For educators

AI for Teachers: Where the Hours Actually Come From

Published August 21, 2026 · By the Kidgeni team

The measured figure is 5.9 hours a week for teachers who use AI weekly — but it does not arrive evenly. It concentrates in administrative writing, first drafts, and differentiation: the tasks where you already know what good looks like and the work is in the typing. It does NOT reliably arrive in grading, planning you care about, or anything requiring knowledge of a specific child. Start with the first list, protect the second, and expect to give some hours back to reviewing output.

Where the hours actually come from

In the Gallup/Walton survey of 2,232 US public K-12 teachers, the largest reported quality gains were in administrative work (74% said their work improved) with grading and feedback at the bottom of the range (57%). That ordering is the useful signal — it tracks how much a task depends on knowing your particular students.

The reliable wins (do these first)

Parent and guardian communication — the newsletter, the field-trip letter, the difficult email you have rewritten four times. You supply the facts and the tone; it supplies the draft.
Differentiation of a text you already chose — same passage at three reading levels, with a vocabulary list. This is the single highest-value task for mixed-readiness classrooms and it is genuinely tedious by hand.
First-draft rubrics, exit tickets, warm-ups, and comprehension questions from material you provide.
Administrative writing: IEP-adjacent drafting (never the determinations), permission slips, meeting agendas, grant-application boilerplate.
Translation of family-facing material into home languages — with a fluent human check before it goes out.

The unreliable ones (where teachers get burned)

Grading with judgement attached. Fine for a first pass on mechanics; poor at recognising the kid who finally attempted a thesis statement.
Anything requiring facts you cannot verify at a glance — dates, citations, science specifics. Models still fabricate confidently.
Lesson planning you actually care about. The output is a competent average of the internet, which is exactly what a good lesson is not.
Anything involving a specific child's history. The model does not know them; you do, and that is the part of the job that is not automatable.

The time it costs back

The Gallup/Walton respondents were clear that saved time is partly offset — by reviewing and correcting output, by learning the tools, and by dealing with AI-related student behaviour. Teachers who report AI as a wash are usually paying the full review tax on every output because they have not yet sorted tasks into 'trust with a skim' and 'verify line by line'. Doing that sort deliberately, once, is what turns a nominal six hours into a real three or four.

A rule of thumb worth adopting

“If I could not spot a mistake in this output in under thirty seconds, it is the wrong task to delegate.”

The policy multiplier

One finding deserves more attention than it gets: the Walton Family Foundation's follow-up work found teachers in schools WITH an AI policy reported a meaningfully larger time dividend than those without. The intuition is straightforward — when the rules are clear, teachers stop spending cognitive energy on whether they are allowed to do this at all, and stop improvising individual answers to student-use questions. A short, clear policy is a productivity intervention, not just a compliance one. Our classroom guide covers what to put in one.

If you have twenty minutes this week

Pick ONE recurring task from the reliable list — the weekly parent newsletter is the classic starting point.
Give the tool your actual context: grade, subject, tone, the four things that must appear, and last week's version as a model.
Time yourself honestly, including the edit. Compare against your real baseline, not your imagined one.
If it saved time, do the same task the same way for three weeks before adding a second one. Tool sprawl is how the dividend disappears.
Check your district's policy and your state's guidance before putting anything containing student information into any tool.

The student-data line you should not cross

Whatever tool you use, treat student names, grades, IEP contents, and behavioural notes as data that does not go into a general-purpose consumer chatbot. Under the FTC's amended COPPA rule — effective June 2025, with full compliance required by April 2026 — using children's personal information to train AI requires separate verifiable parental consent, and retention limits are now explicit. District-procured tools with a signed data agreement exist precisely so you are not the one carrying that risk. When in doubt, de-identify: 'a seventh grader who struggles with topic sentences' gets you the same help as a name would.

Questions parents ask next

What is the single best task to start with?

Family communication or levelled reading passages. Both are high-frequency, low-risk, and easy to check at a glance — which is exactly the profile of a task worth delegating.

Is it cheating for a teacher to use AI for lesson materials?

No — it is drafting, the same as adapting a colleague's worksheet or a textbook resource. The professional judgement stays yours: what to teach, to whom, in what order, and whether the draft is any good. Most district policies say this explicitly.

Should I tell students and parents I use AI?

Yes, and it costs you nothing. Teachers who are open about using AI for a newsletter draft or a practice set find it defuses the double standard students otherwise notice immediately — and it models the disclosure behaviour you want from them.

Can I put student work into an AI tool to grade it?

Only through a tool your district has procured with a data agreement in place, and even then check your state guidance. Student work is student data. A consumer chatbot account is the wrong destination for it, regardless of how good the feedback would be.

Sources

Researched August 21, 2026. Products change; we re-check and re-date this article when they do.

A creative workspace for the whole class

Kidgeni for Schools: pooled Sparks, teacher-managed rosters, moderated end to end, and student work private by default — built for classrooms, not retrofitted from a consumer app.

See Kidgeni for Schools