Three quarters of American middle school educators now use AI every week, and 24% of educators think their education system is adapting well to it. Those two numbers sit a few lines apart in the AI Readiness in US Schools report that IBM published on 2 September 2026, from a Morning Consult survey of 1,019 school staff and 1,029 parents in July. IBM's headline was that adoption is outpacing readiness. The more useful question is what readiness is made of, because you cannot close a gap you cannot name.
Here is the answer I work to. AI readiness has seven dimensions, and an organisation is only as ready as its weakest one.
What are the seven dimensions of AI readiness?
The Seven Dimensions of AI Readiness are the seven things an organisation must have in place before its use of AI can be called ready rather than merely widespread: purpose, permission, ownership, capability, protection, foundations and evidence. Your readiness is your lowest score across the seven, not your average. That is my suggested way of thinking about it, built from the leadership teams I have sat with rather than from a survey, and it is the diagnostic that sits underneath a test some readers will already know.
In an earlier article I set out the Readiness Test: three questions an organisation must be able to answer consistently, from the top to the front line, before it can call itself AI ready: What is AI for here? What may I do with it? Who decides when it goes wrong? The test tells you whether you are ready. It does not tell you why not, or where to start. The seven dimensions do. The first three dimensions are the three questions. The other four are what makes the answers hold when a model changes, a parent complains or the person who wrote the policy leaves.
Each dimension below comes with one diagnostic question you can ask this week, and a picture of weak and strong. Score each from 0 (absent) to 2 (in place and known).
1. Purpose: does everyone give the same answer to "what is AI for here?"
Purpose is in place when three people at different levels give the same answer to what AI is for, without looking anything up. The diagnostic is exactly that: ask a senior leader, a middle leader and someone on the front line, separately, and compare.
Weak looks like three different answers, each sincere. The head says "to reduce workload", the head of department says "to personalise learning" and the teacher says "I don't know, I use it for emails". None is wrong, and that is the problem: an organisation with three purposes has none. Strong looks like one sentence everyone recognises that names what will change, not what AI will be used for. "We are using AI so that teachers get their Sunday evenings back and feedback reaches pupils within a day" is a purpose. "We are exploring AI's potential" is a hobby.
Purpose is the cheapest dimension to fix and the one most often skipped, because it is a leadership conversation rather than a purchase.
2. Permission: could a new member of staff find out in five minutes what they may do?
Permission is in place when anyone in the organisation can find out, quickly and without asking a favour, what they may use AI for, what they may not put into it and which tools are approved. The diagnostic: give a new colleague five minutes and see whether they can answer those three questions from what you have published.
Weak permission is a policy that exists and is not known. Microsoft's AI in Education Report, published 24 June 2026 from 3,345 students, educators and leaders in six countries including the UK, found 58% of education leaders saying their institution is implementing or scaling AI. Scaling without known permission means staff working out the rules for themselves, which is how one department ends up banning a tool that another department requires.
Strong permission fits on one page, names tools rather than categories, and says what is not allowed in the same plain voice as what is. It is the first of the three conditions I described in how leaders build AI confidence across a workforce, and it belongs among the dimensions of readiness for the same reason: people cannot use well what they are unsure they are allowed to touch.
3. Ownership: who decided the last AI question, and how long did it take?
Ownership is in place when a named person can decide an AI question, has had something removed from their role to make room for it, and has a deputy. The diagnostic: name the last AI decision your organisation made, who made it, and how long it took from the question being asked.
Weak ownership is a committee, a keen volunteer with a full timetable, or the head of IT by default. In each case "how long did it take?" is answered in terms, and staff learn that asking is slower than doing. This is the pattern behind why AI adoption is really a leadership problem: the Sponsorship Test asks "have we removed what it replaces?", and ownership fails on that question more than on any other.
Strong ownership is a name, a deputy, a decision route rehearsed for the day something goes wrong, and a leadership team that backs the decision the owner makes. The owner need not be the most technical person in the building. They need to be the person whose judgement colleagues already trust.
4. Capability: how many people used AI on their own work this week, and would they notice a wrong answer?
Capability is in place when most staff have used AI on real work recently and can tell when its output is wrong. The diagnostic has two halves, and the second matters more: count the people who used it this week, then ask a few of them how they checked the result.
This is where the readiness gap shows most plainly in the evidence. IBM found only 20% of US educators had received extensive AI training, and lack of training was the barrier they named most often. Microsoft found 53% of educators without formal training and 66% wanting it monthly or quarterly. In England, Pearson's School and College Report 2026, published in June from more than 11,000 teachers, put the share confident in teaching about AI at 16%, up from 9% a year earlier, with 8% feeling prepared to support pupils for AI-enabled futures. Use is high; the ability to judge the tool is not.
Weak capability is fluency without judgement: staff who can produce a worksheet in thirty seconds and would not spot the invented citation in it. Strong capability is dull and repeated: people using AI on their own work alongside someone who has done it before, and a habit of checking before trusting. Training buys the first half. Only practice buys the second.
5. Protection: when did you last hear about an AI mistake from the person who made it?
Protection is in place when it is safe to be seen using AI and safe to say when it went wrong, and the organisation extends the same openness to the people its AI use affects. The diagnostic is a date: when did a member of staff last tell a leader, unprompted, that AI had got something wrong in their work?
If the answer is "never", you do not have a workforce that makes no mistakes. You have one that has decided not to tell you. Weak protection also shows from outside. IBM found 77% of parents want a say in how AI is used in their child's classroom and 20% clearly understand the guidance their school has given. An organisation whose own community cannot describe its AI use is not protected from the first difficult conversation. It is simply not having it yet.
Strong protection treats the first reported mistake as information and says so publicly, and tells parents, customers or pupils what AI is used for in plain words before they ask. It is the third condition of confidence, and I count it as a dimension of readiness because an organisation that cannot hear about its errors cannot correct them, and a readiness that cannot correct itself will not survive the next model release.
6. Foundations: could you list which tools hold your data and who signed for them?
Foundations are in place when the organisation knows which AI tools hold its data, under what agreements, on what devices and networks, and has checked those tools against the standards it is bound by. The diagnostic: produce the list, with a signature against each line, in an hour.
This is the dimension most readiness tools measure, and measure well. The K-12 Gen AI Maturity Tool from CoSN (the Consortium for School Networking) and the Council of the Great City Schools, first published in April 2024, rates US school districts as emerging, developing or mature across executive leadership, operations, data, technical, security and legal risk, with an academic AI literacy domain added since; Wisconsin's Department of Public Instruction points its districts to it as their readiness self-assessment. If your foundations score is low, use a tool like that; it is more thorough than an article can be.
But notice its shape. Five of its domains sit inside what I am calling foundations. Purpose, permission, protection and evidence barely appear, because a rubric written for technology leaders measures what technology leaders control. Weak foundations are a real risk: a data breach or a procurement failure can end an AI programme in a week. Strong foundations, on their own, are a well-secured organisation that still cannot say what AI is for. Necessary. Nowhere near sufficient.
7. Evidence: what did your last AI decision change, and how do you know?
Evidence is in place when the organisation can say what its use of AI has changed, for whom, against a measure it named in advance, and has a date on which it will look again. The diagnostic: take the most recent AI initiative and ask what it changed and how you know.
Weak evidence is usage. Licences activated, prompts run, staff trained: these count activity, and IBM's figures show why they mislead. Three quarters of middle school educators using AI weekly and a quarter of educators believing the system is adapting well can both be true, because usage was never a measure of readiness. Weak evidence also has no date, which is how pilots become habits. I set out the Exit Question for exactly that failure: if this works, what will change, who has already agreed to it, and on what date will you decide?
Strong evidence is small and specific. Feedback turnaround in days. Hours returned to a named group, and where those hours went. The number of AI decisions that reached the owner and how quickly they were resolved. Pick two measures, name them before you start, and put the review in the diary.
Why adoption is outpacing readiness
Adoption is outpacing readiness because leaders invest in the two dimensions that can be bought, foundations and capability, and leave the five that can only be decided. That is my reading of the evidence rather than a finding within it, but every survey above is consistent with it. Training and infrastructure appear on procurement lists. Purpose, permission, ownership, protection and evidence appear nowhere, because there is nothing to buy; there is only a leadership team that has to agree, write it down and be seen to hold it.
The result is an organisation with good WiFi, a signed data agreement and a well-attended training day, in which nobody can say what AI is for, the owner has no time, mistakes go unreported and success has no measure. It will report high adoption. It will also be the one the newspaper writes about when something goes wrong.
What I See in Practice
Across the schools, trusts and districts I work with, in England and internationally, readiness almost always fails in dimension 1 or 3, and the leadership team almost always believes it has failed in dimension 4 or 6. They ask me for training or a tools review. When I ask three people separately what AI is for here and get three answers, the conversation changes.
The organisations that pass are not the best resourced. The readiest schools I have worked with have often had less technology than their neighbours, because a head had answered the purpose question in one sentence, given one person the job with time attached, and told parents what the school would and would not do before anyone asked.
The pattern I would ask every leadership team to notice is the direction of the fix. Foundations and capability are bought from outside and improve slowly. Purpose, permission, ownership and evidence are decided inside and can improve in a month. Protection sits between the two: leaders decide it, and time earns it.
How to use the seven dimensions
Score all seven, read the lowest number first and fix that dimension before you spend anything on the others. That is the whole method. A hypothetical leadership team scoring 2 on foundations, 2 on capability, 1 on permission, 0 on purpose and 0 on evidence is not "mostly ready". It is not ready, and its next action is a one-hour conversation about purpose, not another training day.
Three rules keep the scoring true. Score from the front line, not the boardroom: leaders describe the organisation they intended to build, staff describe the one they work in. Score by asking, not by reading: a policy that exists earns nothing until people can find it and repeat it. And rescore each term, because readiness that is not re-examined decays quietly as the tools change under it.
Then put the seven words on one slide, with the seven questions underneath, and take it to your board or governors. It is the fastest way I know to move a governance conversation from "are we using AI?" to "are we ready to be?"
Working through this with your leadership team
If the seven dimensions have shown you a gap you cannot close alone, this is the kind of work I support. I work with schools and organisations to turn AI readiness questions into a clear set of priorities and an implementation roadmap, starting from the lowest score rather than the latest tool.
Sources and further reading
- IBM, New IBM Study Finds AI Adoption Is Outpacing K-12 Readiness, press release on the AI Readiness in US Schools report, Morning Consult survey of 1,019 K-12 education professionals and 1,029 parents fielded July 2026; IBM Newsroom, 2 September 2026.
- Microsoft, Microsoft's New AI in Education Report highlights widespread adoption and increasing demand for support, 3,345 respondents across the United States, United Kingdom, Australia, Brazil, Japan and Saudi Arabia; Microsoft Source, 24 June 2026.
- Pearson, The Pearson School and College Report 2026: Foundations for the future, more than 11,000 teachers and tutors and 3,000 students; Pearson UK, June 2026 (figures from Pearson's 4 June 2026 release).
- CoSN and the Council of the Great City Schools, CoSN/CGCS K-12 Gen AI Maturity Tool, 9 April 2024, with the later academic module described on the Wisconsin Department of Public Instruction's AI District Readiness Self-Assessment page.
Dan Fitzpatrick is the founder of The AI Educator, a Forbes contributor and bestselling author on AI in education, and works with schools, trusts and organisations on AI strategy and readiness. About Dan.
