TL;DR: Teach source verification by asking learners to investigate a bounded historical claim using actual documents, then edit an AI-assisted account until every material sentence has a defensible basis. An archive investigation studio makes provenance, ambiguity, and correction visible. The proposed course below uses document stations, editorial conferences, and a public explanation of unresolved questions. It is a teaching design, not a tested intervention, and it does not assume that a citation or a primary source automatically makes a claim true.
A classroom starts with one sentence on the wall
Put a sentence at the front of the room: “The opening of the new railway transformed the town immediately.” Do not present it as a fact about a particular place. It is an invented claim for discussing the studio's method. Its apparent simplicity is the lesson. What counts as transformation? When was the opening? What would immediately mean? Which residents would be included in the town?
A learner can produce a fluent paragraph around that sentence without answering any of those questions. Another can attach a relevant-looking source while leaving the causal claim untouched. A third can discover that a newspaper praised the railway and assume that the praise proves the transformation. Each response demonstrates why verification requires more than locating a link.
The studio treats the sentence as a proposal to be investigated. Learners must discover which parts can be supported, which require qualification, and which should be removed. The intended audience is adult learners in writing, research, journalism, humanities, or technology courses who increasingly work with AI-generated drafts. They do not need programming experience to participate.
The National Archives provides document-analysis materials covering written records, photographs, maps, recordings, and other forms. Its teaching sequence moves from observation and interpretation toward using documents as historical evidence. That sequence supports the studio's starting point; the course design itself is our proposal. National Archives document analysis.
A historical investigation has a practical advantage over a live breaking-news exercise: the teacher can prepare a stable, manageable collection. Stability does not eliminate uncertainty. It gives learners enough time to encounter uncertainty carefully, without a changing news cycle constantly replacing the question they are trying to answer.
Before learners arrive: assemble a small collection with friction
Choose a question with several plausible interpretations and enough accessible documents to test them. A local infrastructure opening, a public exhibition, a scientific announcement, or a dispute over a civic project can work. Avoid a question that requires specialist languages, inaccessible subscriptions, or large amounts of prior knowledge unless those are explicitly part of the course.
Prepare a modest collection rather than an exhaustive archive. Include documents that do different kinds of work: an announcement, a dated image, an official record, a later recollection, and a source that complicates the attractive story. The exact number depends on class time. Each item needs a clear origin, an accessible copy, and enough contextual information that learners can ask what they are looking at.
Do not manufacture contradictions by falsifying real documents. If a simplified source is necessary, label it as an instructional reconstruction and keep it separate from the authentic collection. Learners should never have to wonder whether the teacher secretly altered the evidence in a course about evidential trust.
The friction should come from interpretation. An opening announcement can establish what organizers said would happen. It cannot by itself establish what happened to residents over the following years. A photograph can show a crowd at one moment. It does not count every person in town. A later memoir may illuminate experience while raising questions about distance in time and selective recollection.
Prepare accessibility alongside intellectual challenge. Supply readable text for scanned documents where possible, descriptions of visual material, and a route for students who cannot access a particular medium. If transcription is uncertain, mark that uncertainty. An accessible version should help learners inspect evidence rather than silently substitute an unverified interpretation for the original.
The entrance activity: read the claim before reading the answer
Ask learners to break the wall sentence into smaller assertions. The railway opened on a particular date. Some measurable or describable change followed. The change affected the town in a meaningful way. The interval was short. The railway was responsible for the change, rather than merely preceding it. These are different claims even though one sentence has compressed them together.
Have learners write what evidence they would need for each assertion before searching. This prevents the first available document from defining the entire investigation. A source that establishes an opening date may be sufficient for one part and irrelevant to another. Starting with evidence needs makes that distinction visible.
Let students propose alternative wording. “The railway's promoters promised rapid change” may require evidence different from “the railway rapidly changed the town.” The first describes a claim made by historical actors. The second endorses a historical conclusion. They may look similar in a draft, yet they ask the writer to take very different responsibility.
Only then introduce an AI-generated account, if the course uses one. The teacher should prepare or generate it transparently and label its origin. It is a draft to inspect, not a model answer. If the draft contains invented sources or unsupported details, identify it as a deliberately selected teaching example after learners have investigated the specific problem; do not imply that all systems behave identically.
Students can also begin with a human-written paragraph. The essential skill is evidential reading. Restricting the course to AI errors could teach the misleading lesson that provenance matters only when a machine wrote the sentence. The studio should make the same demands of human and AI-assisted prose.
Station one: the document as an object
At the first station, learners describe the document before explaining it. Who created it? In what form? For whom? What date is visible? Does the copy show the entire item or only an excerpt? Which parts are legible? What information comes from the item itself, and what comes from the catalog entry?
This work can feel slow to students accustomed to summarizing immediately. Explain the purpose with a comparison: a document's identity is part of its meaning. A newspaper advertisement and an editorial may both praise the same project, but they occupy different roles. Treating them as interchangeable quotations erases information that the reader needs.
Ask students to keep observation distinct from inference. “The photograph shows a platform and a crowd” is an observation. “Every resident supported the railway” is an inference the photograph cannot establish on its own. The exercise is not a demand to eliminate interpretation. It is a way of making the path from document to interpretation visible enough to discuss.
A useful station output is a short document note written for another learner. It should explain what the item is, what it can plausibly establish, and what it cannot establish. Avoid long summaries that merely repeat the document. The note is valuable because it limits later misuse as well as enabling appropriate use.
Invite a second learner to challenge the note. Can they find an assumption disguised as a description? Does the note treat a missing date as if it were confirmed? Can they understand which information came from the archive's metadata? The exchange turns observation into a shared practice instead of a private impression.
Station two: establish a timeline without inventing a story
At the second station, learners place the documents in temporal order. Separate the date of an event from the date someone described it and from the date the archive digitized the item. These dates answer different questions. A modern upload of an old letter does not make the letter a recent account, and a later memoir is not contemporaneous evidence simply because it mentions the opening day.
Have learners mark gaps. If the collection contains an announcement and a report from several years later, they cannot fill the interval with an imagined continuous transformation. They can say the surviving packet does not establish the speed of change. Naming that limit is a substantive result, not an incomplete assignment.
The timeline should also show changes in the claim. Perhaps a plan referred to one route while the completed railway followed another. A writer who merges the plan and the result into a single event can create a false account without inventing any individual quotation. The error lies in combining true fragments under the wrong relationship.
AI assistance can be useful here as a formatting aid. Learners may ask it to organize dates they have already extracted. They should retain the extracted notes and check the resulting sequence against them. The learning goal is not typing efficiency. It is understanding what each date refers to and resisting a narrative that outruns the available chronology.
Finish the station with a question that remains open. For example, what additional record would help establish whether the economic change began before the opening? The question should identify a missing comparison or observation rather than vaguely request “more research.” Specific missing evidence gives the next investigation a direction.
Station three: a source can speak and still be interested
At the third station, consider why each item was produced. Organizers may have wanted support. A public office may have needed an administrative record. A resident may have wanted to preserve a personal memory. These purposes do not automatically make the documents unreliable. They affect what can reasonably be inferred from them.
Ask students to rewrite a sentence so attribution does the necessary work. Instead of “the project was a universal success,” write that a named source described it as successful, provided the document supports that wording. The revision does not settle the historical question; it accurately reports what the source says while keeping the writer's endorsement separate.
Then ask whether attribution is sufficient. A paragraph containing only opposing claims may remain unhelpful if the reader cannot understand their evidential status. Learners should explain whether a source observed an outcome, predicted it, reported another person's statement, or defended a position. A list of voices is not yet an investigation.
Use disagreement to clarify questions, not merely to produce balance. If two documents address different neighborhoods, they may not contradict each other. If one describes a proposal and another a completed project, the apparent disagreement may disappear once time is restored. If they do conflict directly, learners should say what the collection permits them to resolve.
The teacher can model a restrained conclusion: “This packet supports the date and the organizers' expectation, but not a town-wide claim of immediate economic transformation.” That sentence demonstrates a valuable achievement. The learner has narrowed the claim to match the evidence without pretending that uncertainty is a personal failure.
The editorial conference: argue for a sentence
Bring small groups together with one revised paragraph each. A group reads the paragraph aloud, then identifies the sentence most likely to be challenged. Another group asks what supports it. The author responds using the document notes, not general confidence or the authority of the AI tool.
Keep the conference focused on the relation between words and evidence. “Rapid,” “major,” “everyone,” “caused,” and “first” deserve attention because they often introduce comparison, scale, or exclusivity. An attractive adjective can quietly demand more evidence than the writer collected. Removing it may improve accuracy more than adding another citation.
Allow students to defend a qualified interpretation. They need not reduce every paragraph to a catalog of obvious facts. An argument can be warranted while remaining contestable. The writer should show the evidence, explain the inference, acknowledge a relevant alternative, and avoid presenting the conclusion as more certain than the collection permits.
The instructor should intervene when the discussion turns into a contest of personalities. A confident reader does not have stronger evidence merely because they speak quickly. Give quieter learners time to point to a document or submit a written challenge. The conference is a method for inspecting arguments, not a performance of certainty.
Make revision part of the conference rather than homework postponed to the end. Students should change the sentence while the objection is clear. Then ask the challenger whether the new wording answers the concern. This exchange teaches that correction can be a normal part of inquiry, rather than a punishment after someone is caught being wrong.
Where an always-on assistant belongs in this studio
An assistant can help organize a list of sources, draft alternative wording, or identify questions to investigate. It should not silently become the authority that determines what a document establishes. The studio's participants remain responsible for inspecting the materials they use and for deciding which claims belong in their account.
The October 2026 Dot documentation describes background work and contextual memory. For this teaching design, that means an instructor should specify what a delegated research task is allowed to collect and how its output will be reviewed. The article does not claim that Dots have an archive-specific feature or that background activity improves learning. OpenAI's Dot documentation.
If learners use an ongoing assistant, have them preserve a clear record of the task they assigned and the material they accepted. A useful record states the question, the sources actually opened, and the changes made after inspection. A transcript alone may be too long to reveal these choices, while a final paragraph alone conceals them.
Do not require students to connect private email, personal cloud storage, or accounts unrelated to the collection. The exercise can operate with a teacher-supplied packet and public sources. Students should have an equivalent route when they cannot or do not want to use a paid or persistent assistant. Tool access should not become an accidental prerequisite for learning how evidence works.
UNESCO's guidance argues for human-centered educational uses and attention to data privacy and pedagogical validation. That establishes a relevant policy context; it does not endorse this particular studio or prove its effects. UNESCO's generative AI guidance.
The correction note is part of the publication
Each learner finishes with a short account and a correction note. The account addresses the chosen historical question. The note explains which material claim changed during the investigation, what prompted the change, and what remains unresolved. Together they show both a result and the reasoning that improved it.
A correction note should not become an apology essay. “I removed the claim of immediate change because the sources established an opening date but did not document the interval's economic effects” is sufficient. It names the evidential mismatch and the editorial action. Readers can understand the improvement without being asked to evaluate the student's character.
Permit a learner to conclude that the original question cannot be answered from the packet. Require a precise explanation of why, including what additional evidence would matter. This option prevents the assignment from rewarding invented closure. It also resembles a real research decision: sometimes the available evidence supports a narrower publication than the original commission.
If work is shared publicly, decide together what will be included. A classroom exercise should not expose a student's private prompts or unrelated personal data merely to demonstrate transparency. The relevant disclosure concerns how the account was produced and supported. The instructor should specify the publication format before learners begin, so consent is part of the course design.
The final presentation can be an annotated wall display, a short audio explanation with a transcript, or a readable article. Different media introduce different editing choices. In every format, the audience must be able to distinguish the historical material, the learner's interpretation, and the limits of the investigation.
A small publication bundle is enough. Each item serves a different reader need:
- The account: the revised answer to the historical question, written for an interested nonspecialist.
- The document notes: the provenance and limitations of the sources actually used, allowing another learner to inspect the argument.
- The correction note: the consequential change in wording and the evidence that prompted it.
- The unresolved question: a specific missing observation or comparison, giving a future investigator a useful starting point.
These items should not repeat one another. The bundle makes an argument readable while keeping its evidential path available.
What the instructor learns from the room
Watch for where students become uncertain. If they can identify a document but cannot connect it to a sentence, the next lesson should focus on inference. If they quote accurately but confuse dates, return to chronology. If they understand the limitations privately yet remove them when writing for an audience, discuss the pressure to sound conclusive.
Collect examples of revision with permission and without presenting individual students as cautionary tales. The examples can reveal recurring instructional needs. A changed verb may demonstrate improved causal reasoning. A new attribution may demonstrate a distinction between testimony and endorsement. A deleted sentence may demonstrate more learning than a longer final answer.
Ask students what they would inspect first in a fresh document set. Their answer can show whether the method transfers beyond the original packet. Be careful about interpreting a confident answer as evidence of lasting learning. The studio is a proposed format; its effects would need appropriate evaluation in the actual teaching context.
Keep the collection modest enough that reading remains central. If students spend most of the session navigating broken links, setting up accounts, or troubleshooting software, the course has drifted away from its intended question. Tool operation may deserve its own lesson, but it should not consume the time reserved for evidential judgment.
A studio succeeds as a teaching design when it gives students repeated opportunities to notice, explain, and revise a claim's relationship to its sources. That is a practical goal for an instructor to observe. It is more defensible than promising that one workshop will make learners immune to misinformation or permanently independent of AI assistance.
A second investigation changes the conditions
For a follow-up session, give learners a different medium rather than a larger version of the same assignment. If the first packet emphasized written announcements, the second might center on photographs and maps. The change forces students to reconsider what a source can establish instead of repeating a familiar sequence mechanically.
Reduce scaffolding gradually. Students who have practiced document notes can decide which observations matter for the new question. They should still explain uncertainty and preserve provenance. Removing the worksheet should not mean removing the reasoning; it means asking whether learners can perform it with less external structure.
Introduce one source discovered by a student, then inspect its suitability together. Does it belong to the relevant period? Is it a copy of another item already in the packet? Does it add evidence or merely repeat a later account? This extension teaches selection without turning the class into an unrestricted search contest.
Ask learners to identify the least useful source for the specific claim and explain why. This is different from declaring the source generally worthless. A valuable historical object may contribute little to a particular question. Learning to leave a source out is part of research judgment, especially when AI assistance makes collecting more material easy.
End by returning to the wall sentence. The sentence may now be narrower, more conditional, or replaced by a different question. That visible change is the point of the studio. A fluent draft has become an accountable account, and the learners can explain the distance between the two.
Preparing the studio for different teaching conditions
A short workshop can investigate a single paragraph with a supplied packet. A longer seminar can let learners choose a question after exploring several collections. These are different courses, and the instructor should resist compressing the seminar's ambitions into the workshop's time. An incomplete but carefully examined claim can teach more than a rushed survey of many documents.
In an online class, pair learners around a shared document rather than asking everyone to browse independently throughout the session. Have one learner describe an observation and the other identify its basis in the item. Rotate the roles. This gives the teacher something concrete to observe and reduces the chance that quiet participation becomes invisible. It also keeps the discussion grounded when screen sharing makes a document harder to read.
For learners writing in a second language, separate difficulties expressing an argument from difficulties understanding its evidence. A student may recognize a causal overclaim yet struggle to find a concise English replacement. Offer examples of qualified wording and allow a first explanation in a language the learner can use comfortably where teaching resources permit. The objective is to improve evidential judgment, not to mistake linguistic speed for research competence.
The teacher should prepare a recovery path for absent or inaccessible sources. A verified local copy with provenance may preserve the exercise where copying is permitted. Otherwise use another item that serves the same instructional purpose and tell learners what changed. Replacing a source silently would undermine the very habit of transparent revision the studio is trying to establish.
Finally, reserve time for the correction note. When schedules tighten, it is tempting to keep the polished presentation and drop the reflection on revision. That would remove the activity that most clearly exposes why the final account deserves more trust than its first draft.
Sources
- National Archives: Document Analysis, source-analysis teaching materials.
- UNESCO: Guidance for generative AI in education and research, human-centered educational guidance.
- OpenAI: Getting started with your dot, product context checked October 1, 2026.
The railway claim, classroom packet, activities, and course sequence are illustrative proposals. No outcomes or classroom measurements are being reported.