92% of Students Use AI, 77% Have No Training: What Microsoft's 2026 Report Really Signals

Quick answer

Microsoft's 2026 AI in Education report finds that 92% of students have used AI for schoolwork, but 77% report no formal training on it. On the educator side, 87% of educators and education leaders agree AI capability matters for students' futures, but 53% of educators say they have not been formally trained themselves. When asked about cadence, 66% of educators and 52% of students want their institution to provide AI training monthly or quarterly. The gap between adoption and capability - not the raw adoption number - is the story. AI has arrived in schools without a curriculum, and the retrofit is now the whole game.

Key takeaways

• 92% of students in Microsoft's global sample have used AI for schoolwork; 77% report no formal training on how to use it.
• 87% of educators and education leaders agree that knowing how to use AI effectively and responsibly matters for students' futures.
• 53% of educators say they have not received formal AI training themselves, at the same time they are being asked to guide students on it.
• 66% of educators and 52% of students want their institution to provide AI training on a monthly or quarterly cadence, not annually.
• Adoption has outrun capability. The policy question is no longer "should we teach AI" but "how often, and to whom."

Why this data matters

Two of the Microsoft numbers are strategy, not survey noise. The first is 92% - the share of students who have already used AI for schoolwork. The second is 77% - the share of those same students who report no formal training on it. Held next to each other, they describe a system where the tool has arrived without the pedagogy. Schools are not deciding whether to introduce AI; they are deciding how to organise a curriculum around a technology students are already using unsupervised. That is a different problem, and it needs a different response.

For school leaders, the second pair of numbers is where the direction becomes concrete. 66% of educators and 52% of students want AI training on a monthly or quarterly cadence. That is the shape of the ask - not a one-off launch event, not an annual PD day, but a recurring cycle of practice, feedback, and updates as the tools change. Any school AI strategy that ignores cadence is solving a different problem than the one the data describes.

What "formal AI training" actually means

Microsoft's report separates use from training. Use is any interaction with an AI tool for schoolwork - from asking a chatbot for an idea to submitting AI-generated text. Training is structured, recognised instruction: a program, a module, a course, a professional-learning session led by someone qualified to teach it. The distinction matters because the 92% figure describes habits that are already forming, while the 77% figure describes the absence of the frame around those habits.

A student who has been using ChatGPT for a year without ever being taught how to check its output has strong habits. They are just not the habits a school - or a university, or an employer - would design for them. The formal-training gap is where those habits either get corrected or get baked in.

The educator side of the same gap

The report's finding that 53% of educators report no formal AI training is the single most-referenced number in school-leadership discussions of the study, and for good reason. It exposes the operational contradiction inside most current school AI positions:

• 87% of educators and leaders publicly agree AI skills matter for students.
• The people expected to teach and model those skills have, in the majority, not been trained themselves.
• The public policy answer to that gap ("teachers should upskill") is not an actionable request without funded, in-service, role-appropriate training.

This is the specific gap that the NSF's July 2026 $11M award to the Computer Science Teachers Association is trying to close in the US - as we cover in the NSF-CSTA teacher training analysis. Australian schools are seeing the equivalent gap, without an equivalent funded response yet.

What "good" looks like from the data

The Microsoft report is unusual in giving school leaders a defensible cadence to point at. If 66% of educators and 52% of students say they want training monthly or quarterly, then any credible school AI program should meet, at minimum, that cadence. In practice, that means:

That is the shape the data supports. Anything less regular than that is a mismatch to what students and teachers say they need.

Practical examples

• A school that runs one AI professional-learning day per year - a common 2025 pattern - is providing training at ~10% of the cadence 66% of educators asked for in the Microsoft data. The gap between the intent and the delivery is where staff frustration compounds.
• A student who has been using ChatGPT unsupervised for eighteen months has developed habits (good or bad) faster than any single school policy can retrofit. The formal-training piece is either introduced early enough to shape those habits, or late enough to inherit them.
• A department that adds AI as an appendix to an existing computing scheme of work is unlikely to shift student behaviour, because student AI use is not limited to computing. Cross-faculty rollout is the only structure that meets a 92% adoption base.

Common mistakes when reading this data

• Treating 92% adoption as the story. Adoption is the setting, not the finding. The finding is the 77% training gap.
• Blaming teachers for the 53% training gap. It is a resourcing and program-design failure at the system level; individual teachers cannot solve it alone.
• Solving a monthly-or-quarterly ask with an annual launch event. The cadence in the data is specific; solving it with a one-off is a category error.
• Assuming students who use AI are trained on AI. The whole point of the Microsoft data is that they are not - and pretending they are means the habits go unaddressed.
• Reading the report as a Microsoft product pitch. The data is directional and consistent with independent studies (Pew, RAND, Turnitin) - the underlying pattern is not brand-specific.

How the Edison Method applies

Understand. Students start with what AI actually is - a language model probability engine, not a search engine and not an oracle - so unchecked outputs stop being surprising and start being expected.

Use. Structured, recurring practice replaces incidental use. Students prompt, iterate, and compare across models on real tasks, not on toy prompts.

Evaluate. Outputs are checked against sources the student can defend. Detection tools sit alongside human judgement, not in place of it.

Build. Every unit produces a portfolio artefact - a document, a workflow, a prototype - that the student can show and explain.

Lead. Students learn to name when and why they used AI, when they overruled it, and what they would do differently next time. That is the disclosure habit universities and employers are about to require in writing.

For the broader Australian picture, see AI education for teenagers in Australia. For the tertiary consequence of not doing this well, see what the Turnitin 53.6% figure means.

The recommendation: treat the Microsoft report as the operational brief it is. Adoption is settled. Cadence, role-based training, and disclosure are the levers - and they are all things a school can start on this term, not next strategy cycle.

Sources

1. Microsoft Source, Microsoft's new AI in Education report highlights widespread adoption and increasing demand for support, 24 June 2026.
2. Pew Research Center, Teens, Social Media, Technology and AI - corroborating adoption data.
3. RAND Corporation, American Youth Panel: AI use in schoolwork - independent survey data.