IELTS Writing Task 2 AI prompts

Almost every IELTS writing prompt circulating online is one sentence long: "act as an IELTS examiner and score my essay." It returns a number, the number is usually flattering, and running the same essay twice returns two different numbers. That is not a marking tool — it is a random band generator with a friendly tone.

The prompts in this category are built around four mechanisms that fix it. First, the rubric is loaded into context rather than recalled from memory, because a model paraphrasing the descriptors is not applying them. Second, an output contract — a fixed table, a fixed field list — so two runs are comparable and inflation has nowhere to hide. Third, evidence before judgement: a model forced to quote your text before assigning a criterion band cannot work backwards from a number it already picked. Fourth, a calibration anchor where possible, so the standard being applied is one you can point at.

What none of them do is give you a trustworthy band on their own. In our own study of 1,200 essays, purpose-built AI grading matched the human examiner consensus within ±0.5 bands 94.2% of the time and exactly 78.3% of the time. A general-purpose chatbot with a pasted rubric is looser than that, and its number moves with the framing of the prompt — which is exactly what the "defend a lower band" prompt below is designed to expose. Use these prompts for the reasoning, not the digit.

The order that works: mark the essay with evidence first, then stress-test that mark by making the model argue for half a band lower, then take whichever specific weakness both passes agree on and fix that one thing. Coherence and cohesion is usually the criterion worth isolating on its own, because it is the one candidates self-assess worst — your own writing always feels logical to you.