AI literacy · for schools & districts
Humy runs the practice inside the class you already teach. Students interview AI figures built on real primary sources, then open the very page each quote came from. Ask, check, ask again, until doubting an AI’s answer is a reflex.
School plans from $600 per school, per year. Free pilot options. No commitment.
Abraham Lincoln
AI-written
What I said at the dedication was brief:
Copied from the document
“The world will little note, nor long remember what we say here, but it can never forget what they did here.”
Source · open the page
Where did you get that?
…we can not hallow—this ground. The brave men, living and dead, who struggled here, have consecrated it, far above our poor power to add or detract. The world will little note, nor long remember what we say here, but it can never forget what they did here. It is for us the living, rather, to be dedicated here to the unfinished work…
Address at the Dedication of the Cemetery at Gettysburg · Bliss copy, 1863
Students adopted the tool before anyone taught them to doubt it.
86% of children ages 9–17 use AI, and only one in three knows AI cannot reliably distinguish fact from fiction. Common Sense Media AI census, June 2026 (survey of 1,204 US children ages 9–17) Just over half say they have been taught to use it safely.
Humy classrooms run it as a three-move routine named for the question itself: the Where-Did-You-Get-That check. It fits any unit where a claim can meet a document.
Students question Napoleon, Frederick Douglass, or a figure the teacher built. The figure answers from an archive of primary sources, and every quote it shows is copied word for word out of the document. The AI never types one. Before the first message, students see which model is answering, how many sources it draws on, and who built it.
One click puts the scanned page on screen with the passage highlighted. Checking an AI's output against evidence is the skill every AI-literacy framework centers on. Here it stops being an abstraction and becomes a habit.
The quotes are verbatim; the sentences connecting them are still generated by the AI. Students learn to see which is which. Every AI they will ever use has the same seam.
“We hold these truths to be self-evident: that all men and women are created equal.”
Character chatbots without sources earned the skepticism. When the Washington Post interviewed an AI Harriet Tubman in 2023, it found factual errors and answers “without any references that would allow for fact-checking.” Washington Post, 2023 (testing another vendor's historical-figure feature) A figure that quotes its documents and opens them on demand is the answer to that objection. Running the check is the assignment. For the hardest cases, read how we approach difficult figures.
What we measured
On the benchmark we published, the same AI model was asked the same 1,250 questions about Napoleon twice: once with 17 historical documents attached, once with nothing but its training data. The difference is what students are usually asked to take on faith. Here they can measure it.
Humy internal benchmark, August 2026: one configuration, 1,250 questions, every quote checked. An AI-literacy tool that hides its own error rate isn’t one, so ours is published.
The classroom move writes itself: put the same question to a general chatbot and to a grounded figure, then open the document and see who survives contact with the page. Numbers, method, and the one citation that opened nothing are in the write-up. Read the benchmark.
AI literacy is the knowledge and skills to critically understand, evaluate, and use AI systems. That is Digital Promise’s definition, the common reference point in US K-12. Digital Promise, 2024 In a classroom it looks like students who know what an AI system is, question what it produces, and check its claims against reliable evidence before repeating them. Every major framework names the middle move, evaluating, as the load-bearing skill. The rest of this page is about practicing it.
Checking an AI’s claims against outside sources can be taught: in a district-wide field experiment, six one-hour lessons measurably improved high schoolers’ ability to judge the credibility of what they read online. Wineburg et al., Journal of Educational Psychology, 2022 (lateral reading; 499 students, treatment vs control)
Digital Promise defines AI literacy as the skills to “critically understand, evaluate, and use” AI, and lists “evaluate the credibility or accuracy of an output or prediction” among its learner skills. Its integration strategy is explicit: embed AI literacy “within the topics they already teach.” Digital Promise AI Literacy Framework, 2024
UNESCO’s AI competency framework for students “encourages integrating AI-related topics into core subjects across the curriculum, emphasizing interdisciplinary learning in both STEM and the social studies.” The history classroom is one of its recommended homes. UNESCO AI competency framework for students, 2024
The newest reference framework, AILit, was finalized in June 2026 with ISTE+ASCD involvement as “a flexible, classroom-ready resource rather than introducing an additional subject area,” shipping learning scenarios embedded across existing subjects. EC-OECD AILit Framework, June 2026
The discipline’s own principles, approved July 2025, warn that generative AI “regularly hallucinates content, references, sources, and quotations” and conclude: “Banning generative AI is not a long-term solution; cultivating AI literacy is.” Their prescribed antidote is sourcing and corroboration, the moves history already teaches. AHA Guiding Principles for AI in History Education, 2025
Five things students do in a Humy assignment, and the named competency each one exercises.
| In the classroom | The named competency it exercises |
|---|---|
| Ask the figure for its evidence, then open the cited page | Digital Promise's Information & Mis/Disinformation practice: “evaluate the credibility or accuracy of an output or prediction.” |
| Compare the AI's framing against the document's own wording | The appraisal of AI output at the heart of the EC-OECD AILit framework's Engage with AI domain. |
| Chase a claim past the archive when the record runs out | Lateral reading, the verification move the field experiment above showed teachable. |
| Anchor the interview in a compelling question, close with an evidence-backed claim | The C3 Framework's inquiry arc: questioning, evaluating sources and evidence, communicating conclusions. |
| Debrief the transcript: where was the AI weakest, who caught it? | Digital Promise's core value of Centering Human Judgment, with the teacher reading every conversation. |
Plain reading, so nobody oversells it to your board: these laws require policies and frameworks, not purchases, and no tool discharges them. What Humy offers is a way to deliver that literacy inside a subject you already staff, on documents you already teach. The rollout and privacy details a district asks for live on the for schools page, the signed agreements on the security page, and the IT approval guide starts an evaluation.
“With Humy, students have to take charge of their learning. They’ve become more adept at critical thinking, forming better questions, and improving their reading comprehension.”
Jacob Chisom
World and American History, Monticello High School, Arkansas
“The feature I use the most is the interview feature. After teaching a topic, most recently Gandhi and Joan of Arc, I have the students ask pointed, focused questions. We practice good conversation techniques beforehand, teaching them to listen (or read) carefully and formulate thoughtful follow-up questions rather than just interrogating.”
Roger Campbell
7th-grade World History, Lancaster County, Pennsylvania
“Humy’s user-friendly platform has enhanced my students’ engagement and understanding of history. It helps them develop critical thinking and engage in insightful discussions.”
Melissa Vanicky
Educator, Staten Island International School, NY
From $600 per school, per year, scaling with the number of teachers. Twenty minutes on a call is enough to see the practice run and get the number.
Most districts start with a handful of teachers. Nyack Public Schools began with a 45-teacher trial.
Digital Promise's definition is the common reference point: the knowledge and skills that let a person critically understand, evaluate, and use AI systems. Every major framework, from UNESCO's student competency framework to the European Commission and OECD's AILit, puts the same skill at the center: evaluating what AI produces instead of taking it on faith. That evaluation skill is the one Humy is built to exercise.
Through practice inside history and social studies. Students interview AI figures built on archives of primary sources. Every quote a figure shows is copied word for word from the document and opens the scanned page it came from, so "where did you get that?" has a checkable answer. Students ask, open the source, compare, and push back: the evaluate-and-verify loop the frameworks describe. The teacher reads every conversation and sets the scope.
With a named routine and a real document. Humy classrooms run the Where-Did-You-Get-That check: interview an AI figure on the unit you're teaching, demand the source behind any claim worth repeating (every quote opens the scanned page it came from), then find the seam between what was copied from the document and what the AI wrote around it. Verification of this kind is teachable: a district-wide field experiment published in the Journal of Educational Psychology in 2022 found six one-hour lessons measurably improved high schoolers' credibility judgments. Humy makes it the weekly assignment instead of a one-off unit.
No. Humy is not a standalone AI literacy program or curriculum, and it doesn't need to be one. Dedicated curricula such as Common Sense Education's lessons, MIT's Day of AI, and Code.org teach about AI in their own sessions, and they pair well with Humy. What Humy adds is the part a slideshow can't: repeated, assignable practice at evaluating a live AI against real evidence, inside a subject the school already teaches. Digital Promise, UNESCO, and the OECD all recommend exactly that embedding.
Yes, and saying so is part of the design. In our August 2026 benchmark, none of the 1,323 quotes the system produced were made up: quotes are copied from documents, never generated. The AI-written prose around them still carried an invented detail in 12.2% of answers, about one in eight, down from 22.6% for the same model without sources. A student catching one of those is the lesson working. Every claim keeps its source one click away so they can.
K-12 through AP. Figures answer at the reading level the teacher sets, and the verification practice scales with age: younger students click the citation and compare the wording; older students corroborate across documents, notice what the archive cannot answer, and critique how the AI framed it. Those are the same sourcing moves as document-based questions (DBQs) and the C3 Framework's inquiry arc.
No purchase satisfies a statute, and we won't tell your board otherwise. Most state action so far requires guidance, district AI policies (mandatory in Ohio since July 2026), or curriculum-framework revisions (California's AB 2876) rather than any specific tool. What Humy gives you is a concrete, supervisable way to deliver AI literacy inside social studies, in the language of the frameworks most state guidance cites. On a demo call, we'll walk your committee through how the practice lines up.
Students join by link or QR code, with no accounts, emails, or passwords. Humy is aligned with FERPA and COPPA, has signed student DPAs covering 16 states, 15 of them through The Education Cooperative and California separately, and student data is never used to train AI models. The signed agreements are published at humy.ai/security.
Still deciding? Write to us at hello@humy.ai.
