Two new speaking tasks, and a human now checks seven question types for memorised answers. Most guides on this are either a thin blog post or a YouTube video. This page is the written reference we keep updated as Pearson changes anything else.
The old 20 question types are unchanged — same format, same timing, same response type expected. Nothing you already knew about Read Aloud, Repeat Sentence, Write Essay's structure, and so on stopped applying. What's new sits on top of the existing test, not instead of it.
Both are designed to be difficult to prepare for with a memorised script — that is the point of adding them.
You hear a brief, everyday scenario — leaving a voicemail, apologising for a miscommunication, giving directions — and respond by voice, appropriately, on the spot.
It tests whether you can pick the right tone, form and vocabulary for a specific social situation, rather than reciting prepared material. There is no fixed template that works across different scenarios, because the scenarios themselves vary in what an appropriate response actually sounds like.
You listen to a short exchange between multiple speakers, then give a spoken summary of the main points and the different perspectives raised.
This tests critical listening across more than one voice, plus your ability to condense several viewpoints into a clear, concise spoken summary. It is closer to a real meeting or seminar than any previous PTE speaking task.
Older PTE prep culture leaned heavily on templates: fixed opening phrases, fixed structural signposting, recycled across almost any prompt. That approach works reasonably well on tasks with a predictable shape, like Describe Image.
It fails on Respond to a Situation because the appropriate tone and content genuinely differ between "leave a voicemail" and "apologise for a delay" — a one-size template sounds obviously wrong on at least one of them, which is exactly the kind of mismatch a human reviewer is trained to catch.
This is the change causing the most anxiety, and the one explained worst elsewhere.
| Section | Question type | Human-reviewed? |
|---|---|---|
| Speaking | Describe Image | Yes |
| Speaking | Retell Lecture | Yes |
| Speaking | Summarize Group Discussion (new) | Yes |
| Speaking | Respond to a Situation (new) | Yes |
| Writing | Summarize Written Text | Yes |
| Writing | Write Essay | Yes |
| Listening / Writing | Summarize Spoken Text | Yes |
| All other types | Read Aloud, Repeat Sentence, Fill in the Blanks, MCQs, and the remaining item types | No — AI only |
All seven human-reviewed tasks are also still AI-scored first. The human layer is an additional check specifically for content — did this response sound like a genuine, original answer to this prompt, or does it read as memorised and templated? Pronunciation and fluency are never scored by a human, in any task, including these seven — that part stays fully automated.
The reviewer is not re-marking your response on the 0–90 scale the way the AI does. They are moving through a queue quickly and making one binary call: does this sound like a real, spontaneous answer, or does it sound scripted?
If your response is judged templated, the score on that specific item gets capped — regardless of how fluent, well-pronounced or grammatically correct it otherwise was. This is a content judgement layered on top of the existing AI marks, not a full independent rescoring of the whole response.
The difference is not complexity or vocabulary — it's whether the answer actually responds to what's in front of you.
Word-for-word repetition of the same opening line and structural phrases across every attempt, regardless of what the image, lecture or scenario actually contains. The content is generic enough to plausibly answer several different prompts.
A consistent underlying approach — the same procedural skeleton each time — but the actual wording, details and phrasing change to match what is genuinely in the prompt. Specific, prompt-relevant content that could not be swapped into a different question.
Pearson added two new speaking question types — Respond to a Situation and Summarize Group Discussion — taking the total from 20 to 22 item types, with total test duration increasing by roughly 10 to 15 minutes. The three-section structure (Speaking and Writing, Reading, Listening) is unchanged. Pearson also introduced human review alongside AI scoring for seven question types specifically to catch templated or memorised responses. The original 20 question types kept their existing format and timing.
A new speaking task where you hear a brief everyday scenario — such as leaving a voicemail, apologising for a miscommunication, or giving directions — and respond appropriately by voice. It tests whether you use suitable tone, form and vocabulary for the specific situation, rather than reciting a memorised structure, which is why a single template does not work well across different scenario types.
A new speaking task where you listen to a short exchange between multiple speakers and then give a spoken summary of the main points and different perspectives raised. It assesses critical listening across more than one voice and your ability to communicate understanding of a multi-speaker conversation clearly and concisely.
Seven types: Describe Image, Retell Lecture, Summarize Group Discussion and Respond to a Situation from Speaking; Summarize Written Text and Essay from Writing; and Summarize Spoken Text. All other question types — including Read Aloud, Repeat Sentence and the multiple-choice items — remain AI-scored only. Pronunciation and fluency are always machine-scored, even within the seven reviewed types; a human never marks those two components.
If a human reviewer judges your response to sound memorised or templated rather than original, the score on that item gets capped, regardless of how fluent, well-pronounced or grammatically correct it otherwise was. The reviewer is making a fast binary content judgement, not fully re-marking the response — so the risk is specifically about sounding scripted rather than about making language errors.
Keep a consistent procedural approach across attempts — for example, the same order of moves for Describe Image — but change the actual wording and specific content each time to genuinely match what's in front of you. Word-for-word repetition of memorised phrases across different prompts is the clearest trigger. Practising reactions to new, unseen prompts rather than rehearsing fixed full-sentence scripts is the most reliable way to prepare for these tasks now.
It changed rather than uniformly got harder. Two new speaking tasks were added and total time increased by 10 to 15 minutes, but the overall scoring scale stayed at 0–90 and the original 20 question types kept their existing format. The main practical shift is that formulaic, memorised-template preparation — a common shortcut before August 2025 — is now actively penalised on seven question types. Candidates preparing with genuine content practice rather than fixed scripts are less affected by the change than those relying on templates.
This is a living reference. Pearson periodically adjusts PTE format and scoring details, and we update this page when that happens rather than leaving it to go stale — that gap in the market, a maintained written breakdown, is exactly why we built it. If you spot something here that no longer matches the live test, tell us and we'll verify and correct it. For the current, complete official specification, always cross-check pearsonpte.com directly.
Our PTE Masterclass covers all 22 current question types, including the two new speaking tasks, with practice built around genuine content rather than fixed scripts.