YOUR SUBJECT · SOCIAL STUDIES

AI in social studies

The thing that actually threatens your course is not a student turning in an essay they didn't write. It is a machine that will summarize any source, argue any thesis, and weigh any evidence in seconds — and will invent a citation, a quote, or a confident historical fact when it hasn't got a real one. What has changed is that it now does the reading and the reasoning for the student, which is either a research tutor or a teacher of invented history, depending on how you set it up.

A STARTING POINT — AND IT NEEDS YOUR JUDGMENT

This page is a draft, and I’m still iterating on it. It wasn’t written by an expert in your field — it is here to start a pathway you then shape to fit you: your subject, your level, your students, your department. Your expertise runs far deeper than anything here, and none of this is finished until your judgment has been on it.

That is the whole point. Judgment is the most important part of this — yours, and the judgment you are building in your students. These pages exist to help you develop and exercise it, never to replace it. Take what is useful, change what is not, and tell me where I have gotten it wrong — that is how these get better.

YOU'LL BE ABLE TO

Configure a tool so it stops doing the source-work and the arguing for your students, and say where in your sequence an evaluation task starts helping rather than hurting.

YOU'LL HAVE

A reason line for your syllabus, five instructions by skill, four tasks worth trying, and four prompts for your own materials.

TIME

The materials prompts pay for themselves the first time you use one. The rest is a unit's worth of thinking.

01 THE IDEA

Say “AI” in a faculty meeting and half the room pictures ChatGPT writing an essay the night before it's due. That happens, but the deeper problem in your department is quieter: the machine will summarize the source, construct the argument, and weigh the evidence — the three skills your course exists to build — and hand the student a fluent, finished result they did none of the thinking for.

And there is a hazard here that other subjects don't have to worry about. The machine fabricates. It invents citations that look real, quotes that were never said, and confident historical claims that are simply false — and it will attribute an invented quote to a real historical figure without a flicker of doubt. A student who trusts it is not just skipping the work; they are learning history that never happened.

What is new is that the machine now does the reasoning out loud. It will explain a document, defend a thesis, and produce “research” on demand. That default completes exactly the cognitive step you are trying to build. Turning it off takes one sentence, and everything else on this page is downstream of that.

02 WHY THIS WORKS

A student is handed a factory owner's 1840 letter defending conditions in his mill and asked what it reveals. Two responses to the same task:

THE DEFAULT

“This source shows that factory conditions were generally reasonable and that workers were well provided for.”

Fluent, one-sided, and the student adopts it wholesale — never noticing that the author had every reason to say exactly this.

AFTER ONE INSTRUCTION

“Who wrote this, and what did he stand to gain? Whose account is missing here? Read it again and tell me whose interests it serves.”

Now the student has to do the evaluating themselves.

This is just sourcing and corroboration — asking who made a document, when, and why before asking whether it's true. The historical-thinking research has said for years that this is the move that separates novices from historians, and that students don't do it unless they are taught to (Wineburg; the Stanford History Education Group). None of that is news to your department. The only new information is that the machine does the un-historical version by default, every time, unless told not to — it reads a source at face value and reports what it “says” — and that a fifteen-year-old will never think to tell it otherwise.

03 START FROM WHAT YOU CARE ABOUT

Everything in this subject turns on one question, and it is a question only your department can answer:

Where in my sequence does a student become able to look at a source — or at machine output — and say “that's not reliable, and here's why,” and catch a citation the machine invented?

Below that threshold, AI “research” is actively harmful — a student who cannot evaluate a source will take the machine's fluent summary and its invented citation as equally true, and you will have taught them to trust history that isn't real. Above it, the same tool becomes something to argue with. Answer in the form “somewhere in the middle of [course]” and then two follow-ups:

What can a student at that threshold reliably catch? An obvious bias, probably. A fabricated citation dressed in perfect formatting, probably not. Be specific — it decides what an evaluation task can legitimately ask for.
In my own courses, where is the analysis or the argument itself the assignment, and where am I actually assessing background knowledge? Reading context you are not testing them on is not the same as constructing the argument you are.

WHAT’S SHIFTING UNDERNEATH

WORTH MORE NOW

Actually reading sources, corroborating a claim, judging where a source comes from, and building an argument from evidence.

WORTH LESS NOW

A perfectly formatted bibliography as proof that the reading happened.

THE UNCOMFORTABLE VERSION

A bibliography was a proxy for having read. A fabricated citation is formatted perfectly and points at nothing.

The new basic: being able to verify a claim and say where it actually came from.

04 WAYS YOU CAN USE IT

The four positions, in this subject, with real tasks. Same four words as every other department — students learn one vocabulary, not nine.

Unassisted

Anything where building the analysis or the argument is the assessment. The DBQ. The document analysis. The essay where the thesis and the weighing of evidence are the whole point.

IN PRACTICE This is most of your summative work, and it is the default. Say it once in the syllabus and you rarely have to say it again.

Cleared

Load you are not assessing. Background context for a unit. A timeline it generates and you verify. Perspective-taking that lets a student test their own view against another.

IN PRACTICE Say what it is cleared for, and verify what it hands back — “cleared” on background quietly becomes cleared for the essay unless you draw the line.

Contested

The tool set up to make it harder. Fact-check the machine against real sources. Find the bias in an AI-written one-sided argument. Spot the fabricated citation the machine slipped in. The protocol in section 6.

IN PRACTICE Requires the instruction in section 5, and real sources for them to check against. Without those it is just unrestricted use with a better name.

Co-authored

Judging the output is the assessed skill — the student defends which claims they kept and which they rejected, against explicit source-analysis criteria.

IN PRACTICE Only above the threshold you named in section 3, and realistically only in your upper courses. A student who cannot yet tell a real citation from an invented one cannot exercise judgment over the machine's work.

05 MAKE IT YOUR VOICE

The reason line — the sentence that goes in your syllabus under the position paragraph. Adapt the wording; keep the logic, which is that the thinking is the thing you are grading:

I need to see your thinking — which sources you trusted, how you weighed them, the argument you built. That is what I teach and what I grade. A polished essay that isn't yours takes that away from both of us, and an argument you didn't construct teaches you nothing, even when it's right — and a source you never checked may not be real.

The instructions that matter, by skill. Each one stops the tool completing a different step. Give it before anything else in the conversation — and if you set up a student-facing tool, build it in rather than trusting a student to type it.

**ONE INSTRUCTION PER SKILL — DRAFTS, MARK THEM UP**
Skill	Give it this first
Summarizing a source	Don't summarize this source for me. Ask me what I think the author's angle is, then push on anything I miss — but let me do the reading.
Building an argument	Give me the strongest counterargument to my thesis and the best evidence against me. Do not write my essay or decide which side is right.
Evaluating a source	Help me weigh this source — who made it, when, and why — by asking me questions. Don't tell me the conclusion about whether to trust it.
Research	Only point me to sources with links I can open and verify myself. If you can't give me a real link, don't cite it — say you don't have one.
Perspective-taking	Argue this from [historical actor]'s point of view so I can test my own reading against it. Stay in that perspective; don't tell me which view is “correct.”

These are drafts and I do not teach your subject. Where the wording is wrong, you are right and I am not — send me corrections and I will fix the page.

06 DEVELOP YOUR JUDGMENT

The one task that turns the machine's biggest weakness into your lesson is the fact-check and bias-hunt — and it works precisely because the machine fabricates so confidently:

You generate an AI historical argument on the unit's question — one that sounds authoritative and cites its sources.
Students check every factual claim and every citation against real sources you provide or they can verify.
Students identify whose perspective is missing and what the argument quietly leaves out.
Students write which claims held up, which were wrong or invented, and where the argument was one-sided.

Step four is the assignment. The machine stops being the source of the answer and becomes the thing they interrogate, and you get a written artefact showing exactly what they caught — better diagnostic information about their sourcing instincts than an essay would have given you.

Now the catch, and it is the whole game. A student who cannot yet tell a real citation from a fabricated one will “verify” the machine by asking the same machine, and come away more confident in the invented version. Somewhere in your sequence a student becomes able to say “that quote is attributed to Lincoln but I can't find it in any real source” — and that is the point where this flips from harmful to genuinely useful. Go back to the threshold you named in section 3: everything here is an upper-level task if your answer was upper level. Below the line, the same instinct is better served by working from a curated set of real sources you hand them — same skill of weighing and doubting, no machine authority in the room.

07 WHAT IS NEWLY POSSIBLE

The bar: not faster, not easier — previously impossible. If it’s merely quicker, it belongs earlier, under everyday use — not here.

The same event told from two national perspectives, on demand. A student can hold genuinely competing framings side by side and interrogate both — a comparison that used to require archives and languages they’d never reach.

08 THE ASSISTANT — YOUR OWN MATERIALS

The uses that help you most are lower risk than anything involving students, and they are almost all materials. Four to start with — and read the caution under them, because it applies to every one:

Here's a primary source I use. Give me a guiding-questions handout at three levels — one for students who are struggling, one at level, one that stretches them. Same source, same questions of reliability and perspective.

Write a deliberately one-sided op-ed arguing [position] on [issue] — the kind a student should be able to pick apart. Make the bias real but not cartoonish.

Here's the source and topic. Give me six discussion questions that get at perspective and reliability, not recall.

Explain [event or concept] to a 15-year-old in five sentences, no jargon — then list every factual claim you made so I can check each one before I use it.

The caution that governs all of these: the machine fabricates quotes and citations and will misattribute them to real historical figures with complete confidence. Verify every source, date, and quotation before it reaches a student, and never present an AI-invented quote as something a real person actually said. That last prompt — make it list its own claims — is there precisely so you can catch what it made up. None of this touches student work.

09 KEEP IT YOURS

Verify everything — fabrication is the failure mode of this subject. The machine invents citations, quotes, and confident facts, and attributes them to real figures. Check every source and quotation before it reaches a student, and teach students that a citation from AI is a claim to be verified, not a source to be trusted.
Write your assignments assuming it happens in the room. The teacher plans run on a licensed workspace built for your own prep and materials — students cannot be enrolled, and no student work or data should go into a personal AI tool without clearing it first. Until a student platform is settled, any AI that is part of an assignment should happen in class with you running it. That is also the only equitable version: access at home runs from paid accounts to nothing, and an assignment requiring AI outside class quietly grades household income.
Contested fails silently. A tool told to “help with research” that has not been configured will hand over conclusions and invented sources the moment a student asks, which is precisely when it matters. An assignment labelled Contested without the setup is unrestricted use with a better name, and it looks fine until you read the outputs. Bring me one and we will set it up properly.
Don't run the fact-check task below your threshold. It is the single most likely way this goes wrong in your department, because the task looks equally good on paper at every level — and a student who cannot spot a fabrication will only certify it.
“Be ready to walk me through your sources” carries more weight here than a prohibition does. Ninety seconds of a student explaining which sources they used and why tells you everything, doubles as the sourcing practice you want anyway, and requires no accusation. Ask it of everyone, always — if it only appears when you are suspicious, it is an accusation mechanism rather than an integrity practice, and students will read it that way immediately.

10 FOR YOUR RECORD

Nothing to submit. The one thing worth writing down is your answer to the threshold question, because it is the piece your department will need when someone proposes a task and nobody can say whether it belongs in year two or year four.

Bring me one actual assignment and we will build it out together — pick the position, write the instruction line, set up the tool configuration if it needs one. Twenty minutes, and more useful than another document. I am glad to send citations for anything here.

The tool that matters here reads and reasons for the student, and it invents sources when it runs short — so the whole game is one setting: make it hand back the source-work and the arguing, not the answer. Everything else is your sequence and your judgment, both of which you already have.

Jason Beyer · Educational Technology · Regis Jesuit High School · A.M.D.G.

HelpDesk@regisjesuit.com · 303.269.8080 · LAST VERIFIED — AUGUST 2026