Who marks your IELTS Writing and Speaking, and is a computer test marked differently?
Certified examiners mark Writing and Speaking whichever way you sit the test. What that means for the practice feedback you use before test day.
ComparisonsGeneralUpdated 6 min read
A trained, certified human examiner awards your Writing and Speaking bands. Not an algorithm, not an "electronic rater" running quietly behind the screen, and not a different standard because you typed instead of wrote. The criteria are published, the marking is criterion-referenced against those published descriptors, and the task types, criteria and marking are the same on computer as on paper. So the question worth your attention is not who marks the real test. It is what any practice marker can actually see, because that gap is where candidates lose weeks believing a number that was never measuring what they thought.
The short answer, and where computers do come in
Writing is marked against four criteria: Task Achievement (Task 1) or Task Response (Task 2), Coherence and Cohesion, Lexical Resource, and Grammatical Range and Accuracy. Speaking is marked against Fluency and Coherence, Lexical Resource, Grammatical Range and Accuracy, and Pronunciation. Each criterion gets its own band from the published descriptors, and those combine into the skill band. Criterion-referenced means you are measured against a described standard, not ranked against the other people in the room.
Computers do mark IELTS, but only where no judgement is needed: Reading and Listening are right or wrong against an answer key. Writing and Speaking need a human because the descriptors describe qualities, not correct answers. Speaking is a conversation with an examiner whichever mode you book. That is why "is computer-delivered marking stricter?" has a flat answer: the criteria and the marker do not change, so the mode is a comfort-and-logistics decision, covered properly in computer or paper.
What the real marker has that a practice marker does not
This is the comparison worth making: what an examiner works from, set against what an automated scorer or a general-purpose chatbot has when you paste an answer into it.
| What the band depends on | A certified examiner | An automated or general-purpose marker | What that means for your practice |
|---|---|---|---|
| The four published criteria | Applies each one separately against its own descriptor wording, then combines | Can name the criteria; in practice reacts to surface features it can measure | Ask for one criterion at a time and read the reasons; bin the single overall number |
| The whole performance | Reads your full script beside the actual task prompt, including the Task 1 chart, map or process you were given | Sees only what you paste. Usually the essay, often not the prompt, rarely the visual, never your voice | Always paste the exact prompt, and describe the visual in full, or every Task Achievement comment is guesswork |
| A shared standard | Certified and monitored, standardised against sample scripts so one examiner's 7.0 means the same as another's | No shared standard behind it. Ask a general-purpose chatbot for a band and the same script can come back a band apart on two different days, and the way you phrase your request nudges it; a tool that marks criterion by criterion is at least steady with itself, but steady is not the same as calibrated to the descriptors | Never quote a practice number as your band. Treat repeated, specific criticism as the signal |
| Exam conditions | Marks what you produced inside the time limit, with no edits after it | Marks whatever you submit, however long you took over it | Time every practice piece. Feedback on an untimed, tidied essay flatters control you will not have on the day |
| Pronunciation and interaction | Hears stress, rhythm and intelligibility live, and can push with follow-up questions in Part 3 | Works from text at best. Cannot hear you and will not press you | Record yourself and listen back. Use text tools for vocabulary and grammar, your ears or a person for fluency and pronunciation |
| Off-topic, underlength, memorised material | Judges relevance to the specific question and applies the consequences | Tends to reward visible structure and miss that the response never answered the question asked | Before asking anything for feedback, check the word count yourself and underline the question's instruction words |
Note what the right-hand column cannot reach: the shared standard, and the full performance under real conditions. That is exactly where a band number you could rely on would have to come from.
So what is practice feedback good for?
Plenty, once you stop asking it for a score.
- Error inventory. Which grammar structures break when you are rushed, which collocations you get wrong every time, where articles and prepositions leak. This is mechanical, repeatable, and a machine is genuinely good at it.
- Structure checks. Does each body paragraph have one controlling idea, does the position stay consistent from introduction to conclusion, are the cohesive devices doing work or just decorating sentence openings.
- Coverage checks. For Task 1, whether you reported the largest movements and compared rather than listing every figure. For Task 2, whether you answered all parts of the question.
What it cannot give you is calibration, and that matters because the distance itself is long: as a rough average, a half band costs something like 200 to 240 hours of instruction and practice, and the climb slows above band 6. That arithmetic, and where the figure comes from, is worked through in how many hours IELTS preparation takes. Chasing a number that wobbles by a band between attempts wastes the only resource that actually buys progress, which is focused repetition on a known weakness.
If you are weighing how much confidence to put in practice feedback against a human reader, the trade-offs are set out in AI feedback compared with a private tutor.