Call it the fluent-output trap: an artificial intelligence (AI) tool can produce a polished paragraph about a pupil in seconds, and the wording can still be unsupported, incomplete or wrong. That is not the same as producing a reliable special educational needs and disabilities (SEND) document, and it is not enough evidence for a school to buy the software.
The right platform should help staff turn recorded evidence into a structured document, keep professional judgement visible, make current records easier to find, and reduce avoidable repetition. It should also be honest about uncertainty, data protection and the work that remains with the school. The wider question is whether it works as SEND software for schools, not just whether it can produce fluent text.
If you are assessing AI-powered SEND software, ask these ten questions before you are impressed by the demo.
What should schools look for beyond fluent output?
Schools should look for software that is grounded in recorded evidence, supports the document structures staff use, keeps pupil and parent voice distinguishable, shows what needs checking, requires professional review, handles existing records honestly, protects sensitive information and makes the approved record easier to manage afterwards.
The software should improve a real SEND workflow. The fact that it uses AI is secondary. That is the standard for SEND management software, whether the platform uses AI or not.
1. What exact SEND documents can it help staff create?
Start with the documents your school uses.
A supplier may say that its platform generates SEND documents, but that phrase can cover anything from a blank template to a fully structured workflow. Ask whether the product supports the document types and sections your staff need now, not a generic form that happens to contain the word SEND.
Ask:
- Can it support SEN Support Plans?
- Does it handle different structures for different document types?
- Can it support profound and multiple learning difficulties (PMLD) planning where that is relevant to the setting?
- Are outcomes, provision, review arrangements and pupil voice kept distinct?
- Can the school adapt the workflow to its local authority context?
The SEND Code of Practice provides a national framework. It does not make every school's paperwork identical. A platform that forces every setting into one rigid template may create tidy-looking documents while losing useful local detail.
2. What evidence is used to produce the draft?
A generic prompt produces a generic answer. A useful SEND workflow begins with information the school has recorded.
Ask the supplier to show the path from source evidence to proposed wording. You should be able to understand what the system used, which sections it is working on and where staff judgement is still required.
Look for a clear distinction between:
- information recorded by the school;
- information reported by a parent or carer;
- the pupil's own words;
- professional recommendations;
- generated wording;
- information that still needs confirmation.
A supplier should be able to explain what grounds the output and what happens when the relevant evidence is missing. "Our model knows SEND" is not a sufficient answer.
3. What does the system do when there is not enough evidence?
This is one of the most important demonstrations to request.
Give the supplier an incomplete example. Leave out the baseline, the pupil's view or the evidence of how an adjustment has worked. Then watch what the software does.
A trustworthy workflow should make the gap visible. It should not fill every empty field with plausible professional language simply because the template has a space to complete.
Ask:
- Does the system flag missing or unmapped information?
- Can staff see which sections need investigation?
- Are qualifiers such as "reported by parent" or "not yet observed" preserved?
- Can a staff member reject a suggestion without losing the source record?
An obvious gap can be followed up. An invented fact can be copied into future documents and mistaken for history.
4. Does it preserve pupil, parent and professional voice?
SEND records contain different kinds of evidence. They should not all be flattened into one polished paragraph.
A pupil's words are not the same as a teacher's observation. A parent's report is not the same as a professional recommendation. A qualified statement such as "may find busy transitions difficult" should not quietly become the definite claim "finds transitions difficult".
Ask the supplier to demonstrate a sample containing:
- a direct pupil quotation;
- a parent's description of what happens at home;
- a teacher observation;
- a speech and language or educational psychology recommendation;
- uncertainty or disagreement between sources.
Then check the output. Can the reader still tell who said what? If not, the platform may be improving the prose while damaging the evidence.
5. Is the document structure controlled by the application?
A long block of generated text is not a SEND workflow.
The platform should help staff work with meaningful sections, clear outcomes, provision detail, review arrangements and the relevant evidence. It should not depend on every member of staff writing a perfect prompt to produce a usable document.
Review timing is a good test. Can the application manage the school's review process, or can generated text accidentally invent a date or present an informal suggestion as an agreed arrangement?
The strongest systems narrow the task AI is asked to perform. They use AI to assist with wording and organisation while keeping document structure, permissions, review state and approval in the application's control.
6. What does human review look like?
"Human in the loop" is not a useful assurance unless the review step is practical.
A staff member should be able to compare proposed wording with the source evidence, edit it, reject it and understand what still needs checking. The final document should not become live merely because someone clicked an approve button under a long block of text.
Ask to see:
- the editing and comparison experience;
- how staff identify generated content;
- how proposed changes are rejected;
- who can approve a document;
- what record is kept of the final decision;
- how corrections are made after approval.
The Department for Education says AI-generated content needs critical judgement for appropriateness and accuracy. The school and its professionals remain responsible for the final record. Good software should make that responsibility easier to exercise, not hide it behind automation.
7. How does it handle the SEND documents you already have?
Migration is where vague software promises become expensive.
Most schools do not have a clean, nationally standardised archive. They have Word documents, Portable Document Format files (PDFs), scans, photographs of paper plans, local authority templates, duplicated records and documents written by different people over several years.
Ask a supplier to process a representative sample that includes both clean and messy records. Then ask:
- What mapped successfully?
- What did not fit the new structure?
- Can staff review the mapping section by section?
- Is the original source retained?
- Can unmapped material be found later?
- Does the supplier claim more accuracy than it has measured?
A useful import process does not pretend that different SEND documents are interchangeable. It shows the best-fit result and leaves staff in control of what is accepted.
8. Can staff find the current record after the draft is finished?
A document is not useful only on the day it is generated.
The platform should help staff find the current version, understand when it was reviewed, see what needs attention and access the relevant history without searching through email and shared folders.
For a teacher or teaching assistant, this may mean finding the current support information quickly. For a special educational needs coordinator (SENCO), it may mean seeing open reviews, assigned actions and documents that need attention. For a headteacher or multi-academy trust (MAT) leader, it may mean understanding whether the workflow is being maintained across settings.
Ask whether the platform supports:
- current and previous versions;
- review dates and reminders;
- assignments and permissions;
- document search and filtering;
- audit history;
- PDF and comma-separated values (CSV) export;
- a clear exit route if the school changes systems.
The generation screen is only one part of the product. The record management around it is what determines whether the tool remains useful after the first month.
9. How are pupil data and AI processing governed?
Do not accept "General Data Protection Regulation (GDPR) compliant" as the whole answer.
Ask the supplier to explain the actual data flow in plain English. A school needs enough information to assess whether the proposed use is lawful, proportionate and consistent with its own policies.
Questions should include:
- What information is processed?
- Where is it stored and processed?
- Is customer data used to train a model?
- Which processors and subprocessors are involved?
- How long are prompts, documents and outputs retained?
- How can the school correct or delete information?
- What access controls and audit records exist?
- What support is available for the school's data protection assessment?
- What happens if the supplier suffers a security incident?
The Information Commissioner's Office (ICO)'s guidance for education technology reminds schools to understand the controller and processor responsibilities in the contract and to check that providers protect children's data, support data protection obligations and process information only as instructed.
The Department for Education's guidance also makes the basic risk clear: AI can generate content that is inaccurate, biased, out of context or out of date. Data protection and output accuracy are connected procurement questions, not separate tick boxes.
10. How will the school prove that the software fits before committing?
A good demo shows possibility. A sensible pilot tests fit.
Before buying, agree a small evaluation using real workflow conditions and appropriately controlled data. Include:
- a typical new document;
- a plan with incomplete evidence;
- an imported PDF or scan;
- a case containing pupil and parent voice;
- a review that needs an amendment;
- a handover where another member of staff needs the current record.
Let the SENCO, a teacher or teaching assistant, a senior leader and the person responsible for data protection see the relevant parts of the process. Record what each person needs to do, how long it takes, what remains unclear and what happens when the software is wrong.
The school should leave the pilot with answers to practical questions, rather than a favourable impression of the interface.
Where MeritDocs fits
MeritDocs is a lean, AI-enhanced whole-SEND provision platform for schools. It supports costed provision mapping, pupil passports, SEN Support Plans, PMLD plans, reviews and reporting, while the Document Wizard helps staff create structured documents using UK SEND terminology and grounded source material. Staff review and approve the result. The system does not replace SENCO judgement.
The Documents Hub then gives schools a place to keep SEND documents findable, track review dates, manage access and export records. For existing archives, MeritDocs supports best-fit mapping from PDFs, Word files and photographs of paper plans. Because SEND documents vary by local authority and setting, staff confirm what is brought into the new structure and what still needs attention.
That is the standard to apply to any supplier. Do not buy an AI label. Buy a workflow that helps staff produce, review and manage better records without pretending that software can make professional decisions for them.
For the broader question of reducing SENCO admin without losing accuracy, see How SENCO software should reduce paperwork without losing accuracy. For the governance questions around AI-assisted SEND documents, see AI for SEND documents: 8 checks before staff use it.
The short buying checklist
Before signing anything, ask:
- Does the platform support the documents and workflow our setting uses?
- Can the supplier show how evidence becomes proposed wording?
- Does it expose missing, uncertain and unmapped information?
- Are pupil, parent and professional voices kept distinct?
- Can staff edit, reject and approve content properly?
- Does migration preserve the source and show what needs checking?
- Can staff find the current record and its review history?
- Are access, retention, deletion and AI processing explained clearly?
- Can the school export its documents?
- Has the product passed a realistic pilot with the people who will use it?
The best AI-powered SEND software is not the product that writes the most impressive paragraph. It is the product that helps a school turn evidence into a usable record, keep the right people involved, and know what still needs a professional decision.