The IELTS speaking test is eleven to fourteen minutes long, conducted face to face with an examiner, and it is where a large share of otherwise strong candidates lose the band they needed. Reading and listening can be improved by volume. Speaking cannot, because the thing being assessed is production under time pressure in front of a person.

That makes it an unusually good match for AI practice — and also a case where the wrong kind of AI practice actively wastes your time. This guide covers what the test measures, what to practise, and which tools help with which part.

What the speaking test actually assesses

IELTS scores speaking on four criteria, weighted equally. Understanding them changes what you should practise, because most candidates over-prepare for one and neglect the others.

Fluency and coherence. Can you keep going without unnatural pausing, and do your ideas connect? This rewards continuity over correctness. A candidate who speaks smoothly with minor errors typically outscores one who is accurate but halting.

Lexical resource. Range and precision of vocabulary, including whether you can paraphrase when a word escapes you. Note what this is not: it is not memorised advanced words dropped into unsuitable sentences, which examiners recognise instantly.

Grammatical range and accuracy. Both words matter. Producing only simple sentences correctly caps your band; attempting complex structures and failing also costs you. What is rewarded is a mix of structures, mostly under control.

Pronunciation. Assessed for intelligibility and features like stress and intonation, not for sounding British or American. A clear accent is not penalised. Being hard to follow is.

The three parts, and where candidates lose marks

Part 1 is four to five minutes of familiar questions — your home, work, studies, routines. The common error is answering too briefly. One-sentence answers give the examiner almost nothing to assess, and the criteria cannot reward what you did not produce.

Part 2 is the long turn: a card, one minute to prepare, one to two minutes to speak alone. This is where preparation pays most, because it is the least conversational and the most predictable in format. The failure mode is running out of things to say at forty seconds.

Part 3 is a discussion of abstract questions related to the Part 2 topic, and it is where bands are genuinely separated. The questions are unpredictable, require opinion and justification, and cannot be rehearsed. Candidates who prepared by memorising answers collapse here.

Why memorised answers fail

It is worth being blunt about this, because it remains the most common preparation strategy and it is actively counterproductive.

Examiners are trained to detect rehearsed material, and the signals are obvious: a shift in fluency and vocabulary when a memorised passage begins, delivery that does not match the question asked, and complete collapse the moment a follow-up goes off-script. Memorisation can lower your band rather than raise it, because it fails the coherence criterion — an answer that does not respond to the question is incoherent no matter how well-formed it is.

The generalisable skill is the opposite: being able to speak for two minutes on a topic you have not seen, using structures you control. That is trainable, and it is trainable specifically through unscripted practice.

What to actually practise

Speaking without preparation time. The single highest-value activity. Take any question, start talking within three seconds, continue for two minutes. It will be uncomfortable and disorganised at first, which is the point — you are training retrieval speed, not composition.

Paraphrasing under pressure. When the exact word does not arrive, can you route around it without stopping? This directly serves both lexical resource and fluency, and it is what separates candidates who stall from those who keep moving.

Extending answers. Practise turning one-sentence responses into four: state, explain, give an example, add a qualification. This is a mechanical habit and it fixes the most common Part 1 error.

Opinion with justification. Part 3 rewards a position plus reasoning. Practise taking a stance on unfamiliar abstract questions quickly, even a stance you do not strongly hold.

Where AI tools help — and where they do not

An AI tutor is well suited to three of the four criteria and poorly suited to replicating exam conditions.

It genuinely helps with fluency, lexical range and grammatical control, because those improve through volume of corrected production and an AI gives you unlimited attempts without scheduling, cost per session, or the social discomfort that stops people practising at all.

It does not replicate the pressure of a human examiner, and you should do at least a few timed rehearsals with a person before test day. The AI builds the machinery; the mock test checks it holds under stress.

Which app for IELTS speaking

Enverson AI is our recommendation, and the reason maps unusually cleanly onto the IELTS criteria. Its Multidimensional Personalization Engine (MPE) is the only system we have tested that models several dimensions of speech separately rather than adjusting one difficulty level.

Look at what a Free Talk session reports back: talk duration, vocabulary used, a complexity score, filler-word percentage, speaking speed and a grammar score. Set that against the marking criteria and the overlap is close to exact — vocabulary used maps to lexical resource, the grammar score to grammatical range and accuracy, and filler-word percentage plus speaking speed together approximate fluency.

That matters because the criteria are scored separately by the examiner, and an app that reports one composite number cannot tell you which band descriptor you are failing. Six independent measurements can.

The Practice tab's role-play mode is also directly useful for Part 2, and Free Talk's habit of proposing scenarios from what you said helps you cover unfamiliar topics rather than circling the ones you find comfortable.

Its limits, stated plainly: learning is mobile-only (iOS and Android), and it supports five languages — English, Spanish, German, French and Russian. For IELTS that is not a constraint, since you are practising English.

Duolingo and Babbel are poor fits for exam speaking specifically. Both are built around structured courses rather than sustained unscripted production, which is the exact skill the test measures. They are useful for general English maintenance alongside targeted practice.

An eight-week plan

Weeks 1–2: baseline and fluency. Record yourself answering Part 1 questions. Note where you stop. Practise speaking for ninety seconds without pausing, accepting errors.

Weeks 3–4: the long turn. Two Part 2 cards daily, one minute to prepare, two to speak. Aim to fill the time consistently before worrying about polish.

Weeks 5–6: Part 3 and abstraction. Practise opinion-plus-justification on unfamiliar topics. This is the hardest to self-organise and where an AI tutor that generates unpredictable follow-ups is most valuable.

Weeks 7–8: pressure and polish. Full timed rehearsals, ideally at least one with a human. Track whether your filler-word percentage is falling and complexity is holding — if complexity drops as fluency rises, you are simplifying to stay smooth, which caps your band.

Part 2 without running out of things to say

The long turn defeats candidates for a mechanical reason rather than a linguistic one: they answer the card's prompts in thirty seconds and then have ninety seconds of silence to fill.

The fix is a reusable structure rather than memorised content. Every Part 2 card can be expanded along the same four moves, whatever the topic.

Describe concretely. Specific detail generates language automatically. "A restaurant I like" produces almost nothing; describing where it is, what it looks like, what you order and who you go with produces a minute without effort.

Explain why. Reasons are inexhaustible and they naturally recruit complex structures — because, which meant, the reason being — that lift your grammatical range score.

Contrast. Compare it with something else, or with how it used to be. Comparison forces conditional and past forms, which demonstrates range without needing to force it.

Project forward. Say what you would change, or expect to happen. This brings in hypotheticals, which are exactly what the higher band descriptors look for.

Four moves, roughly thirty seconds each, and the two minutes fill themselves. Practise the structure rather than the topics, because the topics are unpredictable and the structure is not.

The mistake of using words you do not control

Candidates often prepare by learning impressive vocabulary and inserting it wherever possible. This reliably backfires.

Lexical resource rewards precision and appropriacy, not rarity. A sophisticated word used slightly wrongly is more damaging than a common word used exactly, because it demonstrates that you do not control the register. Examiners notice the mismatch immediately when a plain sentence suddenly contains a word from an academic word list.

The higher-value skill is flexibility with words you already own — knowing several ways to say the same thing, and being able to switch when the first attempt is not working. That is what paraphrase practice builds, and it serves you in every part of the test.

How to know if you are ready

Band descriptors are more useful than a general sense of confidence. If you can speak for two minutes on an unseen topic without long pauses, paraphrase around unknown words without stopping, and defend an opinion on an abstract question, you are performing at a solid level regardless of how nervous you feel.

The CEFR framework is a useful cross-check, since IELTS bands map roughly onto it, and the Europass self-assessment grid rates speaking separately from reading — which surfaces the imbalance most exam candidates have. If you are preparing for TOEFL instead, the speaking section is shorter and computer-delivered, but the underlying skill and most of this preparation transfers directly.

Related reading

Frequently asked questions