IELTS Pronunciation: what the criterion measures, and why your accent isn't it

Pronunciation carries the same weight as grammar in your Speaking band — a full quarter — and it is the criterion candidates prepare for least and misunderstand most. The central misunderstanding: that it grades accent, and that the path to band 8 sounds British or American. The descriptors say otherwise. What is measured is intelligibility — how much effort the listener spends decoding you — and the range of pronunciation features you control. Band 9 speakers keep their accents; L1 influence is explicitly acceptable at high bands so long as it rarely interferes with understanding.

What "features" means

The descriptors credit a range of pronunciation features used with control. Concretely, examiners are listening for four layers:

Individual sounds. Consistent, systematic substitutions that create wrong words are what cost marks — /p/ vs /b/, /l/ vs /r/, the two th sounds, dropped final consonants (fine for find). Occasional slips barely register; a substitution that runs through the whole test does.

Word stress. English stores stress in the word — PHO-to-graph, pho-TO-graph-er — and misplaced stress damages intelligibility more than any single wrong consonant, because listeners retrieve words by their stress shape. It also silently interacts with vocabulary: the "less common words" you learned from reading often arrive with guessed, wrong stress, so every new word should be learned with its stress marked.

Sentence stress and rhythm. English compresses function words and spends time on content words — I've been WORKing on it since JANuary. Equal weight on every syllable is the single most common feature gap at band 6, and it makes accurate English tiring to follow, which is the exact thing the criterion prices.

Intonation. Rising and falling patterns that carry meaning: lists that rise then fall, questions distinguished from statements, contrast marked by pitch (I liked the FILM — not the BOOK). Flat delivery reads as "limited range of features" even when every sound is clean.

Where the bands actually divide

Reading the descriptors as a listener-effort scale is the clearest way in. Around band 5, mispronunciations are frequent enough to cause the listener real, recurring difficulty. Band 6 is generally intelligible with feature control that comes and goes. Band 7 mixes the band 6 and band 8 pictures — sustained stretches of effortless listening with lapses. At band 8, the full range of features is used flexibly and only occasional lapses occur; accent affects nothing. The trajectory is: fewer decoding stumbles, more features under deliberate control.

How to actually train it

Pronunciation improves through feedback loops, not awareness alone:

• Find your systematic errors first. Record two minutes of your speaking and listen for the sounds and stress patterns that repeat — or get AI-graded Speaking feedback that names them for you. Every learner has three to five; drilling sounds you already control is the common wasted hour.
• Shadowing, ten minutes daily. Take a sentence of natural audio, replay it, and speak with it, matching stress and intonation rather than just words. Shadowing trains rhythm and sentence stress faster than any explicit rule, and podcast-plus-transcript is all the equipment it needs.
• Minimal pairs for your two worst sounds — light/right, pass/path — a few minutes a day, listening then producing. Contrast is what re-trains the ear.
• Mark stress in your vocabulary notes. From today, no new word enters your notebook without its stressed syllable underlined. This is the highest-leverage habit on the list because it stops the problem growing.

The reframe that unlocks preparation: stop asking "how do I lose my accent?" — you don't need to, and trying wastes months. Ask "where does my speech make the listener work?", find those five places, and close them. That is the entire criterion.