In 2026, talking aboutformative assessmentin thehumanitiesmeans going beyond “after-the-fact” correction and using light, continuous, interpretable evidence to steer teaching while thinking is being built.educational artificial intelligencecan help identify patterns that slip by in day-to-day practice (revision time, argumentative quality, source coherence), without turning interpretive disciplines into quizzes. In this scenario, tools likeStudierAIcan support reading the signals and designing timely micro-interventions, keeping the teacher at the center of decisions.
Why predictive formative assessment changes teaching in the humanities
Thepredictive assessmentapplied to formative assessment is not meant to “guess a grade,” but to estimate the likelihood that a student will encounter specific obstacles before they become failure. In the humanities, learning is often non-linear: an interpretation can be brilliant yet rest on a shaky quotation; a synthesis can be correct yet lack critical perspective. Prediction, if well designed, helps catch these imbalances in time.
The key point is not to reduce complexity: good predictive formative assessment works onprocess signals(how a draft evolves, how well an argument holds up, how a text is revised) and not only on right/wrong answers. In history, philosophy, literature, and the social sciences, this makes it possible to protect critical thinking: the goal is not to standardize, but to make visible what often remains implicit.
Which data to observe (and how to interpret it) in a humanities context
In a humanities context, useful data are oftennon-invasiveand already present in teaching practices: successive versions of an assignment, revision traces, rubrics, brief metacognitive reflections, participation in guided discussions. AI makes sense when it helps connect these signals to testable instructional hypotheses, not when it “surveils.”
- Progress in drafts: do clarity, structure, thesis, and counterarguments improve? “Cosmetic-only” progress can signal conceptual difficulty.
- Quality of arguments: presence of an explicit thesis, relevant evidence, logical links, handling of objections. Recurring weaknesses indicate misconceptions or a lack of rhetorical tools.
- Citation consistency and use of sources: correct quotations, contextualization, distinction between source and interpretation. Systematic errors may stem from superficial reading or low information literacy.
- Participation and quality of contributions: not only “how much they speak,” but how they connect concepts, ask questions, reframe others’ ideas. Consistent silence may indicate insecurity or poor accessibility of materials.
- Revision time and submission patterns: last-minute revision bursts or lack of iterations can signal planning difficulties or poor understanding of rubric expectations.
Interpreting these indicators requires caution: a long time is not always “risk” (it may be deeper work), and low participation may depend on classroom dynamics. For this reason, prediction should be read as aninstructional hypothesisto be verified through qualitative observations, brief conferences, and authentic tasks.
How to integrate predictive models into the formative assessment cycle: a practical workflow
Effective integration does not start with the model, but with theformative assessment cycle. Below is a replicable workflow, suitable for humanities learning units (short essay, guided commentary, argumentative presentation, debate, source dossier).
- 1) Define observable objectives: e.g., “support a thesis with at least two relevant pieces of evidence and one counterargument.”
- 2) Build essential rubrics: 3–5 criteria, clear descriptors, examples of levels. The rubric is what makes the prediction interpretable.
- 3) Collect light and frequent evidence: micro-writing (150–200 words), concept maps, interpretive exit tickets, annotations on sources.
- 4) Estimate risk in a targeted way: not “risk of failing,” but risk by criteria (e.g., evidence coherence, use of sources, structure).
- 5) Design micro-interventions: 10–15 minutes, highly specific. Examples: mini-lesson on “evidence vs opinion,” a bank of argumentative connectors, paraphrase practice with inference checking.
- 6) Check impact: repeat a similar micro-task and compare rubric criteria. If it doesn’t improve, change the lever (materials, prompt, examples, timing).
Concrete example (literature): before an essay on Leopardi, collect a micro-writing on “historical vs cosmic pessimism” with a commented quotation. If signals of confusion between concept and paraphrase emerge, the intervention can be a commentary grid (context, lexis, rhetorical figure, interpretation) and a short activity comparing two possible readings. Prediction is used to choose the right intervention, not to replace the teacher’s reading.
StudierAI in the classroom: support for predicting difficulties and personalization
In practice,StudierAIcan become an ally in making classroom complexity manageable when working with open-ended, interpretive assignments. The idea is not to automate assessment, but to support three actions: recognize patterns, suggest interventions, monitor development over time.
Some instructionally sound uses: identifying students who show sudden drops in argumentative coherence; flagging the most fragile rubric criteria at the group level; proposingscaffolds(guided outlines, model paragraph examples, checklists) andSocratic questionsto bring out assumptions and logical links. It can also help generate targeted activities: micro-debates on a controversial claim, exercises comparing interpretations, catch-up pathways on disciplinary vocabulary and source methodology. If you want to try it out, you canstart for freeand build a first routine of evidence and feedback in a few weeks.
An often underestimated aspect is longitudinal monitoring: seeing how theses, evidence, citation style, and revision change over time makes it possible to distinguish a temporary stumble from a structural difficulty. In any case, the decision remains with the teacher: AI offers signals and options, not verdicts. To understand the project’s approach and principles, it may be useful to consultwho we are.
Risks, bias, and responsibility: guidelines for ethical and transparent use
Predictive assessment is powerful and, for that very reason, requires clear rules. The main risks are:privacyviolations, use of data without informed consent, poor explainability,cultural and linguistic biases(for example, penalizing non-standard registers or multilingual students), and the risk of labeling (“you’re at risk, therefore…”). In the humanities, where identity and language matter, these aspects are central.
Operational best practices to maintain equity and instructional autonomy:
- Human-in-the-loop: every prediction must be reviewed by the teacher and accompanied by observable evidence (which criteria, which signals).
- Data minimization: collect only what is needed for the instructional task; avoid unnecessary sensitive data and define retention periods.
- Transparency and consent: explain to students and families what is being observed, why, and how it will be used; clarify that it is not an “automatic report card.”
- Audit of rubrics and criteria: periodically check whether some groups are systematically penalized; revise prompts and examples for linguistic inclusivity.
- Preventing labeling: use predictions to activate supports, not to lock in expectations; communicate in terms of “next steps” and strategies.
If these conditions are met, predictive formative assessment becomes an enhancement of teacher professionalism: more time to read deeply, more timely interventions, greater coherence among objectives, evidence, and feedback. It is a way to protect the complexity of the humanities, making the steps of reasoning visible and supporting those at risk of getting lost along the way. To experiment gradually, you can alsosign up for freeand start with a single unit, an essential rubric, and a few well-chosen indicators.
