Use of AI

Overview

Assistant SENDCo offers optional AI writing assistance to help teachers draft parts of SEN Support Plans. AI features are powered by OpenAI and are designed with pupil privacy as the top priority. This page explains exactly what the AI does, what data it sees, and what safeguards are in place.

Who uses it

AI is optional

Consent

AI is off by default. Before AI does anything, you actively turn it on from a permissions card shown after you log in, which explains what is sent, that nothing you type in a plan and no pupil names are included, and who processes it. You must tick a confirmation and choose to enable it before any data is sent — an explicit, opt-in choice.

AI starts switched off at the beginning of every session, so each time you log in you turn it back on from that permissions card and confirm — you opt in every session. This is a deliberate safeguard so AI is never left running unnoticed on a shared school computer. Your agreement is recorded against your account (not just this browser). AI is opt-in for each member of staff individually; you can turn it off at any time in Settings, and a school can turn it off for everyone.

What the AI does

The AI assists with drafting text. It can:

• Suggest SMART outcomes from the assessment statements you tick
• Suggest provisions and strategies from the statements you tick
• Generate provision ideas from an area of need you choose and the statements you tick (standalone tool, also available inline while you work)
• Suggest how to continue an unmet target into the next cycle
• Choose targets from our ready-written list using the assessment answers, and suggest long-term and short-term targets, setting support, and home support strategies (EYFS plans)
• Suggest changes to a lesson you describe, for the needs shared across your class
• Answer questions about using the app, and take you to the right page or change a writing-tip setting for you, one step at a time (the helper)

In addition, a deterministic quality engine runs entirely on your device to check your plan against built-in structural checks drawn from the SEND Code of Practice, Ofsted, and EEF guidance. This includes checking whether outcomes are SMART, whether provisions specify frequency and responsible staff, and whether your plan maintains a coherent “golden thread” across Assess, Plan, Do, and Review. The quality engine produces a score, inline nudges, and guided fix actions — all without AI and without sending any data.

The AI does not make decisions. Every suggestion must be reviewed and accepted or rejected by the teacher before it is saved.

What data is sent to the AI

Nothing you type in a plan is sent to the AI. A full plan is never transmitted. Only the choices you make on screen are sent, and they are limited to:

• A reference to each statement you tick, or each ready-written target you add, from our own library — the reference, not your words
• Fixed choices you select, such as an area of need or a rating
• When you ask for draft targets, the areas of need recorded on the plan, most significant first
• Wording the AI itself produced, sent back unchanged, when you ask it to build on its own suggestion
• Generic Provision Library entries (school-level resources, no pupil data)
• For the helper: the words you type to it, and the name of the screen you are on. Before anything is sent, your device swaps each of your pupils' names for a placeholder and removes the names on plans shared with you. Common requests, such as “download a plan”, are answered on your device and nothing is sent. No plan text is sent. A notice above the box asks you not to type names or anything that could identify a child, because other details are not removed on your device.
• For “Help with a lesson”: the description of the lesson you type, the kinds of activity you tick, and counts of needs shared by two or more pupils in your class (for example “3 pupils find reading and writing hard”), and which of our own support and target statements two or more pupils share (for example “4 pupils have instructions given in small steps”). No names, no plan text and no single pupil's needs are sent, and health needs are never sent.

Free-text boxes you fill in are never sent, except the box you type into the helper (names from your plans, and from plans shared with you, are removed on your device before it is sent), and the lesson description in “Help with a lesson”, which describes the lesson and never a pupil. The app checks it for your pupils' names and will not send it if it finds one. Two features that sent a pasted EHCP to the AI have been withdrawn, because a pasted EHCP cannot be reduced to those choices. The Section E paste helper that remains does its work on your device and sends nothing to the AI.

The plan quality engine (scoring, inline nudges, and fix actions) runs entirely on your device and sends no data anywhere.

What is never sent to the AI

How anonymisation works

The only typed words that are sent are a lesson description, which describes the lesson, never a pupil, and what you type to the helper, after your device has removed names from your plans. This step still runs on our server and checks those words too, as a backstop rather than the main protection. It covers every AI feature except Provision Ideas, which sends only an area of need and our own statements. It has four layers:

1. Name protection — the pupil's name is never sent, only a placeholder; staff names are replaced with [STAFF]
2. PII scrubbing — postcodes, emails, phone numbers, dates, diagnoses, and staff names are detected and replaced with placeholder tokens
3. Common name detection — about 750 common UK first names are checked and replaced, catching references to siblings or other children
4. Full name pattern detection — any two consecutive capitalised words are treated as a potential name and replaced (with a whitelist for SEND terminology)

Output safety filtering

AI responses are checked before being shown to you. The filter blocks content containing profanity, violence, discriminatory language, medical advice, or anything inappropriate for an educational context. If a response is filtered, you see a generic message and the incident is logged for review.

AI provider

AI features use OpenAI (model: gpt-4.1-nano; “Help with a lesson” uses the larger gpt-4.1; Provision Ideas uses gpt-6-sol) via their API. Under OpenAI's API terms:

• Data sent via the API is not used to train AI models
• API data is retained for up to 30 days for safety monitoring, then permanently deleted
• A Data Processing Agreement is in place between Progress 55 Ltd and OpenAI
• Because OpenAI processes this text in the United States, the transfer is protected by Standard Contractual Clauses with the UK International Data Transfer Addendum, supported by a completed Transfer Risk Assessment. OpenAI is not certified under the EU–US Data Privacy Framework; this transfer relies on those clauses, in addition to the anonymisation described above.

Limitations

AI suggestions are generated by a language model and have inherent limitations:

• Inaccuracy — suggestions may contain factual errors, unsuitable strategies, or targets that do not match the pupil's needs
• Bias — language models can reflect biases present in their training data, which may result in stereotyped or culturally insensitive suggestions
• Not a replacement for professional judgement — AI suggestions are starting points only and must be reviewed, edited, or rejected by the teacher

You are responsible for reviewing all AI-generated content before it is saved to a support plan. Do not accept suggestions without checking they are appropriate for the individual pupil.

Questions

If you have questions about how AI is used in Assistant SENDCo, contact us at george@progress55.co.uk