<5 s
End of speech → score
THE RUBRIC
Four criteria. Same as the real test.
Fluency & coherence
Smoothness of ideas. Penalises long pauses, fillers and stop-start delivery.
Lexical resource
Breadth, accuracy and idiom. We track novel words used and flag repeats.
Grammatical range & accuracy
Sentence variety + error count. Subjunctives and conditionals score higher than past simple.
Pronunciation (intelligibility)
Not 'British accent'. Intelligibility, checked against a CMU phoneme map.
WHY WE BUILT IT
A real examiner costs £60/hour. We do the same work for £7/month.
For three decades, the only way to know your IELTS Speaking band was to pay a tutor and wait two days for feedback. That gap — between speaking, and knowing if you got it right — is the single biggest reason learners stay stuck at 6.5. We're closing the gap to five seconds.
THE PIPELINE
From microphone to feedback in five seconds.
Audio capture (browser-side)
MediaRecorder API, 16 kHz mono. WebM on Chrome/Firefox, MP4 on Safari/iOS. Chunked over multipart so the proxy doesn't choke on long answers.
Transcription (Deepgram → Whisper fallback)
Deepgram nova-2 transcribes in ~0.8 s per 30 s of speech. Whisper kicks in only when Deepgram is degraded.
Two-lane grading (GPT-4o)
Lane A returns Angie's spoken reply in <3 s. Lane B runs the full Cambridge rubric in 8–9 s and patches your bands + corrections into the chat afterwards. No loading spinner.
Audit & feedback loop
1% of grading is sampled weekly and re-graded by Cambridge ESOL-trained linguists. The system prompt is retuned monthly to keep AI bands within ±0.25 of human bands.
OUR COACHES
The humans behind the AI.
Every line of feedback Angie produces is calibrated against a panel of human coaches — qualified Cambridge CELTA / DELTA examiners, ESOL-trained linguists, and serving IELTS senior examiners across the UK, Ireland and Australia. They review a sampled 1% of all sessions every week and re-grade them blind.
Cambridge-aligned linguists
CELTA / DELTA Module 1+2. A minimum of 5 years teaching IELTS prep, 2 of those as a school IELTS lead.
Working IELTS examiners
Currently certified Speaking + Writing examiners at British Council / IDP test centres. Anonymous panel; conflicts logged.
Phonetics specialists
MA / PhD in applied linguistics. Own the IPA + phoneme intelligibility models that drive pronunciation scoring.
Safeguarding lead
Designated Safeguarding Lead reviews all under-18 sessions surfacing safeguarding signals (UK KCSiE 2024).
AUDIT PANEL
Dr. Rachel Foster
Cambridge CELTA examiner · Britwise audit panel
“I joined sceptical. After six months auditing 200+ sessions a week, the rubric alignment is honestly tighter than what I see across most physical test centres — and the feedback is instant.”
WHAT WE WON'T DO
Three lines we refuse to cross.
Penalise non-RP British accents during pronunciation scoring
Quietly retain your audio to train third-party voice models
Inflate your band score to keep you subscribed for longer
Hide the rubric so you can't argue with the grade
Score intelligibility, not Britishness — Indian, Nigerian, Caribbean accents are graded fairly
Audio deleted within minutes; transcript-only by default
Your band is your band — even when it's bad news
Publish the rubric and the cut-offs on this page, in plain English
+0.5 band in 30 days — or your money back.
We publish the maths on /efficacy and refund without questions if you complete the schedule and don't move. Email refund@britwise.school.
FAQ
Methodology FAQ.
Why is Britwise faster than a human examiner?
What's the catch with AI grading?
Do you actually grade *non-British* accents the same?
What rubric do you use for non-IELTS speaking?
Can the audit panel be wrong?