Independent learners hit the same wall: Listening and Reading have answer keys, Writing does not. The short answer: self-assessment handles the countable failures well and does not replace a human on content and argument.
A four-step routine
- Step 1 — measure what is countable: word count (150 minimum for Task 1, 250 for Task 2), paragraph count, number of complex sentences, repetitions of any content word.
- Step 2 — score the four criteria separately: Task response, Coherence and cohesion, Lexical resource, Grammatical range and accuracy. One band and one line of reasoning each. No overall figure yet.
- Step 3 — read the response backwards, sentence by sentence. Isolating sentences from the argument exposes grammar errors your eye skips when reading forwards.
- Step 4 — rewrite one paragraph, the weakest, rather than the whole response. Focused revision teaches more than wholesale rewriting.
Each criterion is unpacked in IELTS Writing assessment criteria.
What automated tools do well
Automated feedback, including tools built on language models, is genuinely good at anything with an explicit rule: counting words, flagging repetition, identifying ungrammatical sentences, suggesting alternative vocabulary, spotting paragraphs with no topic sentence. That feedback arrives immediately, which a class of thirty students cannot otherwise get after every piece of writing.
Three failures tools do not catch reliably
- Fluent but off-prompt. Well-organised, well-worded, and answering a different question. This is the most expensive error under Task response.
- Invented evidence. Statistics and examples that read plausibly and are not real. A human marker notices; an automated marker often cannot tell them apart from real ones.
- Prompt-recycling. Rephrasing the prompt until the word count is met. The counter sees 250 words; a reader sees an empty response.
This is why a serious marking system checks the input before the model marks, and holds suspicious work for a teacher instead of publishing it. On LinguaGrade a response is auto-released only when it clears a confidence gate: the marking was not rejected by the validator, every quotation in the feedback traces back to the exact wording of the submitted work, no criterion is still waiting for a human, the input data is intact, and the confidence score reaches the configured threshold. Anything short of that goes into the teacher's queue. See How far you can trust AI marking.
Frequently asked questions
Can self-marking replace a teacher?
No, but it reduces how many rounds a teacher is needed for. Clearing the countable errors first is the most efficient use of a teacher's time.
How many essays should I write per week?
Fewer essays properly revised beat more essays never revised. Two a week with real feedback outperform five abandoned.
Are free online scoring tools trustworthy?
Trust them on grammar and length. Do not trust the band number, especially when the tool will not tell you which criteria produced it.