How a session works
- Pick a card for your profession: setting, your role and four tasks, like the real card (90 cards: 20 nursing, 20 medicine, 5 dentistry, 5 pharmacy, 5 physiotherapy, 5 occupational therapy, 5 dietetics, 5 optometry, 5 podiatry, 5 radiography, 5 speech pathology, 5 veterinary science).
- Prepare for 3 minutes, then speak for 5. The patient has a personality and a worry they only share if you ask the right way.
- Talk naturally. Interrupt, check understanding, slow down. The patient reacts to what you actually say.
- Get your marks: an estimated band for each of the 9 criteria, the lines from your transcript that earned or lost marks, and the three things to fix first. Then say it again.
What gets marked
| Criterion | Scale | How we estimate it |
|---|---|---|
| Intelligibility | 0–6 | Pronunciation assessment of your recording |
| Fluency | 0–6 | Speech rate and pauses measured from your recording |
| Appropriateness of Language | 0–6 | Your transcript: register, jargon explained, tone |
| Resources of Grammar and Expression | 0–6 | Your transcript: range and accuracy |
| Relationship Building | 0–3 | Your transcript: greeting, empathy, respect |
| Patient's perspective | 0–3 | Did you draw out and respond to their worry? |
| Providing Structure | 0–3 | Signposting and logical order |
| Information Gathering | 0–3 | Open questions, listening, summarising |
| Information Giving | 0–3 | Chunks, checking understanding |
When a criterion can't be judged from the evidence, it says “not assessed” instead of guessing. How accurate are we?
What to look for in any AI Speaking tool
- It talks back out loud and lets you interrupt; typing to a chatbot doesn't train speaking.
- It marks the 9 official criteria, not a single made-up score.
- Pronunciation comes from your audio. A tool that only reads a transcript can't hear you.
- It shows the lines it judged, so you can check it.
- Cards match your profession: a pharmacist talks to the people a pharmacist would.
Questions candidates ask
Is there an AI tool to practise OET Speaking?
Yes. On OETAce you pick a role-play card for your profession, prepare for 3 minutes, then speak for 5 minutes with an AI patient or carer who answers out loud. Afterwards you get an estimate for each official criterion with quotes from what you said. Your first role-play is free.
Can AI mark OET pronunciation?
Only if it listens to your audio. We run a pronunciation assessment on your recording for Intelligibility and Fluency. If too little of your speech can be assessed, those criteria are shown as not assessed instead of being guessed from text.
How is OET Speaking marked?
Each role-play is marked on 4 linguistic criteria (Intelligibility, Fluency, Appropriateness of Language, Resources of Grammar and Expression, 0–6 each) and 5 clinical communication criteria (Relationship Building, Understanding and Incorporating the Patient's Perspective, Providing Structure, Information Gathering, Information Giving, 0–3 each).
Is an AI Speaking score the same as the real OET score?
No. It is an estimate from a practice tool, shown as a range. Only OET's trained assessors give official scores. We publish how our estimates compare with real results candidates report on the accuracy page.
Can I interrupt the AI patient?
Yes. When you start talking, the patient stops, as a real patient would. Turns usually come back in about 2–4 seconds.
Sources
Last checked 22 Oct 2026. Rules and prices change: confirm on the official page before you book or apply.
OETAce is an independent practice site. OET is a registered trademark of Cambridge Boxhill Language Assessment (CBLA); we are not affiliated with or endorsed by CBLA. All practice material here is original.