Same essay · same prompt to every AI · very different bands
Ask a free tool, and a real 4.0 student hears 1.5–3.0 — Gemini says 2.0. We say 4.0: the examiner's own band, exactly.
The examiner says 5.0; we read this 144-word answer 5.5. Half a band high — level with Gemini and DeepSeek, both half a band low. Printed because you can only trust the wins if we show the ties.
A 5.5 answer, read 4.5–5.0 by most chatbots. We read 6.0 — half a band high, and DeepSeek is the one that read it exactly.
Believe the chatbots and you’d think you’re a 4.5 student — this answer was a 6.0 all along.
A Band-7 essay, told it’s 5.5 — that’s paying for another course you never needed.
Half a band low, and Gemini read it exactly. We print the paper we lose, next to the ones we win.
Nobody's AI gives IDP's own perfect answer its 9 — half call it 5.5–6.5. We read 7.5, and Gemini reads it closer than we do. This is the card we lose, printed next to the ones we win.
You paste your essay into ChatGPT. It calls a real 7.0 essay a 6.0 — we checked, on examiner-scored answers. We grade on the official criteria and show exactly where the band went — in Vietnamese. Here’s the receipt.
Free · no account, no card · practice estimate, not an official IELTS score
Full access from $2.99/week (~73.000đ) · cancel anytime
Don’t take our word for it — Cambridge 19 is in every bookstore, examiner scores printed inside. Check the answers yourself, or read the full methodology →
Across all ten examiner-scored answers — our average miss: 0.30 of a band. Gemini 0.70 · DeepSeek 0.90 · ChatGPT 0.95 · Claude 1.50. Nine of the ten land within half a band. The tenth — a Band 4.0 answer we read 5.0 — is our worst of the set, and it is counted in that average.
Published, examiner-scored Cambridge 19 and 20 answers plus IDP's published Band 9 sample — you can look every one of them up. Every system, ours included, graded every answer three times with the identical prompt, and every number shown is that system's median (9 Aug 2026 (chatbots) · 2 Sep 2026 (ours)). Model versions and per-answer raw numbers are on the accuracy page. We're closest, not perfect: our average miss is 0.30 of a band — five of the ten exact, four out by half a band, one a full band high — and we print all of it.
Average miss vs the examiner’s band — smaller is better.
Worst single miss: IELTS Ace 1.0 · ChatGPT 1.5 · DeepSeek 1.5 · Gemini 2.0 · Claude 2.5 bands.
“Overall the essay fits the band 7 description. Good range of complex sentences. To improve, use more advanced vocabulary and check your grammar carefully.”
✗ which sentence lost the band — not shown✗ your words, marked — no✗ a corrected rewrite — no
See this full report →“Overall, the speech fits well within the band 7 description due to its smooth flow, despite the slow pace and lack of cohesive devices affecting clarity.”
✗ which word was mispronounced — not marked✗ where you hesitated, on your transcript — not shown✗ a model answer — no
See this full report →
See this full report →
See this full report →
Liz evaluates, teaches, and practices with you. One voice across writing, speaking, and your daily study plan.
Submit writing or speaking and get a band estimate with margin notes — in your language.
Part 1, 2, and 3 drills with real pronunciation feedback and follow-up questions.
Structured courses with short lessons, practice, and progress tracking tailored to you.
Any Task 1 or Task 2 prompt, any speaking part. No account needed for your first check.
Independent observers count your errors, devices and evidence; the official band descriptors turn the counts into a band. No single AI’s opinion.
Your band with every criterion, the exact sentences that cost you, and the fastest path up — explained in Vietnamese.
Writing and Speaking scored on the official criteria — plus Reading and Listening feedback that shows the exact sentence that proves each answer. Nobody else does that part.
Your goal band, exam date and weak areas, referenced in every lesson. Text chat or live voice conversation.
Timed, band-weighted, all four skills — with a calibrated study plan after each one.
Every explanation available in Vietnamese — because “improve your cohesion” helps nobody in a language you’re still learning.
Free
$0
always free · no card
Weekly
$2.99/wk
~73.000đ · cancel anytime
Monthly
$9.99/mo
~243.000đ · cancel anytime
Exam Pack
$19.99
one-time · 30 days
No. It’s a practice estimate. On published examiner-scored answers our average miss is 0.30 of a band — measured, and published on this page — but only IDP and the British Council issue real scores.
Every essay in the comparison is published with its examiner score — Cambridge 19 is in any bookstore. Check the answers yourself, or read the full methodology with raw numbers.
Because it never hears you. Two of the four official criteria — Pronunciation, and the fluency half of Fluency & Coherence — live in the sound: pauses, pace, fillers, hesitation, intonation. A chatbot reads a transcript you typed, and typed transcripts flatter you: the “um”s disappear, autocorrect fixes your grammar, and you write what you meant to say, not what you said. We score the recording itself — word-level pronunciation, timing, fillers — and then apply the official descriptors to what was actually said.
Yes — the scoring reads your English, and every explanation of what to fix is available in Vietnamese.
Anytime, in one click — weekly and monthly plans stop at the end of the paid period. The Exam Pack is a single payment; there’s nothing to cancel.
Two minutes, free, no account. A practice estimate — not an official IELTS score.
Check my band — free