and how smooth delivery is—not speed alone. Long pauses mid-idea, circular answers, and off-topic drift lower FC even when vocabulary is strong. BAND9AI flags missing extensions in Part 3 and “false fluency” from memorized chunks.

Development: do you finish the logical arc of the question?
Coherence: do follow-ups connect or reset?
Flow: pauses at clause boundaries vs random freezes

Lexical Resource (LR)

LR is precision and flexibility—not rare words dropped at random. Scoring looks for topic-appropriate paraphrase, collocations that sound natural, and repair when you miss a word. Over-repetition of basic adjectives or template phrases caps LR even if grammar is clean.

Grammatical Range and Accuracy (GRA)

GRA balances error frequency with range: mix of simple and complex structures, control in extended speech, and errors that block meaning vs slips. Part 3 often exposes tense and clause errors Part 1 hides.

Pronunciation (Pron)

Pron scores intelligibility in context—segment clarity, stress, and rhythm that help the examiner follow ideas. Accent is fine; systematic sound errors or flat stress on key words lower Pron. Pair with how AI evaluates pronunciation.

How to use BAND9AI Speaking scores

Record under light time pressure—not scripted perfection.
Note lowest criterion; drill that one dimension for seven days.
Re-test same question type; compare sub-scores, not only overall band.
Book human mock when FC and Pron diverge—sign of delivery vs content mismatch.

Key takeaways

Four sub-scores: FC, LR, GRA, Pron—averaged like examiners.
Audio + transcript beats text-only Speaking feedback.
Fix one criterion per week based on lowest leak.
Validate spontaneous Part 3 with humans before exam day.

FAQ

Examiner rounding differs slightly—use sub-scores for study focus.

Often—always add unscripted Part 3 in the same session.

Yes—track criterion trends on identical prompts weekly.

Get four-criterion Speaking diagnostics—not generic praise.

Get Speaking Reality Check →