Speaking is the IELTS paper candidates understand least well, because it is the only one where you never see the marking happen. There is no answer sheet to review afterwards and no raw score to convert. An examiner sits opposite you for eleven to fourteen minutes and produces four numbers.
Those four numbers are more knowable than they feel.
The three parts
Part 1 (4–5 minutes). Identity check, then straightforward questions about familiar subjects — where you live, your work or studies, food, weather, hobbies. The questions are easy on purpose. This part settles you and gives the examiner a baseline.
Part 2 (3–4 minutes). The long turn. You get a cue card with a topic and prompts, one minute to prepare with paper and pencil, then you speak for one to two minutes uninterrupted. One or two short follow-up questions close it.
Part 3 (4–5 minutes). A two-way discussion of abstract themes connected to your Part 2 topic. If Part 2 asked about a teacher who influenced you, Part 3 moves to education systems, how teaching has changed, whether technology can replace teachers. This is where the highest bands are separated.
The whole test is recorded. The recording exists for quality control and remarking, not to catch you out.
The four criteria, each worth 25%
Your Speaking band is the average of four equally weighted criteria. No single one dominates, which means a serious weakness in any one of them costs you roughly a quarter of your score.
Fluency and Coherence
The ability to keep going at a natural pace, and to organise what you say so it follows.
Examiners are listening for hesitation that comes from searching for language — not hesitation from thinking about the idea. Pausing to consider a genuinely difficult Part 3 question is normal and does not hurt you. Pausing mid-sentence because you cannot retrieve a word does.
Self-correction works the same way. Occasional correction is natural speech. Repeatedly restarting sentences signals that you are assembling language rather than using it.
Coherence is the other half: connecting ideas so an argument develops. Linking words help, but a string of firstly, secondly, moreover laid over disconnected points does not create coherence — it advertises the absence of it.
Lexical Resource
Vocabulary range and, more importantly, precision.
The reward is for using the right word naturally, not for using an impressive one. A memorised advanced word deployed slightly wrongly damages this criterion more than a simple word used correctly. Examiners recognise rehearsed vocabulary easily, because it does not match the register of everything around it.
Two things genuinely lift this score. Paraphrase — being able to express an idea another way when the exact word does not come — is explicitly valued, and it is also the practical solution to hesitation. Collocation — knowing that we say heavy traffic and strong opinion rather than the reverse — is much of what separates Band 7 from Band 8.
Grammatical Range and Accuracy
A mix of structures, used accurately.
Range and accuracy pull against each other, and how you handle that tension is what is being assessed. Speaking only in short simple sentences protects accuracy while capping range. Attempting complex structures you do not control produces errors that obscure meaning.
Band 7 is usually described as frequent error-free sentences with good control of complex forms; some errors persist but rarely impede understanding. That is the realistic target: attempt complex structures, accept occasional errors, and make sure the listener never loses the thread.
Conditionals and past tenses cost candidates the most marks, mainly because Part 3 questions invite hypotheticals and Part 2 usually requires narrating something that already happened.
Pronunciation
Being understood without the listener working at it.
This criterion is widely misunderstood as an accent test. It is not. There is no target accent and no advantage to sounding British or American. What is assessed is intelligibility: individual sounds, word and sentence stress, rhythm, and intonation.
Stress matters more than most candidates expect. English carries meaning in stressed syllables, and misplacing stress can make a correctly pronounced word unrecognisable. Flat, monotone delivery also costs marks here even when every sound is technically accurate, because intonation carries meaning too.
How the four combine
The Speaking band is the average of the four, reported in whole and half bands. It then joins your other three skills and is averaged and rounded again into your overall band.
The practical consequence is that the four criteria are worth attacking unevenly. Most candidates are not equally weak across all four, and lifting the weakest one is worth four times more than polishing the strongest. A candidate at 7 / 7 / 7 / 5 in the four criteria averages 6.5; fixing only the pronunciation weakness reaches 7.0 without touching anything else.
What actually improves it
Speaking is the skill where practice most often means rehearsal, and rehearsal is the thing the test is built to detect. Memorised answers break down the moment Part 3 pushes somewhere unscripted, and the shift between rehearsed and spontaneous language is audible.
What works better is producing spoken answers under time pressure and then reviewing them against the four criteria separately — which is difficult to do for yourself, because the criterion you are weakest on is usually the one you cannot hear. Recording yourself on real Part 2 cue cards and getting the four criteria scored independently is the version of this that closes the loop; speaking practice in IELTS Studio is built around exactly that.
For what the band numbers mean across all four skills, see what every IELTS band level means. The writing equivalent of this breakdown is how the writing band scale works.