This week the Department for Education published updated guidance on using AI in education settings. One section in particular landed differently for those of us working in SEND: an explicit statement that teachers, teaching assistants, and SENDCos can use AI to draft initial versions of support plans, create staff guides for specific needs, and generate communications to parents.
I have spent the last year building an AI-powered inclusion platform alongside my day job as an Inclusion Teacher. So I have thoughts on what this guidance actually means in practice — and some honest cautions about what it does not mean — for SENDCos reading this on a Tuesday evening with thirty plans still to write before the end of term.
What it says
The guidance — developed by Chiltern Learning Trust with the Chartered College of Teaching and the DfE — is specific rather than vague. It names support plans. It names staff guides. It names parent communications. These are not peripheral tasks. They are the things that keep SENDCos at their desks until seven o'clock.
The guidance also makes a point that anyone who has worked in inclusion will recognise immediately: the administrative burden of plan writing is not where our professional value lies. Our value is in knowing the child. In the review meeting. In the conversation with the parent who is frightened for their child's future. In the professional judgment that no algorithm can replicate.
The DfE is saying, carefully, that AI can take on some of the paperwork so that we can spend more time on the work that actually requires us. That is a reasonable position. It is also, frankly, overdue.
What it does not say
This matters more than what it does say, so I want to be clear.
The guidance does not say AI should decide whether a child has a special educational need. It does not say AI should write EHCPs or contribute to statutory documents without significant professional review. It does not say any of us should hand a plan to a family that we have not read, considered, and made our own.
AI does not know your pupil. It does not know that a particular child presents completely differently on a Monday after a difficult weekend at home. It does not know that the strategy that works for most children with a similar profile is actively counterproductive for this child because of something that only the family knows. It does not carry the professional and legal responsibility that sits with the SENDCo who signs the plan.
What it can do — when it is built properly — is produce a well-structured starting point grounded in relevant evidence. The SENDCo then does what only the SENDCo can do: reads it critically, changes what needs changing, adds what only they know, and signs off on something that genuinely reflects their professional judgment.
That is the distinction the guidance draws. It is the right distinction.
The data protection question
This is the question I know most SENDCos will ask first, because it is the right question to ask.
Pasting a child's name, diagnosis, and confidential learning profile into a public AI chatbot is not appropriate. The guidance says this clearly and it is correct. General-purpose AI tools are not built for sensitive data about children. They do not have the data protection architecture, the GDPR compliance frameworks, or the sub-processor agreements that this kind of work requires.
But this is not an argument against AI in SEND planning. It is an argument for using tools built specifically for this purpose — where the pupil data stays within a secure, controlled environment, where the AI processes that data without it being exposed to external systems, and where the Data Processing Agreement is something you can actually read and understand before you sign up.
There is a significant difference between pasting a child's details into ChatGPT and using a purpose-built, GDPR-compliant platform designed specifically for inclusion. Any SENDCo evaluating tools in this space should know which one they are using before they start.
What this looks like in practice
Here is what AI-assisted plan drafting actually looks like when it is done well, based on a year of building and using it.
It starts with the pupil, not the technology. The AI works from the information you give it — the child's strengths, their needs, their assessment data, what they say about themselves, what their parents say. The quality of the draft is entirely dependent on the quality of that information. If the profile is thin, the draft will be thin.
The draft is a starting point. It has structure. It has targets that make sense for the need type. It has strategies that are grounded in the evidence base rather than invented from thin air. And then the SENDCo reads it, amends it, removes what does not fit this specific child, and adds the detail that no AI could know. The SENDCo writes the plan. The AI drafted a scaffold.
Every suggestion should be traceable. When an AI proposes a strategy or a target, a well-built tool should be able to show you where that comes from — which evidence source, which research framework, which professional guidance. If the AI is producing plausible-sounding text with no grounding in the actual evidence base, that is a problem. Not all tools in this space are built to that standard.
Professional review is not optional. No AI-generated plan should go to a family without being read carefully by a qualified practitioner. Not skimmed. Read. The professional review is where accountability sits and where the plan becomes genuinely useful to the child rather than a well-formatted document that says the right words without meaning them.
How we have built OMNIA
OMNIA was designed from the start on the assumption that the SENDCo is the expert and the AI is the assistant. The platform drafts. The SENDCo reviews, edits, signs off. Every plan carries an evidence trail showing which research base the strategies are drawn from — EEF, nasen, the British Dyslexia Association, the PDA Society — so what arrives in front of you is not generated text, it is a draft grounded in the same evidence base your professional training rests on.
That review step is not a formality. It is where accountability sits and where the plan becomes genuinely useful to the child rather than a well-formatted document that says the right words without meaning them.
A note on EAL
The guidance also addresses AI for EAL pupils — translating materials, tailoring support to English proficiency levels, and communicating with multilingual families. Worth noting separately because in many schools, particularly international ones, EAL and SEND overlap significantly and frequently.
The guidance is right that disentangling a language acquisition barrier from a genuine additional learning need is one of the most complex professional judgments a SENDCo makes. AI can help with translation and communication. It cannot make that clinical distinction. That stays with the practitioner.
The practical questions worth asking
If you are a SENDCo reading this and thinking about what to do next, here are the questions that actually matter before you adopt any tool in this space.
Where does the pupil data go? Can the provider give you a Data Processing Agreement that you can understand and that your data protection officer is comfortable with?
Is the AI grounded in evidence or is it generating plausible text? There is a difference between an AI that suggests strategies based on the EEF guidance and nasen resources and one that produces well-formatted sentences that sound professional but have no evidential basis.
Does the tool understand context? Does it know the difference between an EHCP and a DLP? Does it know what an IDP is, or a CSP? Does it understand that the strategies appropriate for a pupil with a PDA profile are fundamentally different from those appropriate for a pupil with a broadly autistic profile? If the tool treats all additional needs as interchangeable, it will produce plans that look right and are wrong.
The DfE guidance is a green light for AI-assisted SEND planning done thoughtfully and done properly. It is not a green light for AI-assisted SEND planning done carelessly. The responsibility to know the difference has always sat with us. It still does.