Tell teaching assistants three things: which tasks AI may help with, which it must not replace, and who to ask when a task falls between the two. The line is not the tool. It is whether AI ends up standing between the assistant and a pupil who needs a person.
Most schools have not said any of this to their support staff. The guidance goes to teachers, the briefing happens when assistants are with pupils, and the adults who sit beside the children who find school hardest are left to guess. (Teaching assistants are called paraprofessionals or instructional aides in many US districts. Everything below applies to them too.)
What should a school tell its teaching assistants about AI?
A school should tell its assistants which of their tasks AI may help with, which it must not touch, and where to take a doubt. That is one conversation and one page of writing, not a training course.
I suggest sorting everything an assistant does into three zones: preparing, recording and interacting. This is my own way of dividing the work, not an established model. Each zone gets a different default.
| Zone | Typical tasks | Default | Why |
|---|---|---|---|
| Prepare | Adapting a text to a lower reading level, drafting a visual timetable, writing practice questions | Allowed, and an adult reads every word before a pupil sees it | A mistake costs least when a person catches it first |
| Record | Case notes, intervention logs, summarizing an observation, transcribing a conversation | Only in a tool the school has approved, with no pupil details in any other tool | This is where pupil information leaves the building |
| Interact | Explaining an idea, giving live feedback, calming a distressed child, delivering an intervention | A person does it. AI is not the voice the pupil hears or the answer they get | The interaction is the job. A chatbot in that seat changes the job, not just the tool |
The rest of this article takes each zone in turn, after one question: why assistants need their own guidance at all.
Why do teaching assistants need their own AI guidance?
Assistants need their own guidance because the three zones fall on them differently from teachers, and the guidance schools issue is usually written for teachers.
The evidence on that second point is thin, so here is what I can support. A Gallup and Walton Family Foundation report, covered by Axios on May 27, 2026, found that around eight in ten teachers said they had received no formal guidance on applying AI to their work. I have read the Axios coverage, not the Gallup report itself, and the coverage does not state the sample. My inference, which is mine and not the report's: if teachers are short of guidance, the support staff who are least likely to be in the room where it is written are at least as exposed.
There is also an inspection reason. England's inspectorate, Ofsted, says it does not look at AI as a stand-alone part of inspections and says: "We do not expect or require providers to use AI in a certain way, or to use it at all." The same page says inspectors "may ask leaders how they make sure that any use of AI supports the best interests of children and learners." The page was updated on October 21, 2025, and describes itself as not a final position. So the question a leader should be ready for is not "are you using AI?" It is "how do you know your support staff's use serves children?" My longer reading of that position is in what Ofsted expects from schools on AI.
What can AI safely help teaching assistants prepare?
AI can help assistants with preparation, provided an adult checks the output before a pupil sees it, because that is the zone where the evidence is most encouraging.
The Education Endowment Foundation (EEF) ran a randomized trial, evaluated independently by the National Foundation for Educational Research, with 259 teachers in 68 English secondary schools. Teachers who used ChatGPT with a guide to plan Year 7 and 8 science lessons spent an average of 25.3 minutes less per week on planning, a 31 percent reduction. The independent panel found no noticeable difference in the quality of the resources. The EEF itself says that finding rests on a limited sample and "should be treated with caution."
That trial studied teachers, not assistants, and one subject at one stage. It is a reason to believe preparation is the low-risk zone. It is not proof that any particular assistant task is safe. The practical rule is short: the adult reads everything, and anything adapted for a pupil with an education plan goes past the SENCO first. Where to draw the line on AI and SEND support plans covers that boundary in more detail.
Where must a school draw the line on AI and pupil interaction?
A school must draw the line at the moment AI replaces an adult's response to a pupil, because the pupils who most need that response are the ones assistants sit beside.
The EEF's refreshed guidance on deploying teaching assistants says that "pupils who struggle most should spend at least as much time with the teacher as other pupils, if not more." It also recommends assistants deliver structured interventions and use scaffolding techniques to build pupils' independence. My reading, and it is an argument rather than an EEF finding: a chatbot that answers on demand works against scaffolding, which exists to hand control back to the pupil a step at a time. An adult can see when a child needs the answer withheld. A chatbot has no reason to.
Here is a hypothetical to make it concrete. A school is short of adults for the morning and sets a small group of Year 5 pupils to work with an AI tutor on tablets, with an assistant "keeping an eye". Nothing in that arrangement is dishonest. But the school has quietly moved an interaction from the Interact zone to a tool, without deciding to. The default in the table exists to make that a decision someone owns.
What should assistants never put into an AI tool?
Assistants should put no pupil information into any tool the school has not approved, because that is the point where good intentions turn into a data problem.
In England, the Department for Education's generative AI product safety standards, last updated on January 19, 2026, say products should not use personal data for commercial purposes, including further model training, without confirmation of a lawful basis. The standards are mainly written for developers and suppliers, and schools can use them to judge which products are safe. They give a school a fair test to apply, not a ruling on any single tool. Information about a pupil's special educational needs is among the most sensitive a school holds, so the Record zone is where an assistant's helpful shortcut is most likely to do harm.
The rule to give assistants fits in one sentence: if the school has not approved the tool for pupil information, no pupil information goes in, including a first name. The school's approval checklist for AI tools and the rule on putting student data into ChatGPT show how to write it.
The mistake I see most often
The mistake I see most often is announcing an AI rule to teachers and assuming it has reached everyone. This is a pattern across the leadership teams I talk to, not a single case.
As a former classroom teacher and assistant headteacher, I know how often a briefing is timed for the people who can attend it, which is rarely the people who most need it. Two things then happen. Some assistants stop using anything, and a useful hour of preparation is lost. Others carry on privately, often on a personal phone, and that is the worse outcome because nobody can see it.
Another hypothetical: an assistant supporting a Year 5 pupil with reading difficulties pastes a paragraph of the pupil's writing, with their name, into a free chatbot to simplify a text. The intent is exactly right. The tool and the data are wrong. A school that has told its assistants where preparation is welcome, and which tool to use, turns that moment from a risk into a good habit.
What should the conversation with teaching assistants cover?
The conversation should find out what is already happening before it announces anything. Thirty minutes with the SENCO in the room is enough, and it should be clear nobody is in trouble. Six questions do the work:
- Which AI tools do you already use for work, including on your own phone?
- What do you use them for: preparing, recording, or talking to pupils?
- Is there a task you would like to use AI for but are unsure you are allowed to?
- Has anyone asked you to do something with AI that you are uncomfortable doing?
- What pupil information, if any, has gone into a tool?
- Who would you ask if you were not sure?
Questions three and five are where the real information turns up, which is why the SENCO should hear the answers. Then write one page: the three zones, the approved tools, one named person to ask, and a date to review it. Resist writing more. Assistants will use a page that fits on one side.
Training follows from that page, not before it. The same logic applies as for teachers: look at the work, not the attendance sheet when you judge whether training helped. And if some assistants fear the technology is aimed at their jobs, that conversation deserves its own space: see what to tell staff who fear AI will replace them. The Interact zone is a better answer than reassurance, because it shows in writing which part of the work a school has decided stays human.
What to do next
Hold the conversation this half term, write the one page, and name who owns it. A SENCO usually owns the Interact line for pupils with SEND, and whoever leads AI owns the approved list. The headteacher's job is to make sure the support staff are in the room.
If your leadership team is working through how AI changes what support staff do, this is the kind of work I support through staff training and advisory sessions, and through keynotes and leadership sessions for conferences and leadership days.
Sources and further reading
- Education Endowment Foundation, EEF launches updated guidance to help schools deploy their teaching assistants (England; the release carries an embargo note for March 26, 2025).
- Education Endowment Foundation, Teachers using ChatGPT alongside a guide to support them to use it effectively can cut lesson planning time by over 30 per cent (England; independent evaluation by NFER; the page carries no date).
- Ofsted, How Ofsted looks at AI during inspection and regulation (England; updated October 21, 2025).
- Department for Education, Generative AI: product safety standards (England; updated January 19, 2026).
- Axios, AI's education explosion leaves teachers in the dark, reporting a Gallup and Walton Family Foundation report (May 27, 2026; coverage read, original report not read).
Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and international keynote speaker, and has trained more than 150,000 educators across 30+ countries.
