How we grade: method, evidence, and our accuracy challenge

Every report on this site is produced by the same published method. It is written down here so you can check our work — before and after you pay.

Four official criteria, equally weighted

Task 1 is scored on Task Achievement, Task 2 on Task Response; both are scored on Coherence & Cohesion, Lexical Resource, and Grammatical Range & Accuracy — the four official IELTS Writing criteria, each worth a quarter of the overall band. The overall band is the average of the four, rounded the way IELTS rounds it. There is no fifth, hidden criterion and no score for effort.

Every band is backed by your own sentences

Each criterion in your report carries a sentence copied verbatim from your essay, plus one line on what caps that criterion at its band. Above the four, the report names the single criterion costing you the most — your #1 issue — again with your own sentence as evidence. If a claim in a report cannot be pointed to in your text, it does not belong in the report.

Same essay, same report

Before any grading runs, your essay is fingerprinted. Submit the identical text again — by accident or to test us — and you get the stored report back: no new grading run, no credit charged, and no score drift between submissions. Scores only change when the text changes.

What this is — and what it is not

Our bands are examiner-calibrated AI estimates against the official descriptors; they are not official IELTS results, and only a test centre can issue those. Estimates can be wrong — which is exactly why the accuracy challenge below exists, and why we publish this method instead of asking you to trust a black box.

The accuracy challenge

If an official Enquiry on Results (EOR) re-mark of the same essay you submitted here comes back 1.0 band or more away from our overall band, send your Test Report Form to xm@sweetai.work and we refund that purchase in full. Competitors guarantee outcomes ('reach your target or your money back'); outcomes depend on the candidate. We guarantee the one thing that is ours alone: the accuracy of the score we gave you.

See a full sample report See our measured accuracy — 12 essays, four runs: high tier re-grade (Run 4) mean error 1.33 on the same 3 high-band essays, the ceiling unmoved; 0.40 on mid-band essays (Bands 5.5–7); 1.50 on high-band essays (published Bands 7–9) at the low tier See pricing and the four guarantees