Treat the application as a document, not as the candidate. A polished letter now tells you that someone had a good afternoon with an assistant, not what that person does in front of thirty children. Move the weight of your decision to what you can watch happen in the room, ask every finalist how they think about AI, and write down a rule for any tool that sorts your pile.
Postings for next September are a few weeks off, and that is the moment to decide. Once forty files land in one folder on a Thursday night, nobody redesigns anything. Education Week reported in April 2026 that the share of teachers using AI in some capacity rose from 34 percent in 2023 to 61 percent in 2025, citing EdWeek Research Center survey data. Assume some of your applicants used it on their applications, and that some of your own systems use it on them.
This article is written for a principal (a headteacher, in the UK) who sits on hiring panels. The legal detail differs by country, and I flag where.
Can you tell which applications were written with AI?
No, and you should stop trying, because a guess dressed up as detection decides who gets an interview.
Detectors misfire, and I have set out why in do AI detectors actually work. A hiring file adds a twist: the candidate cannot appeal a hunch. Tes, advising candidates in June 2026, noted that for some school leaders, spotting AI use could be enough to put an application in the "no" pile. That rule punishes the candidate who used an assistant to tidy a sentence and rewards the one who wrote everything unaided and claimed a result they never achieved.
The question that matters is whether the claims are true, and you can test that. Pick one claim in each shortlisted file and ask about it later, in the room. Here is a hypothetical. A letter says: "I led a reading intervention that closed the gap for students from low-income families." At interview you ask what the candidate did in week one, and what they would change. Someone who ran the intervention answers with texture: which children, which room, the session that fell flat. Someone who did not has only the letter to recite. No software was needed, and the assistant's role in the letter stopped mattering.
What is the Room Test?
The Room Test is the filter I suggest for any piece of evidence in a hiring file, because it asks where the evidence came from and what you will still need to see.
It rests on the Three Ps, my way of splitting any piece of work into product, process and performance. Applied to hiring, it reads like this:
The Room Test is three questions I ask of any piece of evidence in a hiring file: Would this look the same if a machine had written it? Could the candidate show me how it came to be? And what will I see them do, live, that no file could show me? Evidence that survives all three is a candidate. Evidence that fails the first and has no answer to the others is a well-made document.
Evidence comes from three places, and AI has changed what each is worth:
| Place | What it is | What AI does to it |
|---|---|---|
| The file | What they sent: letter, form, CV | Makes it uniformly fluent, so it separates candidates less than it used to |
| The trail | How they got there: a claim they can unpack, a reference who can say what they saw | Little. An assistant cannot remember a child |
| The room | What they do while you watch: a lesson, a task, a conversation | Nothing, unless you let them bring a script |
A polished application tells you about the document, not the person. I offer the Room Test as my suggested way of thinking, not as a validated instrument. Its practical effect is to move your decision from the first row to the third.
What should you ask candidates about AI?
Ask how they think about it, not how much they know, because judgment is what you are hiring and fluency is what a weekend course gives.
Education Week asked principals and assistant principals in April 2026 what AI questions they put to teaching candidates. The answers share a shape. One principal said what he listens for is "not whether they are an expert in AI, but how they think about it." Others described being wary of a candidate who dismisses AI outright, and equally wary of one who embraces it without a thought for its effects on students. These are practitioners' views in a short feature, not survey data, and I use them as a sensible starting point.
Here are five questions I would put to finalists, and what I would listen for:
| Question | A strong answer sounds like | An answer that gives me pause |
|---|---|---|
| When have you decided that students should not use AI on a task? | A specific task, a reason tied to what the task was for | "Never" or "always" with no task attached |
| When have you let them use it, and what did you check afterward? | A named use, plus how they knew learning happened | Enthusiasm with no check |
| How do you use it in your own planning, and what do you never let it do? | Honest use and a line they drew themselves | Pretending not to use it, or no line at all |
| A student hands in work that looks too good. What do you do first? | Talk to the student before any accusation | A detector score treated as proof |
| What would you want to learn about AI in your first year here? | A real gap, named without fear | "I'm already confident" or "I'd rather not" |
None of these has a right answer. All of them have a thin one, and the thin ones sound the same in every school.
Try the fifteen-minute critique
The one task I would add to any final round is a critique, because it tests subject knowledge and AI judgment together and cannot be written in advance.
Give every finalist the same AI-drafted lesson plan in their subject and phase, with three flaws planted by you: a factual error, an activity that cannot fit the time, and a step that ignores a learner who needs more support. Ask them to fix it aloud in fifteen minutes, then say what they would check before teaching it.
Score three things on a simple scale: did they find the error, did they repair the timing and the support gap with a specific change, and did they name a check before it reaches the class. Tell candidates the format in advance, offer adjustments as you would for any task, and use the same draft for everyone, or the comparison means nothing.
The task rewards teachers who can edit a machine and exposes those who would hand its output to a class unread. If you teach ten-year-olds or run a sixth form, change the plan, not the task. For context on newly qualified candidates, see early-career teachers and AI.
What if your own hiring tools use AI?
Then the tool needs a rule before it needs a renewal: it may sort the pile, and a person decides who is out.
Education Week's April 2026 reporting puts numbers on how common this is. It cites an EdWeek Research Center survey of 270 district recruiters in which 53 percent said they use AI tools. In a connected survey of more than 700 job seekers, only 2 percent said they had applied in the past year to a district using AI in hiring, which suggests most candidates do not know. The same article cites RAND for the finding that about one in ten districts has an AI-use policy and virtually none covers hiring. It also reports University of Washington research in which people followed an AI recommendation about 90 percent of the time, even when they judged it poor. That last finding is the one to remember: a ranked list changes the human who reads it.
On the law, the picture depends on where you work. In the UK, the Information Commissioner's Office (ICO) published an audit of AI recruitment tools in November 2024, covering tools that source, screen and select candidates. It says "the accuracy being better than random is not enough to demonstrate that AI is processing personal information fairly," and that recruiters "must ensure that they inform their candidates how they will process their personal information by AI." The report excluded generative AI tools, so read it as a floor, not a full map. In the US, New York City's Local Law 144 prohibits using an automated employment decision tool unless it has had a bias audit within a year and the required notices have been given, with enforcement beginning July 5, 2023, according to the city's Department of Consumer and Worker Protection. EdWeek reports that the law has been hard to enforce. Most districts sit outside it, so the rule you write yourself is the one that applies.
Five checks cover most of the ground:
- Ask the supplier in writing whether your applicant system ranks, scores or filters candidates, and on what basis.
- Tell candidates. One sentence in the posting is enough.
- Screen it for data protection. In England that means asking your data protection officer for a DPIA decision, as does a school need a DPIA for every AI tool sets out.
- Keep a person behind every rejection. Read a sample of whatever the tool puts at the bottom, and never let a score alone remove someone.
- Name who answers when a candidate asks why. The test in who is accountable when AI gets it wrong applies: a person, not a product.
The tool should also sit on your approved list, with an owner and a review date, like any other. What to require before approving an AI tool shows how.
What I See in Practice
The pattern I see most often is a leadership team spending its energy on the candidate's AI and none of it on the school's.
I come to this as a former classroom teacher, assistant headteacher and Director of Digital Strategy in further education, now working with leadership teams on AI. Across the teams I work with, the faculty lounge argues about whether a cover letter "sounds like ChatGPT," and almost nobody has asked whether the applicant tracking system ranked the pile before a person saw it. The first is a matter of taste. The second is a decision about people, made by a system nobody on the panel can explain.
The same teams tend to write a rule for what candidates may do and none for what the school does. A rule for the school is shorter, and it is the one a candidate can hold you to.
The habit in the head of sixth form's guidance on UCAS applications carries over: let AI work in the margin while the person writes the page. A teacher's application has the same shape as a student's statement. And once someone is hired, AI skills for staff beyond basic training picks up where the interview leaves off.
What should you decide before the postings go out?
Three decisions, each with a name beside it, before the first posting for next September goes live.
- Write the candidate notice. Two sentences are enough. Here is a version to adapt: "We use AI in some of our own work and expect you may in yours. Tell us if you used it to prepare your application, and be ready to talk through any claim in it. People make every shortlisting decision."
- Fix the question set and the critique task. Choose three of the five questions and the fifteen-minute critique. Agree what a thin answer sounds like before anyone interviews, so the panel scores the same thing.
- Settle the tool rule. Send the five checks to whoever owns your applicant system and ask for answers by a date. Where the answer is "we cannot tell," switch the ranking feature off until you can.
Then do one thing differently on the first day of shortlisting: put one claim from each file at the top of your notes. It is a small habit and it moves the decision from the file to the room.
If your leadership team is working through these questions, this is the kind of work I support through AI strategy sessions for schools and through leadership keynotes and workshops. My newsletter, on the AI Educator site, covers the changes worth acting on as they arrive.
Sources and further reading
- Sarah D. Sparks, AI Is Changing Teacher Hiring. Here's How, Education Week, April 2, 2026 (US; reports EdWeek Research Center surveys, RAND and University of Washington findings, and New York City's law).
- Elizabeth Heubeck, The Interview Topic That Could Trip Up This Year's Job-Seeking Teachers, Education Week, April 23, 2026 (US; principals' interview questions; teacher AI-use figures from EdWeek Research Center data).
- Information Commissioner's Office, AI tools in recruitment: Audit outcomes report, November 2024 (UK; audits ran August 2023 to May 2024).
- Tes, Should you use AI for your teaching job application?, June 5, 2026 (UK; advice to candidates).
- New York City Department of Consumer and Worker Protection, Automated Employment Decision Tools (AEDT), undated page (US; Local Law 144).
Dan Fitzpatrick is a keynote speaker, Forbes contributor, bestselling author and the founder of The AI Educator, who has trained more than 150,000 educators across 30+ countries. About Dan.


