# Is AI Widening the Gap Between Your Students? What a Principal Should Check Canonical URL: https://blog.theaieducator.io/posts/is-ai-widening-the-gap-in-schools Publication: Dan Fitzpatrick Insights Author: Dan Fitzpatrick Topic: AI Readiness and Benchmarking Published: 2026-11-05T07:25:00.000Z Modified: 2026-11-05T07:13:39.637Z The students who lean hardest on AI for schoolwork are not the ones with the most support around them. Here are three questions a principal can answer this half term, with the evidence behind them. ## In brief Yes, AI can widen the gap between students, and the gap is in support, not access. US and UK evidence shows students in lower-income households reporting the heaviest reliance on chatbots for schoolwork, while schools serving them are less likely to have trained staff or given guidance. Dan Fitzpatrick's Gap Question gives principals three checks to make this half term. ## Key takeaways - The gap to watch in AI and equity is not who has access to the tools but who is sitting beside the student when they use them. - In Pew's February 2026 survey of 1,458 US teens, one in five in households making under $30,000 a year said they do all or most of their schoolwork with AI chatbots' help, against 7% in higher-earning households. - RAND found principals in the highest-poverty US schools were about half as likely as those in the lowest-poverty schools to report that AI guidance had been provided (13% against 25%). - The Sutton Trust's 2025 research found 45% of private school teachers had received formal AI training against 21% of state school teachers. - The three studies measure different populations on different lines, so they cannot be added together; they are hints, and a school's own survey is the evidence that counts. - The Gap Question asks which students lean on AI hardest, who sits beside them, and what their work shows by group, and it can be answered in a half term. - A principal should report a position to governors, not an average, because an average hides exactly the students this question is about. Yes, it can, and not in the way most leaders expect. The gap that matters is not who has access to AI. It is who is sitting beside the student when they use it. Students in lower-income households report leaning on chatbots hardest for schoolwork, while the schools serving them are less likely to have trained staff or given guidance. Most principals hear one of two answers to this question. The optimists say the tools are free, so they level the field. The pessimists say the tools favor students who already have advantages. Both are talking about access. The evidence points somewhere more useful, and more uncomfortable, because it is something a school can actually change. ## What does the evidence say about AI and disadvantage? Three studies from the last two years point the same way: heavy student use in lower-income homes, and thinner adult support in the schools around them. On the student side, the [Pew Research Center's February 2026 survey of 1,458 US teens](https://www.pewresearch.org/internet/2026/02/24/demographic-differences-in-how-teens-use-and-view-ai/), fielded from September 25 to October 9, 2025, found that one in five teens in households making less than $30,000 a year say they do all or most of their schoolwork with an AI chatbot's help. A similar share in households making $30,000 to just under $75,000 said the same. Among teens in higher-earning households it was 7%. Pew also found that about six in ten Black or Hispanic teens had used chatbots to get help with schoolwork, against roughly half of White teens. On the adult side, RAND's report on the 2023-24 school year, [Uneven Adoption of Artificial Intelligence Tools Among U.S. Teachers and Principals](https://www.rand.org/pubs/research_reports/RRA134-25.html) (February 2025), found that teachers and principals in higher-poverty schools were less likely to use AI tools than those in lower-poverty schools. Principals in the highest-poverty schools were about half as likely to report that guidance had been provided: 13% against 25%. RAND's district survey, [More Districts Are Training Teachers on Artificial Intelligence](https://www.rand.org/pubs/research_reports/RRA956-31.html) (April 2025), found that by fall 2024, 67% of low-poverty districts had trained teachers on AI use, against 42% of middle-poverty and 39% of high-poverty districts. District leaders expected almost all low-poverty districts to have trained their teachers by the start of the 2025-26 school year, but only about six in ten high-poverty districts. That last figure is an expectation, not an outcome. The UK picture has the same shape, drawn along a different line. The Sutton Trust's July 2025 research, [Artificial advantage?](https://www.suttontrust.com/our-research/artificial-advantage/), used Teacher Tapp polling to compare state and private schools. As the Trust's [own summary](https://www.suttontrust.com/news-opinion/all-news-opinion/mind-the-gaps-three-areas-to-watch-for-ai-disadvantage/) puts it, private school teachers (45%) were more than twice as likely to have had formal training as their state school counterparts (21%), and 38% of state school teachers had received no training at all, against 18% in private schools. Private school teachers were three times as likely (21% against 6%) to say their school had a clear school-wide staff strategy on AI. Here is my reading, and it is an interpretation, not a finding. Put the three side by side and the students who rely most on AI are, on average, in the settings where the adults around them have had the least preparation to respond. The section on limits below explains why that sentence is weaker than it sounds. ## Why is access the wrong thing to measure? Because the tool reaches students whether or not your school has a position on it, while support reaches them only when a school decides it should. A student with a chatbot and a teacher who knows what to ask has an assistant. A student with the same chatbot and nobody who asks, "what did you do yourself?" has a substitute. The software is identical. What differs is whether anyone is placed to tell the two apart. That is why a school can report, truthfully, that every student has equal access to AI, and still be widening a gap. It also explains why the usual measures mislead. A usage figure tells you the tool arrived. It says nothing about whether the work changed or who noticed. If you have not yet read my argument on [students becoming dependent on AI](/posts/students-dependent-on-ai-what-principals-can-do), the short version is that the support has to come down again as the student gets stronger. It cannot come down if it never went up. And the support has to survive the home login, which is the problem I describe in [what a student's AI agent can do with a learning platform](/posts/can-ai-agents-complete-student-coursework). ## The Gap Question: three questions before you tell a board AI is helping everyone The Gap Question is three questions I ask before a school says AI is helping all its students equally: Which students lean on it hardest? Who is sitting beside them when they do? And what would we see in their work, by group, that we are not looking at today? A school that can answer all three has an equity position. A school that cannot has an average, and an average hides exactly the students this is about. I offer it as my suggested way of checking a school's position, not a validated instrument. It is built to be answered in a half term, with information you can collect yourself. ### Which students lean on it hardest? You probably do not know, and that is the first finding. Ask students one anonymous question about how much of a typical piece of homework they do with AI's help, and split the answers by groups you already track: disadvantaged students, students with special educational needs, students learning English, year groups. Merge any group too small to protect anonymity; ten students is my working floor, though your data protection lead may set it higher. The Pew result tells you where to look. It cannot tell you what is true in your building. ### Who is sitting beside them? Name the adult, group by group. For many students who lean on AI hardest, that adult is not the subject teacher. It may be a teaching assistant, a homework club leader, a tutor, a form tutor or a parent. Then ask whether your AI guidance has reached that person. Guidance that lives in a staff handbook and a briefing for teachers has not reached the adult at the homework club. I cover the task-level version of this in [what AI guidance teachers need](/posts/ai-guidance-for-teachers-by-task), and the support-staff version in [what to tell teaching assistants about AI](/posts/what-to-tell-teaching-assistants-about-ai). ### What would we see in their work? Look at the work, not the log. Take ten pieces of independent homework and ten pieces of work done in class from the same students in each group you care about, and read them together. You are not hunting for cheating. You are asking whether the voice, the reasoning and the errors in the homework still belong to the student. If the gap between the two is wider for one group than another, you have found something no usage dashboard would show. It is the same instinct I argue for in [evaluating AI training for teachers](/posts/how-to-evaluate-ai-training-for-teachers): look at the work. ## What I See in Practice The mistake I see most often in leadership teams is reporting AI in averages. Across the leadership teams I work with, the governors' paper arrives with a staff figure, a student figure and a reassuring sentence. Each number is true. None of them can answer the question a trustee ought to ask, which is whether the school's answer is working for the students who most need it to. I have been the assistant headteacher who presents the data and I am now a school trustee who reads it. In both seats, I found the same thing: an average is the number that ends the conversation. The strongest schools I see do something plain. They ask the same question twice, once of everyone and once of the group they worry about, and they treat the difference between the two answers as the finding. ## A worked example (hypothetical) Take a hypothetical secondary school that adds the survey question and splits the answers. Students receiving free meals report doing more of their homework with a chatbot than their peers do. The principal asks who sits with those students in the evening and finds the answer is a teaching assistant who runs the after-school homework club and who has had no briefing on AI at all. The decision that follows is small and specific. The teaching assistant is the first person to get a session on what AI is for and what it is not for. One recurring homework task is restructured so that its first ten minutes happen in class. Next half term, the same survey question is asked again. The principal now has a position to report: here is the gap we found, here is who we put in place, here is what we will look at next. Nobody has banned anything, and nobody has pretended the average was an answer. ## What should a principal do this half term? Four moves are small enough to finish before the holiday. 1. **Ask one question.** Add a single anonymous question on AI-assisted homework to the next student survey, and split the results by the groups you already track. 2. **Name the adult.** For each group where use is heaviest, write down who sits beside those students when they work, and check whether your AI guidance has reached them. 3. **Read twenty pieces.** Compare independent homework with in-class work for the same students, group by group. 4. **Report a position, not an average.** Give governors one sentence per question of the Gap Question, and say plainly which you cannot yet answer. My piece on [what to report to governors about AI](/posts/what-to-report-to-governors-about-ai) shows how to frame it. If you want a starting point for the staff side of the same exercise, [where your school's AI readiness gap sits](/posts/ai-readiness-gaps-school) is the companion, and [the staff survey questions that get honest answers](/posts/staff-ai-survey-questions-school-leaders) will help with the adults. ## What can this evidence not tell you? It cannot tell you what is true in your school, and it cannot be added together into a single national finding. Pew counts students by household income and relies on what teens say about themselves in a single survey of US teens. RAND counts schools and districts by poverty level, in the US, and its district panel was a small sample: 291 districts completed the fall 2024 survey out of 8,017 invited, a 4% response. The Sutton Trust compares state with private schools, which is a different line from deprivation. I looked for a UK study that splits students' own AI use by deprivation and did not find one. The RAND figures describe 2023-24 and fall 2024, and the training gap may have narrowed since. Treat the three studies as three hints pointing the same way, and treat your own survey as the evidence that counts. ## Where to take this next If your leadership team wants to see where its guidance and training have actually reached, and for whom, that is the work behind my [school AI readiness support](https://theaieducator.io/school-ai-readiness?utm_source=blog.theaieducator.io&utm_medium=referral&utm_campaign=is-ai-widening-the-gap-in-schools). I also speak to school leadership teams and conferences on this question through my [education keynotes](https://theaieducator.io/education-keynotes?utm_source=blog.theaieducator.io&utm_medium=referral&utm_campaign=is-ai-widening-the-gap-in-schools). ## Sources and further reading - Pew Research Center, [Demographic differences in how teens use and view AI](https://www.pewresearch.org/internet/2026/02/24/demographic-differences-in-how-teens-use-and-view-ai/), February 24, 2026 (US; 1,458 teens aged 13 to 17; fielded September 25 to October 9, 2025). Companion report: [How Teens Use and View AI](https://www.pewresearch.org/internet/2026/02/24/how-teens-use-and-view-ai/). - RAND Corporation, [Uneven Adoption of Artificial Intelligence Tools Among U.S. Teachers and Principals in the 2023-2024 School Year](https://www.rand.org/pubs/research_reports/RRA134-25.html), February 11, 2025 (US). - RAND Corporation, [More Districts Are Training Teachers on Artificial Intelligence: Findings from the American School District Panel](https://www.rand.org/pubs/research_reports/RRA956-31.html), April 8, 2025 (US; 291 districts). - The Sutton Trust, [Artificial advantage?](https://www.suttontrust.com/our-research/artificial-advantage/), July 16, 2025 (UK; Teacher Tapp polling), and its summary, [Mind the gaps: three areas to watch for AI disadvantage](https://www.suttontrust.com/news-opinion/all-news-opinion/mind-the-gaps-three-areas-to-watch-for-ai-disadvantage/). *Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and a bestselling author on AI in education. [About Dan](https://www.theaieducator.io/about).* ## Frequently asked questions ### Is AI widening the gap between disadvantaged students and their peers? It can, and the evidence points to support rather than access. Pew's 2026 US survey found the heaviest reliance on chatbots for schoolwork in the lowest-income households, while RAND found less AI guidance and training in higher-poverty schools. No study yet proves the gap is widening, so measure it in your own school. ### Do students from lower-income households use AI less than other students? Not in the US survey Pew published in February 2026. One in five teens in households making under $30,000 a year said they do all or most of their schoolwork with chatbot help, against 7% in higher-earning households. I found no UK study that splits students' own AI use by deprivation. ### What should a school measure to see whether AI is widening gaps? Measure three things: which student groups lean on AI hardest, using one anonymous survey question split by groups you already track; which adult sits beside them; and what independent homework shows against in-class work. Usage logs alone tell you the tool arrived, not whether the work changed. ### Who in a school should own AI and equity? Give it to one named senior leader, normally whoever already answers for disadvantaged students, working with the AI lead. The test is whether that person can state, for each group, who sits beside those students when they use AI and whether your guidance has reached that adult. ### Does a school need a separate AI equity policy? Usually not. A short line in your existing AI policy, plus a half-termly check using the Gap Question, does more than a separate document nobody reads. What matters is that governors receive a position on which groups lean on AI most and who supports them, not an average. --- Source: [Is AI Widening the Gap Between Your Students? What a Principal Should Check](https://blog.theaieducator.io/posts/is-ai-widening-the-gap-in-schools) Publisher: [The AI Educator](https://theaieducator.io)