Most artificial intelligence (AI) writing tools approach a special educational needs (SEN) Support Plan as a formatting problem. Give the model a few pupil details and a heading structure, and it will produce something that reads like a support plan. The problem is that nobody can tell from the output whether the content came from recorded evidence or from the model's sense of what a support plan usually says.
We took a slower route. We studied special educational needs and disabilities (SEND) document examples from local authorities across England, worked with practising special educational needs coordinators (SENCOs) to build synthetic test cases, and used both to develop a system for UK SEND documents. It is designed around UK SEND structure and source material, with staff deciding what the record finally says.
This is the development story, including what we can and cannot claim about the results.
Why one generic template is not enough
Ask five special educational needs coordinators (SENCOs) in five different local authority areas to share their SEN Support Plan and you will get five different documents.
One separates strengths, needs, outcomes, and provision into distinct sections. Another folds most of the useful detail into a single review table. A third leads with pupil voice. A fourth uses graduated response language throughout. Special schools working with profound and multiple learning difficulties (PMLD) pupils need something different again, because a plan built around curriculum targets does not describe a pupil communicating through eye gaze and physical response.
The SEND Code of Practice gives every school the same statutory framework and the same Assess, Plan, Do, Review cycle. It does not standardise the paperwork. That is deliberate, and it is not a flaw. Local variation reflects local practice.
For anyone building software, though, it means a single template is a design failure waiting to happen. Force every setting into one form and you get plans that fit nobody, headings that make no sense to the staff reading them, and content quietly dropped because there was nowhere to put it.
So we started by collecting the variation rather than flattening it.
Who actually makes an education, health and care (EHC) plan?
There is an important distinction before we go any further. In England, an Education, Health and Care Plan is a statutory plan prepared and maintained by the local authority. Section 37 of the Children and Families Act 2014 says that where an EHC needs assessment shows that special educational provision is necessary through an EHC plan, the local authority must secure that the plan is prepared and must maintain it once prepared.
Schools do not issue the EHCP. They contribute the evidence, advice, pupil and family views, provision information and review material that help the local authority make and maintain the plan. When a child or young person moves between local authority areas, the transfer and maintenance duties remain matters for the authorities involved, as the Department for Education's move guidance explains.
That means the useful school-side software question is not, "Can this platform make an EHCP for us?" It is, "Can it help us collect, structure, check and hand over the evidence that an EHC needs assessment, review or transfer depends on?" MeritDocs is built around that evidence and document workflow. It helps schools work with their current SEND records and supported document types, while the statutory decision and plan remain with the local authority.
Why 2026 may bring more common SEND document structures
The paperwork has not suddenly become a single national form. Local authority variation is still real. But the direction of travel is changing.
The 2026 SEND reform proposals would introduce digital Individual Support Plans for children and young people with SEND, new National Inclusion Standards, and nationally defined Specialist Provision Packages for children with more complex needs. Those are proposals, not a new statutory template that schools must use today. They do point towards more common language, digital records, and centrally defined expectations over time.
The funding arrangements reinforce that direction. The DfE's Inclusive Mainstream Fund guidance describes over £500 million per financial year for three years, including £400 million for schools, and requires schools to publish an inclusion strategy by 31 December 2026. That strategy must explain the barriers affecting the cohort, the activity funded to address them, and the intended outcomes. The guidance also says that future national inclusion standards will help schools develop those strategies and that funding is intended to become simpler and more consistent over time. For the immediate record implications of this funding and guidance, see our guide to what the new inclusion guidance means for SEND records in schools.
It is therefore reasonable to expect more centralised structures in SEND documentation: common fields, digital plans, shared definitions, and clearer expectations about what support and evidence should be recorded. That does not necessarily mean one rigid form for every school. A better description is standardisation around the important information while allowing local and setting-specific detail where it matters.
That creates a different design test for SEND software. Future-proofing does not mean forcing every school into an unfinished national form. It means keeping the evidence and core structure organised separately from today's output format, so schools can work with their current local-authority documents and adapt when shared national fields or digital plans become clearer.
That is where MeritDocs is designed to help. It supports different document structures and helps staff review imported information before adopting it. Today, that means working with local variation rather than pretending it has disappeared. If national or local requirements become more consistent, a school has a more useful starting point when its evidence, versions, review dates, and approved documents are already structured and exportable. We cannot promise a government schema before it exists. We can build around the principle that the record should remain usable when the format changes.
What we learned from local authority examples
We gathered published and shared SEND document examples from local authorities across England and looked at how they actually differ.
Some of the differences are cosmetic. Many are not. Documents vary in how they express outcomes, how they separate universal provision from targeted support, how they record parent and pupil views, how much detail they expect about baseline and impact, and how they handle review timing.
Those examples helped us support the document structures schools actually use, rather than forcing every plan into one generic template.
Why we test with synthetic pupil cases
You cannot responsibly develop and test this kind of system on identifiable pupil records.
So we built synthetic test cases instead, working with practising SENCOs to make them realistic. These are invented pupils with invented histories, but the combinations are drawn from professional experience: the Year 4 pupil with speech, language, and communication needs whose parents disagree with the school's assessment; the Year 9 pupil whose social, emotional and mental health (SEMH) needs surface as attendance decline; the pupil with a two-year-old educational psychology report and no recent evidence; the pupil whose record contains a professional's carefully qualified opinion that must not be flattened into a statement of fact.
Synthetic cases let us test realistic combinations without using real pupil records. They help us check that gaps remain visible and that qualified professional language is not flattened into certainty.
The DfE's guidance on generative AI in education is clear that personal data must be protected under data protection law and that staff remain responsible for checking outputs. Synthetic development data is not a substitute for those obligations in live use, but it does mean our testing does not depend on real children's records.
What this means for schools
MeritDocs helps staff turn recorded pupil information into structured SEND documents using relevant UK SEND material. Different document types can follow the structure they need, rather than forcing every plan into the same format.
The workflow keeps source context and gaps visible for staff review. Review arrangements stay part of the school's working process, and staff read, edit and approve every draft before it becomes part of the record.
For a closer look at what happens when source material does not fit neatly into a destination structure, see why unmapped content matters when schools import SEND documents.
The edge cases that screenshots miss
The hardest variation is not always a different logo or table. It can be a change in the facts behind the document.
One composite synthetic case makes the point. It describes a fictional Year 11 pupil who turns 16 during an annual-review window, whose family moves across a local-authority boundary, whose record includes a personal budget and social-care provision, and whose plan moves between draft and final states. This is an illustrative case study, not a report of one verified production incident.
The case asks questions that a visual check cannot answer. Does age-sensitive wording follow the pupil's current status rather than the date the document was first created? Does a move update the relevant authority context without mixing current rules with stale presentation data? Do status-sensitive sections stay correct when the plan changes from draft to final? Can staff see that a dependency changed and confirm the consequences before export?
The lesson is useful beyond this scenario: when a pupil's circumstances change, staff need a clear review step to check that the document still reflects the current evidence before export.
That is also why our existing synthetic cases deliberately include thin evidence, contradictory sources, qualified professional language and different local-authority structures. A document that looks polished is not necessarily a document that has preserved the facts that matter.
What SENCOs told us
Here is our internal result, stated plainly. SENCOs we consulted rated the generated documents 9.5/10.
That is a consultation result from practitioners we worked with during development. It is not an independent benchmark, an accuracy percentage, a statutory quality mark, or a claim about how the system will perform on your pupils in your local authority format. We are not going to dress it up with a sample size, a rubric, or a significance claim we have not published.
What we think it does support is a modest claim: the documents are high quality, and the system is built to reflect expert practice. A serious state of the art claim would need a published benchmark with a defined method, and we do not have one. When we do, we will say so.
A checklist for evaluating any AI SEND document tool
Ask any supplier, including us:
- Does it support more than one document structure, or one generic template?
- If it discusses EHC plans, does it clearly distinguish school evidence and review work from the local authority's statutory responsibility to prepare and maintain the plan?
- What reference material grounds the output, and is it UK SEND material?
- Was it developed and tested without identifiable pupil records?
- Does it preserve pupil, parent, and professional voice with attribution intact?
- Does it show uncertainty and unmapped content, or fill every field?
- Who controls review dates, the application or the model?
- What happens to pupil data, and is it used to train anyone's model?
- Is human approval a required step or an optional one?
- Can the supplier explain the workflow in plain English?
If a supplier answers only the last question well, keep asking the others.
Where this leaves schools
None of this makes a generated document automatically compliant, guarantees an inspection outcome, or removes the professional work of knowing a pupil. It cannot, and any supplier suggesting otherwise is selling something schools should not buy.
The result is a more useful starting point for staff: grounded in the school's recorded information, clear about gaps, and still subject to professional judgement.
That is the standard we hold MeritDocs to, and it is a reasonable standard to hold any AI SEND tool to before it touches a live pupil record.
Sources
Children and Families Act 2014
Department for Education's move guidance