IELTS Speaking Band Descriptors: What Fluency and Coherence Rewards
The IELTS speaking band descriptors reward something narrower than most candidates assume: not speed but a turn that keeps its thread. Here is the published fluency and coherence wording at Band 6, 7 and 8, one answer annotated at both bands, and a two-minute recording check.
On this page
Most candidates who are told they "sound fluent" but stay at Band 6 are not short of speed. They are short of one specific behaviour that the IELTS speaking band descriptors name in their first column: keeping a long turn together while you repair it. The criterion is not a word-per-minute limit, and it is not a penalty for thinking. It rewards speech that keeps its shape when the speaker pauses, backtracks or changes direction — and the published wording says so with more precision than most preparation advice admits.
This guide reads the fluency and coherence column on its own. It compares the Band 6, 7 and 8 wording against the same answer, separates the linking language that helps from the fillers that quietly cost you, and ends with a two-minute recording check you can run without an examiner in the room.
What the criterion actually measures
The Speaking test is a face-to-face interview of 11 to 14 minutes in three parts, assessed by a certificated examiner on four criteria: fluency and coherence, lexical resource, grammatical range and accuracy, and pronunciation (IELTS Academic: Speaking test format, accessed 4 November 2026). IELTS states that the four criteria carry equal weighting and that their average gives the Speaking score (IELTS scoring in detail, accessed 4 November 2026).
Two judgements in one column
Fluency and coherence bundles two separate questions. Fluency asks whether speech keeps moving: how often it stops, whether the same word keeps returning, whether a speaker can repair a sentence and carry on without restarting it. Coherence asks whether a listener can follow the thread: whether ideas are introduced, developed and closed in a sensible order, and whether the linking language matches the relationship between two ideas rather than merely announcing that a new one is coming.
The two can drift apart, which is why a single answer often feels fluent and unclear at the same time. A speaker who never pauses but jumps between three unrelated points has fluency without coherence. A speaker who is slow, deliberate and easy to follow has coherence without much fluency. The bands above 6 are where the two have to arrive together.
What the descriptors do not reward
Three assumptions do not survive a careful read. Speed is the first: no band wording mentions words per minute, and a fast answer with three broken repairs reads lower than a slower answer that holds its thread. Accent is the second: the fluency column does not mention accent at all, and the pronunciation column refers to it only through its effect on intelligibility. Memorised openers are the third: the descriptors reward a "willingness to speak at length" at Band 6 and speaking "at length without noticeable effort" at Band 7, and a scripted answer tends to lose both the moment the examiner asks a follow-up question.
One published note matters more than any of these before you score yourself. The descriptors state that a candidate must fully fit the positive features of the descriptor at a particular level, and that the rating is based on their average performance across all parts of the test (IELTS Speaking band descriptors, accessed 4 November 2026). "Fully fit" is the phrase behind the mock-test comment many candidates receive. An answer can sound fluent in Part 1 and still sit at Band 6 overall if Part 3 keeps losing its thread.
Band 6, 7 and 8 in the examiner's wording
The table below quotes the public descriptors for this criterion only. The right-hand column is the reading that matters while you practise: what actually changes between one band and the next.
| Band | The published fluency and coherence wording | What it changes in your answer |
|---|---|---|
| 6 | willing to speak at length, "though may lose coherence at times due to occasional repetition, self-correction or hesitation"; connectives used "not always appropriately" | You finish the turn, but the listener sometimes has to rebuild the thread for you |
| 7 | "speaks at length without noticeable effort or loss of coherence"; some "language-related hesitation"; connectives used "with some flexibility" | Effort stops being audible, and repairs no longer cost the thread |
| 8 | "only occasional repetition or self-correction"; hesitation "usually content-related and only rarely to search for language" | Pauses carry meaning: you stop to choose an idea, not to hunt for a word |
| 9 | hesitation is "content-related rather than to find words or grammar" | Flow and connection stop being something the listener notices at all |
Read down the third column and the movement is a subtraction rather than an addition. Band 6 is not Band 5 plus better grammar; it is a performance where the listener stops having to repair your sentence for you. Band 7 is Band 6 with the effort made inaudible. Band 8 is Band 7 with the repairs aimed at content instead of language. Nothing in that progression asks you to speak faster, and nothing asks you to eliminate hesitation altogether.
One answer, annotated at Band 6 and Band 7
The card below was written for this guide in the style of a Part 2 task. It is a teaching example, not a past paper question.
Describe a skill you picked up later than most people and how the first month went.
The annotation compares one constructed answer against the two bands, and it deliberately ignores vocabulary, grammar and pronunciation. Everything marked here belongs to fluency and coherence alone, which is the only way to see what this one criterion is doing.
| What the answer does | Read against the Band 6 wording | Read against the Band 7 wording |
|---|---|---|
opens with a stack of hedges: So, um, I guess I started, like, quite late, maybe at twenty-five | Willing to talk, but four fillers pass before the first idea arrives, so coherence starts late | I started learning to swim at twenty-five. Content first, no runway before the claim |
repairs a sentence by repeating it: I mean, I could float by week two. | The repetition restarts the sentence and the listener loses what the repair was for | What I mean is, I could float, but I could not swim a length. The repair adds the contrast instead of restating the line |
links with a connective whose referent is missing: And because of that, the coach moved me down a group. | The connective is correct in form, but "that" points at nothing, so the link is decorative | So the coach moved me down a group. The consequence is named in the same breath as the event |
hesitates inside a clause: It took me about, um, six weeks | The pause lands mid-phrase, which reads as a search for language and breaks the clause | It took me about six weeks, actually, to stop panicking in deep water. The pause moves to a boundary and becomes emphasis |
The four rows carry the same story, the same facts and the same speaker. What separates them is not ambition but placement: where the pause goes, what the repair adds, and whether the connective has something to point at. That is the part of this criterion a candidate can change in a week of deliberate practice, and it is also the part that examiners describe as "not noticeable effort" once it becomes automatic.
Linking words that help and fillers that hurt
Candidates are usually told to "use more linking words", and the advice is half right. The descriptors do ask for discourse markers and connectives, but they ask for them with flexibility, and the Band 6 wording adds the warning that they are used "not always appropriately". A connective earns its place when it marks a real relationship: contrast, cause, sequence, concession. It costs you when it announces a relationship the answer does not actually contain.
| Device | What it does for coherence | Where it turns into a problem |
|---|---|---|
contrast markers such as however, on the other hand | Signals that the next idea contradicts the last one | Used to start a sentence that agrees with the previous one |
cause markers such as so, as a result | Makes a consequence explicit and closes a thought | Attached to a cause the answer never stated |
sequence markers such as firstly, after that | Gives the listener a map for a long turn | Applied to a list that is not actually ordered |
fillers such as um, like, you know | Buys a moment, nothing more | Repeated four times before the first idea, as in the Band 6 row above |
The Band 5 wording names the failure mode precisely: candidates at that level "may over-use certain connectives and discourse markers". The fix is not to delete them but to give each one a job. If a linking phrase can be removed without changing the relationship between two sentences, it was decoration, and decoration is what the band 7 and 8 wording stops rewarding.
Self-repair that reads as control
Self-correction is not a fault in the descriptors. Band 9 allows "only very occasional repetition or self-correction", and Band 7 explicitly permits "some repetition and/or self-correction" as long as it does not cost coherence. What separates a repair that reads as control from one that reads as panic is whether the listener can hear what the repair was for.
Repair the phrase, not the sentence
Compare the two repairs from the table above. The Band 6 version restates the whole claim, so the listener has to work out which part was wrong. The Band 7 version replaces the phrase and adds the missing contrast, so the repair carries information:
The breathing — sorry, the timing of the breathing — was the hard part.
That single line shows the pattern. Name the word you want, mark the repair with a short signal, and continue the clause. The listener hears a speaker who noticed a problem and fixed it, which is exactly the behaviour the band 7 wording places under "without loss of coherence".
The three repairs that cost you coherence
- Restarting from the beginning of the sentence after a small error, which throws away the part that was already clear.
- Repeating the same phrase three times while searching for a synonym, which reads as a language problem rather than a correction.
- Abandoning the point and starting a new one, which leaves the listener with a topic that was never closed.
None of these three is a grammar error. Each one is a coherence error, and each one is fixable with a rehearsed repair habit rather than more vocabulary.
A two-minute recording check for fluency alone
The test of this criterion is what a listener hears, so the check has to be a recording rather than a self-assessment in your head. Set a timer, answer one Part 2 card, and score nothing except fluency and coherence.
- Record a full two-minute answer on your phone, without stopping the recording to restart the answer.
- Listen once without writing anything, and mark only the moments where you lost the thread.
- Listen again and label every hesitation as content-related or language-related. The Band 8 wording turns on that distinction.
- Count your repairs. For each one, write down what the repair added: a correction, a contrast, or nothing at all.
- Check your connectives against the table above, and delete the ones that could be removed without changing the meaning.
- Compare the recording against the wording, not against a model answer, and change one habit before the next recording.
The numbers are for comparison, not for scoring: two language-related repairs in two minutes is a different performance from eight, and the descriptors describe the difference in words rather than in counts. Keep the recordings dated, and compare recordings two weeks apart rather than two days apart.
FAQ
Does speaking faster improve my fluency and coherence score?
No, and it often works the other way. The published wording describes fluency as speaking "at length without noticeable effort" and coherence as development that keeps its thread; neither depends on speed. A faster delivery compresses the pauses but also compresses the repairs, so a listener hears more restarting per minute. In the recording check above, most candidates who slow down by roughly a tenth find that their repairs become audible as repairs rather than as breakdowns, which is the direction this criterion rewards.
How much hesitation is still acceptable at Band 7?
The Band 7 wording allows "some hesitation, repetition and/or self-correction" as long as it does not affect coherence, and it adds that such hesitation often occurs mid-sentence and can indicate problems with accessing language. So the amount is less important than the effect: hesitation that leaves the thread intact is compatible with Band 7, while hesitation that forces the listener to reconstruct the point is not. Band 8 narrows the same allowance to hesitation that is usually content-related, and Band 9 to hesitation used only to prepare the next idea.
Do memorised linking phrases make an answer more coherent?
They make an answer sound organised for about thirty seconds, and then the effect turns against you. The descriptors reward connectives used with flexibility and, at Band 6, warn about ones used not always appropriately; a memorised list produces the second pattern rather than the first, because the phrases are chosen before the ideas exist. A more reliable route is to keep a small set of connectives with clear jobs, as in the table above, and practise them on new topics until the choice is made by the relationship rather than by habit.
What to change in your next recording
Read the fluency and coherence column once, in the examiner's words, and then stop reading it. The rest of the work happens in recordings: one card, two minutes, one habit changed at a time. If you want the four criteria side by side before you start, the wider guide to the IELTS Speaking band descriptors explains how the columns combine into a single score, and a direct comparison of the same answer at two bands is a useful second reference once this criterion is clear in your head.
The change that moves most Band 6 candidates is not a longer answer. It is a repair that keeps its thread and a connective that has something to point at.