AI can produce a convincing SEN Support Plan before a SENCO has finished making tea. That is precisely why schools need to be careful.
Fluent writing is easy to mistake for reliable writing. A generic tool may smooth over missing evidence, paraphrase a pupil's words, invent a plausible target or produce language that sounds professional without saying anything measurable. Sensitive pupil information creates another layer of risk.
The Department for Education supports safe and effective use of AI in education, but its position is clear: staff must use professional judgement, check outputs for accuracy and remain responsible for the final document.
Before any AI tool touches a SEND workflow, ask these eight questions.
What should schools check before using AI for SEND documents?
A school should confirm that the tool protects pupil data, uses recorded evidence rather than filling gaps, keeps the document structure controlled, preserves source voice, shows uncertainty, supports human review, manages versions and produces an auditable final record. If the supplier cannot explain those controls plainly, do not use the tool with live pupil information.
That is the short answer. The detail matters.
1. What happens to pupil data?
Start here, not with the demo.
The DfE's policy on generative AI in education says personal data must be protected under data protection law. It recommends that personal data is not used in generative AI tools. Where its use is strictly necessary, the school must make sure the product and process comply with data protection law and its own privacy policies, and that pupils and families understand the processing.
Ask the supplier:
- What personal data enters the AI process?
- Is identifiable information removed before processing?
- Is customer data used to train a model?
- Which suppliers process the data, and in which countries?
- How long are prompts and outputs retained?
- Can the school delete the data?
- Is there a data processing agreement?
- What access and audit controls apply?
Do not accept "GDPR compliant" as a complete answer. Compliance is not a badge. The school needs to understand the actual flow of data and decide whether it has a lawful, proportionate process.
2. Does the AI use evidence or merely a prompt?
A generic prompt invites a generic answer.
Suppose a member of staff writes: "Create a SMART target for a Year 5 pupil with ADHD who struggles to focus." The model can produce something polished, but it has no reliable baseline, no observed context, no pupil voice and no evidence about what has already worked.
A safer workflow starts with recorded information. It should distinguish source evidence from generated wording and avoid unsupported inference.
MeritDocs uses section-aware UK SEND exemplars to help staff turn recorded pupil information into a structured support document. Staff still check and approve the result. The point is not to make AI sound more certain. It is to constrain the job AI is being asked to do.
3. Can it say "there is not enough evidence"?
This is one of the best tests of an AI workflow.
A weak tool treats every empty section as a writing task. A better tool preserves the gap. If the source does not say how the pupil responds to a particular adjustment, the system should not invent a response because the template expects one.
Ask for a demonstration using incomplete evidence. Then look for three things:
- Does the tool flag the gap?
- Does it preserve qualifiers such as "sometimes", "reported by parent" or "not yet observed"?
- Can staff see which content came from the source and which wording was generated?
A blank can be fixed through assessment. An invented fact can sit in a pupil's record for years.
4. Is the document structure controlled?
An AI-generated block of text is not the same as a usable SEND document.
Schools need the right sections, consistent terminology, clear outcomes, provision detail and review arrangements. Those elements should not shift because a member of staff changed the wording of a prompt.
Check whether the tool:
- uses a defined document structure
- separates needs, outcomes, provision and review information
- keeps required fields visible when evidence is missing
- prevents generic filler from replacing measurable detail
- handles different UK document types as distinct structures
- keeps review timing under application control rather than allowing generated prose to set it
MeritDocs builds UK SEN Support Plans as structured documents rather than loose AI prose. Its workflow uses document-specific sections and keeps review arrangements controlled by the application. That is safer than hoping every user writes the perfect prompt.
5. Does it preserve pupil and parent voice?
Pupil voice often gets tidied until it stops sounding like the pupil.
"I hate the dining hall because everyone bumps into me" should not quietly become "the pupil experiences sensory challenges in busy environments". The second sentence may be a reasonable professional interpretation, but it is not a quotation.
The system should keep verbatim words separate from staff analysis. It should also distinguish information reported by a parent from information observed by school staff or recorded by a professional.
Test this directly. Give the supplier a sample with:
- a pupil quotation
- a parent's description
- a teacher observation
- a professional recommendation
Then check whether the output preserves who said what. If every voice becomes one polished narrative, the record has lost useful evidence.
6. What does human review actually look like?
"Human in the loop" is often used as a comforting phrase. It means little unless the review step is practical.
A staff member needs to be able to compare the proposed content with the source, edit it, reject it and understand warnings before anything becomes part of the live record. A single approve button beneath a long document encourages rubber-stamping.
The DfE says AI content requires critical judgement for appropriateness and accuracy. It also says the professional and their organisation remain responsible for the final document.
Ask who is expected to review the output, what they can see during review and what evidence the school keeps of that decision.
7. How does the tool handle versions and corrections?
An accurate plan today can become misleading next term.
Check what happens after publication:
- Is there a clear current version?
- Can staff see the previous approved version?
- Are review notes linked to the document being reviewed?
- Can an incorrect output be corrected without erasing the history?
- Are draft, approved and archived states distinct?
- Can staff tell who changed what and when?
This is where a document workflow matters more than a clever generation screen. The school needs a dependable record after the novelty of the first draft has worn off.
8. Can the school explain the workflow to a parent?
Use plain language.
Could the school explain:
- why AI is being used
- which information is processed
- what the AI does and does not decide
- who checks the output
- how errors can be corrected
- how the pupil's views are protected
If the explanation collapses into model names and vague assurances, the workflow is not ready.
The strongest explanation is usually modest: the tool helps organise recorded evidence and draft a structured document; a responsible professional reviews and approves it; the school remains accountable.
The takeaway
Schools do not need to choose between banning AI and using it carelessly. They need a controlled workflow.
MeritDocs is built around that middle ground: recorded evidence, structured UK SEND documents and staff approval. The useful question is not whether the AI can write. It is whether the school can trace, challenge and trust the record that remains afterwards.