More than one pupil in five in England now carries an English as an additional language (EAL) flag. The Department for Education's school census for 2025/26 records 21.6% of pupils with a first language known or believed to be other than English, up 0.2 percentage points on the year before. For a trust education lead weighing a multilingual AI tutor, that number looks like a ready-made target cohort. It is not, and the reason sits in the census definition itself.
An AI tutor that can teach a whole course in a pupil's home language is a newer option for schools. Used well, it gives a recently arrived pupil access to Year 9 maths while their English catches up. Used badly, it hands a fluent bilingual pupil a crutch they never needed, or quietly keeps a new arrival out of English for longer than the school intends. The difference is a decision the school makes about each learner, not a feature the tutor ships with.
This guide is written for a multi-academy trust or college curriculum lead. It covers what the EAL flag measures, what the national attainment data says, how to decide which learners benefit from home-language tutoring, what an AI tutor can and cannot do for them, and how to pilot one. It deliberately cedes three neighbouring questions. Adult ESOL and the funding rules around it belong to our guide to AI tutors for adult essential skills. Why a tutor should ask before it tells belongs to our Socratic AI tutor explainer. Head-to-head product comparisons belong to our Khanmigo and Third Space Learning reviews, linked further down.
This guide is shipped by the team behind Dilr Academy, an AI tutor that builds interactive, multilingual courses from your curriculum, including right-to-left scripts. Or see DATS, our senior-led consulting practice for placing AI inside an institution's own operations.
What does the EAL flag in the school census actually measure?
The EAL flag in the DfE school census records exposure, not ability. A pupil is recorded as having English as an additional language if they are exposed at home to a language known or believed to be other than English. The flag says nothing about how well that pupil speaks, reads or writes English, so a fluent bilingual sixteen-year-old and a pupil who arrived last month can carry the same marker.
The Department for Education is explicit about this limit. Its 2025/26 census release states: "This measure is not a measure of English language proficiency or a good proxy for recent immigration." The same release shows how unevenly the flag is spread: from 7.1% of pupils in alternative provision to 30.6% in nurseries, with primary schools at 23.8% and secondary schools at 19.3%. The overall count of pupils of all ages across all schools in England was 8,920,227.
The DfE's key stage 4 performance release for 2024/25 adds the other half of the caveat. It says that being a pupil whose first language is other than English does not mean the pupil is necessarily fluent in another language, or that they cannot speak English. A trust that buys a multilingual tutor "for its EAL pupils" has, on the department's own terms, bought it for a group defined by the language heard at home. That group contains pupils who need home-language support alongside pupils who do not.
Are EAL pupils behind, and why does the answer change the AI tutor decision?
Nationally, pupils with a first language other than English are not behind as a group. In the DfE key stage 2 results for 2025, 64% of them met the expected standard in reading, writing and maths combined, against 62% of pupils with English as their first language. Because the flag records exposure rather than proficiency, it cannot by itself identify the learners an AI tutor should serve.
The DfE key stage 2 attainment release for 2024/25 gives the subject detail. Pupils with a first language other than English, who were 23% of the end of key stage 2 cohort in 2025, led on maths by 5 percentage points (78% against 73%) and on writing by 1 point (73% against 72%). Reading is the one subject where they trailed, at 74% against 76%, a gap the department notes has narrowed to 2 percentage points.
The pattern holds at sixteen. In 2024/25, 19.5% of pupils at the end of key stage 4 in state-funded schools had a first language other than English, up from 16.7% in 2018/19. The DfE reports that EAL pupils had better attainment than non-EAL pupils across the headline measures in 2018/19, 2023/24 and 2024/25. The gap in grade 5 and above in English and maths narrowed to 3.3 percentage points, from 4.8 the year before, driven by a fall in EAL pupils' own attainment.
None of this means every EAL pupil is fine. It means the national averages, together with the DfE's own definition, give a school no way to pick out the pupils who need language support from the flag alone. A trust reading these figures should conclude that the flag is the wrong unit of decision. The right unit is the individual learner's proficiency in English and their literacy in their home language, which the census does not record and the school has to assess itself. Our AI placement diagnostic starts from the same principle in any institution: find the decision the data actually supports before buying the tool.
How should a school decide which EAL learners need home-language support?
A school decides by assessing two things the census flag does not capture: the learner's proficiency in English and their literacy in their home language. A pupil new to English who reads fluently in their first language is the clearest case for home-language tutoring in subject content. A pupil who is already competent in English, or who has little first-language literacy, usually is not.
For the first measure, most English schools do not need to invent anything. The Bell Foundation's EAL Assessment Framework for Schools and its digital tracker are free to every school in the UK, and assess proficiency across listening, speaking, reading and writing. The framework was built with assessment experts at King's College London and the University of Cambridge. It uses five levels of proficiency in English: New to English, Early Acquisition, Developing Competence, Competent and Fluent. The foundation describes New to English learners as working at band A and, as a general rule, in their first two years of learning English.
The second measure is the one a multilingual tutor depends on, and it is easy to skip. The Bell Foundation's guidance on home language assessment notes that many EAL learners have good home-language skills, but that first-language literacy varies. Age of entry matters: a pupil who starts school in England at five will only just be developing literacy in their first language, while one who arrives at fifteen may be highly literate in it. A tutor that teaches algebra in written Arabic helps the second pupil and may not help the first at all.
In practice this produces three rough groups. Fluent and competent bilingual pupils learn subject content in English like their peers. Pupils new to English with strong first-language literacy can learn subject content in their home language while English is taught separately and deliberately. Pupils new to English without first-language literacy need English, oral language work and specialist support first, and a written home-language course is a poor fit for them. The school's EAL lead owns these calls; the AI operating model question is simply who records them and when they are reviewed.
When does teaching in the home language help, and when does it delay English?
Teaching in the home language helps when the goal is access to subject content and the learner can already read and reason in that language. It risks delaying English when the goal is English itself, or when home-language teaching replaces, rather than sits beside, deliberate English teaching. An AI tutor makes the first easy and the second easy to drift into, so the school has to name the goal for each learner.
Consider a Year 9 pupil who arrived from abroad in September, reads well in their first language and studied algebra at home. In English they cannot yet follow a maths lesson. In their home language they can work through simultaneous equations at the level they had reached. A home-language course keeps their mathematics moving while the school's EAL programme builds their English. Without it, a capable pupil can spend a year relearning work they had already mastered because the explanation is the barrier, not the maths. Our AI tutor engagement guide looks at why learners disengage when a tool does not meet them where they are.
Now consider a pupil who has been in English schools since reception, speaks a language other than English at home and reads mostly in English. The census still flags them. A tutor set to their home language would teach them in the language they read less well, which is the opposite of help. And for a pupil whose actual goal this term is English, such as a new arrival at the Early Acquisition level, an always-on home-language tutor can become the comfortable route that crowds out the harder practice.
Our adult essential skills guide sets out the same principle for adult ESOL learners, where first-language support is a scaffold the tutor team chooses rather than a default. At school altitude the decision is usually made per subject, not per pupil: home language for maths or science content, English for English, with the balance shifting as the pupil moves up the proficiency levels. That shift is the thing to plan and review, because no tutor will make it on its own.
What can an AI tutor that follows the learner's language actually do?
An AI tutor that follows the learner's language can teach a complete course, with explanations, questions, feedback and diagrams, in that language. Dilr Academy rebuilds the whole course in the learner's language, including right-to-left scripts, and its tutor asks, listens and explains in the language the learner answers in. Once a course is built in that language, a new arrival with strong home literacy can study subject content they could not yet follow in English.
The Dilr Academy live page is specific about what follows the language. It is not just the menu labels: the syllabus and stated outcomes, the Socratic questions and feedback, and the diagrams and interactive elements are all written in the course language. The tutor works in voice or text, so a pupil can answer by speaking or typing, and it asks before it tells. Knowledge tracing tunes each step to the level the learner has actually reached, which matters for a pupil whose prior schooling did not map neatly onto the English curriculum.
The courses themselves are real syllabuses, as our explainer on what Dilr Academy is sets out. The live page shows built courses ranging from a five-hour topic to a 180-hour school year, with Year 9 Mathematics for England at key stage 3 as one example. Those courses are shown as examples of what Academy builds rather than a ready catalogue: institutions license it for their own cohorts, and courses are built from the institution's curriculum in the languages its cohort needs. For the arriving Year 9 pupil above, that is the scale that matters: a year of maths content, taught one to one, in a language they can already reason in, once the school has scoped that course and language.
Two design questions matter for any tutor serving EAL pupils, not only ours. The first is whether it does the thinking for the pupil. The DfE's generative AI product safety standards, updated on 19 January 2026, expect products not to give final answers or full solutions by default. Those standards are written for edtech developers and suppliers, though schools may use them when assessing products. A translated full answer is still a full answer, so a home-language tutor needs the same ask-first discipline our Socratic tutor explainer describes. The second is whether the school, not the tutor, controls which language each course runs in.
Where does an AI tutor stop for EAL learners?
An AI tutor stops at four boundaries for EAL learners. It does not assess English proficiency, so it cannot tell a school which pupils need home-language support. It is not an English acquisition programme. It does not replace the school's EAL lead, teaching assistants or the relationships that help a new arrival settle. And it does not decide safeguarding or wellbeing questions, which stay with named staff in the school or trust.
The first boundary is the most important for procurement. Dilr Academy follows the language the course is built in and the language the learner answers in; its live page describes no assessment that places a pupil on the Bell Foundation's proficiency levels or any other scale. The assessment that decides who gets a home-language course is the school's job, done by staff who know the pupil. A tutor that claimed to make that call would be overreaching. Our AI teacher buyer's guide covers the wider questions to put to any vendor before buying.
The second boundary is about scope. A subject course taught in Polish or Urdu does not teach English, and it should not be counted as EAL provision in a school's plan. English acquisition needs structured teaching, speaking and listening with people, and feedback from staff trained in it. The Dilr Academy live page makes no claim to EAL outcomes, and a trust should be wary of any tutor that does without published evidence behind it. Our DATS consulting practice works with institutions on what to ask before they buy an AI system.
The third and fourth boundaries are about people. Many arriving pupils carry experiences a school needs to understand through its own staff, families and interpreters. Our AI safeguarding guide for schools and trusts covers the duties that stay with the school whatever tutor it uses. The same principle underpins our AI execution office, the supervised model for running AI systems inside an organisation once they are live: the tool does the repeatable work and named people own the judgement.
How should a trust pilot a multilingual AI tutor with EAL learners?
A trust should pilot a multilingual AI tutor with a small, assessed group rather than the whole EAL cohort. Start with new arrivals who have strong home-language literacy, set the course language subject by subject, run the tutor alongside the existing EAL programme, and review each pupil's English proficiency and subject progress at the end of a term before widening the group.
A workable sequence for one secondary school looks like this:
- Assess before you assign. Place each candidate pupil on the Bell Foundation framework or the school's own equivalent, and record whether they read and write fluently in their home language.
- Pick one subject. In our view maths is a sensible first subject: prior learning carries across languages, and it leans less on extended prose than history or English literature.
- Set the language deliberately. Home language for the subject course, English for everything else, with the EAL lead deciding which pupils take the home-language course.
- Keep English teaching intact. The tutor sits beside the EAL programme, never in place of a session, which reflects our approach of placing AI beside people rather than instead of them.
- Review at a fixed point. At the end of the first term, look at the pupil's proficiency level, their subject progress as the class teacher sees it, and the pupil's own view. Move pupils towards English for the subject as their proficiency rises.
The trust measures what it can observe in its own data, not what a vendor promises. It should also record the decision for every pupil, because the hardest question in a review is usually why a given pupil was or was not included. Our AI teacher buyer's guide covers the wider pilot and procurement steps, and the education hub maps where AI pays across schools, colleges and universities.
What is the best AI tutor for EAL pupils in 2026?
The best AI tutor for EAL pupils in 2026 is the one that teaches a whole subject course in the learner's language, including right-to-left scripts, asks before it tells, leaves the decision about who learns in which language with the school, and makes no unsupported claims about English acquisition. Dilr Academy is built for the first three. No general AI tutor, Dilr Academy included, replaces a school's EAL programme.
Four criteria separate a useful home-language tutor from a translated chatbot:
- Depth of translation. Does the whole course follow the language, including questions, feedback and diagrams, or only the interface?
- Script coverage. Does it handle right-to-left scripts such as Arabic and Urdu properly, or only left-to-right languages?
- Teaching discipline. Does it ask before it tells in every language, in line with the DfE's expectations, or does translation become a shortcut to the answer?
- School control. Ask any vendor, Dilr Academy included, how staff choose the course language for a learner and what they can see of each pupil's progress. That is the governance question our AI operating model work is built around.
There are scenarios where Dilr Academy is not the right choice. A school whose real need is English acquisition for new arrivals will get more from specialist EAL staff and the Bell Foundation's free resources than from any AI tutor. A school already running a maths intervention it trusts may have no reason to switch for language alone; our Third Space Learning comparison looks at that case, and our Khanmigo alternative review covers schools comparing Khanmigo. For a wider shortlist, see the best AI education tools of 2026 comparison.
Frequently asked questions
Does Dilr Academy teach English to new arrivals?
Dilr Academy is not an English acquisition programme for new arrivals, and it makes no claim to improve English proficiency. It teaches subject courses, and it can teach them in the learner's home language, including right-to-left scripts. A school should keep its EAL programme for English and use a home-language course for subject access while that English develops.
Does the EAL flag tell a school which pupils need home-language support?
The EAL flag does not tell a school which pupils need home-language support. The DfE census records whether a pupil is exposed at home to a language other than English, and the department states it is not a measure of English proficiency. A school needs its own proficiency assessment and a check on home-language literacy before assigning any pupil to a home-language course.
Can a pupil answer in English inside a course taught in their home language?
The Dilr Academy live page states that the tutor asks, listens and explains in the same language the learner answers in, and that a learner can answer by speaking or typing. How a school uses that with a bilingual pupil is a teaching decision, so the EAL lead should agree the course language and the expected answer language before the pupil starts.
For anything the live page does not answer, ask the Dilr Academy team directly.
Where to go next
Want to see a multilingual course in practice? Read what Dilr Academy is, book an AI placement diagnostic, or see our approach to placing AI where an institution can govern it. More education posts sit in our industries collection.
Give new arrivals the subject, not the wait.
30-min scoping call · No deck · Confidential. We will tell you whether a multilingual tutor fits your EAL plan, and where it does not.
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.