Oracy has moved from a niche enthusiasm to a named national priority. The Curriculum and Assessment Review, whose final report was published on 5 November 2025, lists oracy among the five areas that drew the most attention from young people, parents, carers and other stakeholders, alongside financial, digital and media literacy and climate education. At the same time, AI tutors that let a pupil answer out loud have arrived in schools and colleges. A curriculum lead is now being asked, often by the same supplier, whether the second can help with the first.
This guide is written for that curriculum lead, or for the education lead at a multi-academy trust who has to sign off a tool. It answers one narrow question: what can an AI tutor that accepts spoken answers contribute to oracy, and what does oracy, as the Review defines it, still need from people? It is a reading of public documents, not a claim about any pupil outcome.
The scope is deliberate. How a Socratic tutor questions before it explains is covered in our Socratic AI tutor guide. Voice answering in a pupil's home language belongs to our post on AI tutors for EAL learners. The timetable for the revised curriculum, and what it means for re-mapping a tool, sits in our adaptive learning and curriculum guide. The DfE product safety standards in full are in our AI tutor safeguarding checklist. This post takes the oracy question only.
This guide is shipped by the team behind Dilr Academy, an AI tutor that builds interactive, multilingual courses and asks before it tells. Or see DATS, our five-stage system for placing AI inside an institution's own operations.
What does the Curriculum and Assessment Review mean by oracy?
The Curriculum and Assessment Review defines oracy broadly: speaking, listening and communication, including non-written forms such as sign language, non-verbal communication and Alternative and Augmentative Communication. That definition matters for AI tutors, because a tool that turns speech into text captures only one part of it. Listening to other people, non-verbal signals and AAC sit inside the Review's definition and outside a microphone.
The Review puts it in one sentence, on page 42 of the Curriculum and Assessment Review final report:
"The Review considers oracy to incorporate speaking, listening, and communication, including verbal as well as other forms of non-written communication, such as sign language, non-verbal and Alternative and Augmentative Communication (AAC)."
Three findings in the Review explain why the Review wants more attention on it. First, in 2024 over a fifth of children did not meet the expected standards in all the early learning goals for communication and language in their Early Years Foundation Stage profile, and pupils eligible for free school meals and those with special educational needs were less likely than their peers to meet them. Second, in the same oracy section, the Review concludes that the emphasis on spoken language in the curriculum aims is not always reflected in classroom practice, and that stakeholders found the guidance on spoken language limited in scope and specificity. Third, the English chapter records that Ofsted found spoken language is not always well understood or taught effectively.
Read together, those findings describe a teaching problem, not a technology gap. The Review's remedy is a new oracy framework, more prominent speaking and listening in English and Drama, and greater specificity on oracy in Citizenship. Its oracy recommendations are about guidance, curriculum and teaching. That is the right starting point for a curriculum lead weighing an AI tutor: the tool has to fit a plan built around teaching, rather than stand in for one. Our buyer's guide to AI teachers sets out that ordering for the wider procurement.
What has the government committed to on oracy?
The government's response to the Curriculum and Assessment Review, published on 5 November 2025, commits to a new primary oracy framework and a combined secondary oracy, reading and writing framework. The primary framework is meant to help teachers make pupils confident, fluent speakers and listeners by the end of key stage 2. It is a commitment to guidance, not yet a statutory test.
The government response is specific about the shape. It says the primary framework will sit alongside the national curriculum, will be underpinned by evidence-based teaching approaches, and will give teachers guidance on how to formatively assess and support pupils' oracy. It also says the department will consider the role of non-statutory assessment for oracy in light of that work. At secondary, the combined framework is intended to help teachers in every subject connect reading, writing and talk as a whole-school strategy.
The wider timetable in the same document is to publish the final revised national curriculum by spring 2027 for first teaching from September 2028. The English and Drama programmes of study are to be revised to exemplify speaking and listening. We found no published DfE oracy framework as of this run, so any supplier claiming a tool is aligned to it is describing something that does not yet exist in final form. A sensible question back is: aligned to which document, dated when?
For a curriculum lead, the practical consequence is timing. A spoken-practice tool bought now will be used under the current 2014 programmes of study, and the new framework may arrive during its contract. The mechanics of re-mapping a tool when programmes of study change are covered in our guide to adaptive learning and the UK curriculum, and the same discipline applies here: ask each supplier how it will update when the framework lands.
What can an AI tutor with voice mode actually do for oracy?
An AI tutor with voice mode lets a pupil answer a question out loud instead of typing, and responds to what was said. Used well, that gives a pupil low-stakes rehearsal: explaining reasoning aloud, putting a method into words, or trying an answer before saying it in class. It does not give the pupil an audience of peers, which is where much of oracy actually happens.
The useful part is real. A tutor that asks a question and waits for an answer, rather than handing over a solution, creates a reason to talk. The DfE's generative AI product safety standards, updated on 19 January 2026, already expect learner-facing products to prompt learners for input before providing answers or explanations, and to show a full solution only after a genuine attempt. A pupil who has to explain a step aloud to get the next hint is doing a version of the national curriculum statement "articulate and justify answers, arguments and opinions".
Dilr Academy works this way. Its live product page states that the tutor runs Socratic questioning in voice or text, that the learner can answer by speaking or typing, and that the tutor asks, listens and explains in the language the learner answers in. Knowledge tracing then sets the next step from what the learner has shown. That is a tutoring mechanism for subject learning. It is not an oracy assessment, and Dilr Academy does not claim to measure fluency, audibility, register or any other spoken-language skill.
That distinction is the one to hold onto. Voice mode is an input channel. It changes how a pupil answers a maths or science question; it does not, by itself, make the session an oracy lesson. The ask-before-tell loop that makes it useful is explained in more depth in our Socratic tutor explainer, and the general case for what an AI tutor is for is in what Dilr Academy is.
The same diagnostic logic underpins our AI placement diagnostic, a four to six week review of where AI belongs in an institution's work and where it does not.
Which spoken language statements can an AI tutor rehearse?
The current national curriculum for English lists twelve statutory spoken language statements for years 1 to 6. Read against what a one-to-one AI tutor can do, about four can be rehearsed alone, four only partly, and four need other people. That split is Dilr.ai's reading of the statements, not a DfE classification, and it applies to primary English in England.
The English programmes of study set the twelve statements out as things pupils "should be taught to" do. The table below takes each one in order and asks a single question: can a pupil practise this with a tutor that listens and replies, with no classmate in the room?
| Statutory statement (years 1 to 6) | Rehearse alone with an AI tutor? | Why |
|---|---|---|
| Listen and respond appropriately to adults and their peers | No | Peers are part of the statement |
| Ask relevant questions to extend their understanding and knowledge | Yes | A tutor can take and answer questions |
| Use relevant strategies to build their vocabulary | Yes | Vocabulary practice works one to one |
| Articulate and justify answers, arguments and opinions | Yes | The ask-before-tell loop invites justification |
| Give well-structured descriptions, explanations and narratives for different purposes, including for expressing feelings | Partly | Explanations yes; personal feelings sit awkwardly with the DfE standards (our reading) |
| Maintain attention and participate actively in collaborative conversations, staying on topic and initiating and responding to comments | No | Collaboration needs other speakers |
| Use spoken language to develop understanding through speculating, hypothesising, imagining and exploring ideas | Yes | Thinking aloud suits rehearsal |
| Speak audibly and fluently with an increasing command of Standard English | Partly | A tool can hear speech; judging it is a teacher's call |
| Participate in discussions, presentations, performances, role play/improvisations and debates | Partly | Role play can be rehearsed; the rest needs an audience |
| Gain, maintain and monitor the interest of the listener(s) | No | A listener's interest is a human signal |
| Consider and evaluate different viewpoints, attending to and building on the contributions of others | Partly | A tutor can pose a viewpoint; it is not a classmate's |
| Select and use appropriate registers for effective communication | No | Register depends on a real audience and setting |
The pattern is consistent. The statements an AI tutor can carry are the ones about a single pupil's thinking made audible: questioning, vocabulary, justification and exploration. The statements it cannot carry are the social ones: responding to peers, collaborative talk, holding a listener's interest and choosing a register for a real audience. The non-statutory guidance on the same page makes the point directly, asking for opportunities to work in pairs, small groups, large groups and as a whole class, and for pupils to understand how to take turns.
That sequence is the honest place for voice mode: an on-ramp to classroom talk for a pupil who needs to try an answer privately first, and a way to make subject reasoning audible. It is not a replacement for the second and fourth nodes. Our guide to measuring AI tutor engagement covers how to tell whether pupils use the first node at all.
Why can a one-to-one AI tutor not teach oracy on its own?
A one-to-one AI tutor cannot teach oracy on its own because the EEF names interaction with adults and peers as a key aspect of oral language interventions, and because the DfE standards ask AI products not to imply personhood or elicit personal disclosures. Oracy, in the Review's terms, involves other people as listeners and speakers. A tutor is not one.
Start with the evidence. The Education Endowment Foundation's Teaching and Learning Toolkit rates oral language interventions as high impact for very low cost, with an average of six months' additional progress, based on extensive evidence. Three limits travel with that figure. The average is based mainly on reading comprehension tests, not on spoken language itself. The EEF says it is less confident in the oral language outcomes. And it names interaction for learning, between the practitioner and pupils and between peers, as a key aspect of the interventions. None of that is evidence about AI tutors, and the figure should not be borrowed to sell one.
The DfE standards add a principled limit. The emotional and social development section of the product safety standards expects conversation prompts to be task-bounded for learning and not to elicit personal or affective disclosures. It also expects products not to imply personhood, with an exception for time-limited, pedagogically justified role play such as role-based language practice. The fifth spoken language statement asks pupils to give narratives "including for expressing feelings". In Dilr.ai's reading, a tutor built to those standards is a poor place for a child to practise the personal side of that statement, even if a neutral narrative task stays within them.
Then there is the breadth of the Review's own definition. Sign language, non-verbal communication and AAC are inside it. A pupil who communicates through an AAC device or through sign is doing oracy in the Review's terms, and a speech-input channel offers that pupil nothing new. For those pupils the special educational needs team, not a voice feature, decides what practice looks like, just as a school's EAL lead decides how home-language answering is used.
None of this makes voice mode pointless. It makes it narrow. The standards themselves are aimed mainly at edtech developers and suppliers, and say schools and colleges may find them helpful when assessing products; our safeguarding checklist walks through every section a trust should ask about.
What should a curriculum lead test before using voice mode for speaking practice?
A curriculum lead should test five things before using an AI tutor's voice mode for speaking practice: what it does with a pupil's recording, whether it hears the pupils it will actually serve, whether it stays task-bounded, how a teacher sees what was said, and whether the supplier's claims rest on evidence. Each can be checked in a short pilot with a teacher present.
1. The recording. A child's voice is personal data. Ask where audio goes, whether it is stored or only transcribed, for how long, and who can access it. The ICO's Children's code guidance is the reference point for online services likely to be accessed by children, and the school's own data protection officer should see the answers before pupils do.
2. Your pupils' voices. Do not accept a supplier's general accuracy figure. Run the tool with a small group that reflects the cohort, including younger pupils, regional accents and pupils who speak English as an additional language, and have a teacher compare what was said with what the tool heard. Where it mishears, a pupil learns that speaking does not work, which is the opposite of the aim.
3. Task boundaries. Try prompts that drift: a pupil talking about their weekend, or about how they feel. The DfE standards expect conversation prompts to stay task-bounded for learning and not to engage learners in personal or emotionally sensitive topics. Note what it actually does.
4. Teacher visibility. In our view the judgement of a pupil's speaking belongs with a teacher. Check that a teacher can see a transcript or summary of a pupil's spoken answers, so that solo rehearsal feeds the teacher's view rather than replacing it.
5. The claims. The DfE standards say suppliers should not exaggerate the impact or capabilities of their tools and that claims should be supported by robust and transparent evidence. Ask for that evidence on any oracy claim. A supplier saying its tool "develops oracy" should be able to say which of the twelve statements it means and how it knows.
The wider governance around a pilot like this, from ownership to review points, is what our AI operating model work sets up for institutions running several AI tools at once. If the pilot raises questions about whether AI belongs in this part of the timetable at all, talk to us before buying anything.
What is the best AI tutor for oracy practice in 2026?
There is no best AI tutor for oracy in 2026, because the oracy approaches the EEF summarises centre on interaction with adults and peers, which no AI tutor provides. The better question is which tutor suits spoken rehearsal of subject reasoning. Judge that on four criteria: it asks before it tells, accepts spoken answers, stays task-bounded, and shows teachers what pupils said.
Two of those criteria come from the DfE standards: asking before telling is the cognitive development expectation, and task-bounded behaviour is the emotional and social development expectation. The other two are Dilr.ai's own: spoken answers are the input that makes rehearsal possible, and teacher visibility keeps the judgement of a pupil's spoken language with a teacher rather than a tool.
On those criteria, the concessions are clear. For oracy as the Review defines it, the approach the EEF Toolkit summarises is not a product at all: structured talk with peers and a teacher. For an individual learner rather than an institution, Dilr Academy's own product page says it is not sold to individual learners today and names Khanmigo as a sensible place to start for someone learning on their own in the United States; our Khanmigo alternative comparison covers where each fits. For a school whose priority is a specific spoken-language programme, an oracy programme run by teachers is the more direct route than a general tutor, and it should be checked against the framework once it is published.
Dilr Academy fits a narrower case: a college, school or training provider that wants its own courses taught one to one, with questions a learner can answer by speaking or typing, as part of subject teaching. That is a real use, and a smaller one than oracy. The category overview for AI strategy posts and our education hub set it against the other places AI pays in education.
Where does Dilr Academy fit in a school's oracy plan?
Dilr Academy fits a school's oracy plan as a rehearsal step inside subject courses, not as the oracy programme itself. Its tutor asks a question first, accepts an answer by speaking or typing, and moves on once knowledge tracing shows the idea has landed. That makes reasoning audible in maths or science, while discussion, debate and feedback on spoken language stay with teachers.
The courses Dilr Academy has built so far are shown on its product page as examples rather than a catalogue. The longest follows the England KS3 programme of study for Year 9 Mathematics, and others cover professional subjects such as Python and generative AI. Academy is licensed to universities, colleges and training providers for their own cohorts, so any oracy-adjacent use is a decision the institution makes about its own curriculum, not a feature the tool switches on.
In practice that means three things for a curriculum lead. Put voice answering where reasoning aloud helps subject learning. Plan the peer and whole-class talk separately, against the national curriculum now and the oracy framework once it is published. And keep the assessment of speaking and listening with teachers, because Dilr Academy does not assess it. Our DATS methodology is how we help an institution make placement decisions like that one across all of its AI, and our approach explains why we start from the work rather than the tool.
Frequently asked questions
Does an AI tutor's voice mode count as oracy teaching?
An AI tutor's voice mode does not count as oracy teaching on its own. The Curriculum and Assessment Review defines oracy as speaking, listening and communication with others, including non-verbal forms and AAC. Voice mode lets one pupil answer a tutor aloud, which supports rehearsal and explaining reasoning, but the listening, collaboration and audience parts of oracy still need classmates and a teacher.
A useful test is to ask which of the twelve national curriculum spoken language statements a session served. If the answer is only "articulate and justify answers", that is valuable subject practice, and it should be recorded as such rather than as oracy coverage.
Will the new oracy framework include a test?
The government's response to the Curriculum and Assessment Review does not commit to an oracy test. Published on 5 November 2025, it says the department will consider the role of non-statutory assessment for oracy in light of developing the framework, and that the framework will give teachers guidance on formatively assessing and supporting pupils' oracy. A statutory test is not part of the commitment.
Until the framework is published, treat any product described as "oracy assessed" with caution, and ask what standard it assesses against. For a broader view of how to judge AI tutors before buying, our AI teacher buyer's guide sets out the criteria we use.
Planning AI around the new curriculum? Compare AI education tools for 2026, book an AI placement diagnostic, read our AI teacher buyer's guide, or see our approach to placing AI inside an institution's own teaching.
Put AI where talk and teaching both win.
30-min scoping call · No deck · Confidential. We will tell you where an AI tutor helps your courses, and where teachers should keep the work.
Written by the Dilr.ai engineering team, practitioners who ship enterprise AI in production. Follow us on LinkedIn for shipping notes, or subscribe via the RSS feed.