# AI Tutoring That Improves Outcomes, Not Just Engagement > Higher Education · AI Course · HE-4 > Source: https://ibl.ai/solutions/higher-education/course/ai-tutoring-that-improves-outcomes > Last updated: 2026-08-25 **Design a tutoring agent that produces measurable learning gains — Socratic scaffolding, answer-withholding, misconception detection, and honest outcome measurement.** ## The Short Answer **A tutor that gives answers raises satisfaction and lowers learning. ibl.ai builds tutoring agents that withhold answers, diagnose the misconception rather than the wrong answer, and ground in your own course materials — deployed inside your perimeter where you own all the code and the data, so student work never trains an external model.** On ibl.ai you own all the code and the data, run it model-agnostic across any LLM, and pay with no per-seat pricing — so you can deploy anywhere, from your own cloud to a fully air-gapped network. [Request Access](https://ibl.ai/contact) · [Explore Higher Education](https://ibl.ai/solutions/higher-education) ## Course facts - **Level:** Intermediate - **Duration:** 6 hours across 8 modules - **Format:** Cohort workshop with a build lab - **Modules:** 8 - **Catalog code:** HE-4 - **Frameworks covered:** FERPA, WCAG 2.2, NIST AI RMF ## What is this course about? Most AI tutors optimize for the wrong thing. A tutor that answers quickly produces high satisfaction and low learning, and the two metrics move in opposite directions. This course covers the design decisions that separate a tutor from an answer service — scaffolding, withholding, misconception diagnosis — and the measurement discipline needed to prove a learning effect rather than an engagement one. ## Who is this course for? - Directors of tutoring and learning centers - Faculty developing course-embedded AI support - Instructional designers - Institutional research staff evaluating learning interventions ### What do I need before starting? - Familiarity with one course's learning objectives and common student errors - No technical background required ## What will I be able to do afterwards? - Explain why answer-giving tutors improve satisfaction and depress learning - Implement Socratic scaffolding as a system-level policy rather than a prompt trick - Design misconception detection that diagnoses the mental model, not the answer - Ground a tutor in your own course materials so it teaches your curriculum - Design an outcome evaluation that would survive institutional research review ## What does each module cover? ### Module 1 — Why do students rate answer-giving tutors highest? The core tension of the course: the metric that is easiest to move is the one least connected to learning. _(40 min)_ **Objectives** - Describe the divergence between satisfaction and learning gain - Identify the design choices that trade one for the other - Set an explicit position on the trade before building anything **Topics:** Satisfaction–learning divergence · Desirable difficulty · Metric selection · Stakeholder expectation setting **Activity:** Compare transcripts from an answer-giving and a withholding tutor and predict each one's ratings. ### Module 2 — How do you implement Socratic scaffolding that holds up? Scaffolding as an enforced policy, and the failure modes that appear when a determined student pushes. _(55 min)_ **Objectives** - Specify a scaffolding policy at the system level - Anticipate and test the ways students defeat withholding - Calibrate hint progression to the learner's demonstrated state **Topics:** System-level policy · Hint laddering · Withholding defeat patterns · Adaptive calibration **Activity:** Red-team your own tutor by trying to extract a direct answer, then patch what worked. ### Module 3 — How do you diagnose the misconception rather than the error? The distinguishing capability of a real tutor: identifying the wrong mental model behind a wrong answer. _(55 min)_ **Objectives** - Build a misconception catalog for one topic - Design diagnostic questions that discriminate between misconceptions - Route remediation by diagnosed model rather than by error **Topics:** Misconception cataloging · Diagnostic discrimination · Remediation routing · Common error taxonomies **Activity:** Build a misconception catalog for one topic you teach and encode it into the tutor. ### Module 4 — How do you make the tutor teach your curriculum? Grounding in course materials so notation, method, and sequence match what the instructor actually taught. _(50 min)_ **Objectives** - Ground a tutor in course materials and instructor notation - Handle divergence between the textbook and the internet's default method - Keep grounding current as the course evolves **Topics:** Course material grounding · Notation consistency · Method divergence · Term-over-term refresh **Activity:** Ground the tutor in one unit's materials and test it on problems using your notation. ### Module 5 — How do you measure learning rather than usage? Evaluation design that produces evidence rather than a dashboard. _(50 min)_ **Objectives** - Design a pre/post assessment tied to learning objectives - Construct a defensible comparison condition - Pre-register the analysis to prevent post-hoc metric selection **Topics:** Pre/post design · Comparison conditions · Pre-registration · Effect size interpretation **Activity:** Draft an evaluation protocol and submit it to institutional research for critique. ### Module 6 — Where is the line between tutoring and doing the assignment? Integrity boundaries built into the tutor rather than left to the student's judgment. _(45 min)_ **Objectives** - Define the integrity boundary for specific assignment types - Implement assignment-aware behavior changes - Design disclosure so instructors know how the tutor was used **Topics:** Assignment-aware policy · Graded versus practice work · Usage disclosure · Instructor visibility **Activity:** Write the integrity policy for three assignment types and encode the tutor's behavior for each. ### Module 7 — How do you make the tutor usable by every student? Accessibility and multilingual support treated as core requirements rather than later additions. _(45 min)_ **Objectives** - Apply WCAG conformance to a conversational tutoring interface - Support multilingual learners without degrading precision - Test with assistive technology and real users **Topics:** WCAG in chat interfaces · Screen reader behavior in streaming output · Multilingual precision · Assistive technology testing **Activity:** Run an accessibility audit of the tutor with a screen reader and log every failure. ### Module 8 — Building a course-grounded tutor with a withholding policy The hands-on module: deploying a tutor with configurable withholding and a misconception catalog. _(60 min)_ **Objectives** - Deploy a tutor grounded in real course materials - Configure the withholding policy per assignment type - Validate against the red-team and accessibility suites **Topics:** Tutor deployment · Withholding configuration · Validation suites · Pilot design **Activity:** Deploy the tutor for one unit and pilot it with five students, recording every place it over-helped. ## What is the capstone project? **Course-embedded tutor with a pre-registered evaluation.** Design and deploy a tutor for one real course unit, including the misconception catalog, withholding policy, integrity boundaries, accessibility conformance, and a pre-registered evaluation with a comparison condition. _Deliverable:_ A deployed tutor plus an evaluation protocol accepted by institutional research. ## How are learners assessed? - Red-team exercise — the learner's tutor must resist direct-answer extraction - Misconception catalog reviewed for diagnostic discrimination - Capstone evaluated on measurement rigor, not tutor polish ## What ships with the course? - **Facilitator guide.** Session-by-session running order, discussion prompts, and the questions that reliably derail a room. - **Learner workbook.** Exercises, checklists, and the templates each module's activity produces. - **Hands-on lab environment.** A sandboxed ibl.ai deployment so exercises run against real agents, not screenshots. - **Assessment bank.** Scenario questions and rubric criteria mapped to each stated learning outcome. - **Source bibliography.** Every primary regulation and standard cited on this page, linked and dated. ## Which AI agents does this course use? - [Tutoring Agent](https://ibl.ai/solutions/higher-education/agent/tutoring-agent) - [Faculty Agent](https://ibl.ai/solutions/higher-education/agent/faculty-agent) - [Student Services Agent](https://ibl.ai/solutions/higher-education/agent/student-services-agent) - [Retention Agent](https://ibl.ai/solutions/higher-education/agent/retention-agent) ## Where does the course material come from? Every module is grounded in primary sources — the regulation, standard, or research itself, not a summary of it. Each was resolved at authoring time. - [Office of Educational Technology](https://tech.ed.gov/) — U.S. Department of Education. Federal guidance on AI in teaching and learning, framing the design principles. - [AI and the Future of Teaching and Learning](https://www.ed.gov/sites/ed/files/documents/ai-report/ai-report.pdf) — U.S. Department of Education. The report's guidance on human-in-the-loop instructional design. - [AI Index Report](https://hai.stanford.edu/ai-index) — Stanford HAI. Baseline evidence on model capability and reliability in educational tasks. - [Web Content Accessibility Guidelines](https://www.w3.org/WAI/standards-guidelines/wcag/) — W3C. Conformance target for the accessibility work in Module 7. ## Delivery notes Binding guidance for anyone preparing and delivering this course: - The red-team exercise in Module 2 is the course's signature. Build a library of extraction attempts from real student behavior — flattery, false premises, claiming the deadline passed, role-play — and keep adding to it. - Module 3 needs a genuine misconception catalog from a real discipline. Physics force concepts and introductory statistics are both well documented; pick one and build it properly rather than sketching several. - Be honest in Module 1 that the evidence base for AI tutoring effect sizes is still thin. Do not manufacture a headline number — the course's credibility with institutional research depends on this. - The integrity module must not become a surveillance design session. Frame disclosure as instructor visibility into how the tool was used, not monitoring of the student. - Accessibility testing needs a real screen reader user in the room. A sighted facilitator tabbing through is not a substitute and should not be presented as one. ## Why run AI training on a platform you own? - **You own the course, not a licence to it.** Course content, learner data, and the platform run inside your perimeter — you own all the code and the data. - **Model-agnostic delivery.** Run the course's AI components on any LLM — Claude, GPT, Llama, Gemini, Command — and switch anytime. - **No per-seat training licences.** Usage-based or self-hosted, so cost tracks actual use rather than headcount. - **Deploy anywhere.** Cloud, private VPC, on-premise, or fully air-gapped — including for cohorts that cannot use public AI tools. ## Frequently asked questions ### What does the AI Tutoring That Improves Outcomes, Not Just Engagement course cover? Most AI tutors optimize for the wrong thing. A tutor that answers quickly produces high satisfaction and low learning, and the two metrics move in opposite directions. This course covers the design decisions that separate a tutor from an answer service — scaffolding, withholding, misconception diagnosis — and the measurement discipline needed to prove a learning effect rather than an engagement one. It runs 6 hours across 8 modules across 8 modules, at intermediate level, and closes with a capstone: Course-embedded tutor with a pre-registered evaluation. ### Who should take AI Tutoring That Improves Outcomes, Not Just Engagement? It is written for Directors of tutoring and learning centers, Faculty developing course-embedded AI support, Instructional designers, Institutional research staff evaluating learning interventions. Prerequisites: Familiarity with one course's learning objectives and common student errors; No technical background required. ### Can we run this course on our own infrastructure? Yes. ibl.ai is model-agnostic and deploy-anywhere — cloud, private VPC, on-premise, or fully air-gapped — and you own all the code and the data. Cohort data, submissions, and any material learners upload stay inside your perimeter, which matters for higher education teams that cannot send work to a public AI tool. ### How do we get access to AI Tutoring That Improves Outcomes, Not Just Engagement? Request access and we will set it up for your cohort — hosted by ibl.ai, or running against your own deployment. Tell us the group size and timing you need, and whether it should run inside your own perimeter. ### How much does AI training for higher education cost on ibl.ai? There is no per-seat pricing — you pay for usage or self-host and pay only for the infrastructure, so a 5,000-person rollout does not cost 5,000 licences. 1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University. ## More Higher Education courses - [FERPA-Compliant AI: Deploying Agents on Student Data](https://ibl.ai/solutions/higher-education/course/ferpa-compliant-ai): Run AI agents against your SIS and LMS without a vendor ever seeing a student record — the school official exception, vendor DPAs, and the architecture FERPA implies. - [AI Academic Advising at Scale: Design and Guardrails](https://ibl.ai/solutions/higher-education/course/ai-academic-advising-at-scale): Build an advising agent that handles degree audits and registration at 20,000-student scale without ever giving a student wrong graduation advice. - [Enrollment and Yield AI: Agents Across the Funnel](https://ibl.ai/solutions/higher-education/course/enrollment-and-yield-ai): Deploy AI across inquiry, application, admit, and melt — where agents lift yield, where they damage trust, and how to keep the funnel on infrastructure you own. - [Assessment Redesign for the AI Era](https://ibl.ai/solutions/higher-education/course/assessment-redesign-for-the-ai-era): Detection does not work. Rebuild assessment around what AI cannot fake — process, oral defense, local context, and in-class artifacts — with department-ready rubrics. - [Writing a Campus AI Policy That Survives Accreditation](https://ibl.ai/solutions/higher-education/course/campus-ai-policy-that-survives-accreditation): Draft institutional AI policy a regional accreditor, a general counsel, and a faculty senate will each accept — with the governance to keep it current. - [RAG on Institutional Knowledge: Catalogs, Policies, Handbooks](https://ibl.ai/solutions/higher-education/course/rag-on-institutional-knowledge): Retrieval-augmented generation over the documents a campus runs on — chunking a catalog, versioning policy, and stopping the agent citing a 2019 handbook.