Begin with one lesson or planning task

This playbook helps Mississippi teachers, instructional coaches, librarians, school leaders and district teams decide where AI can support a specific activity and what must remain the learner’s and educator’s work. Use it before introducing a tool, changing an assignment, reviewing AI-assisted material or responding to a concern.

Leave with an activity decision, clear student directions, a verification record and a named person who can pause or resolve the work. A first planning conversation can take about 30 minutes; district approval, accessibility testing and any required family process may take longer. Bring the current learning objective, adopted materials, district rules and approved-tool information.

1. Define: name the learning or planning problem and what a person must demonstrate.
2. Check: confirm authority, account, information, safety and access conditions.
3. Prepare: test the bounded task and verify material before student use.
4. Decide: retain evidence, communicate the rule and continue, revise, pause or stop.

MAIN provides an implementation method, not a district policy, product approval, curriculum mandate or certification. It complements MDE’s AI guidance and Digital Learning resources. Follow the applicable district, school, assessment and student-support requirements. MDE participation in a teacher pilot does not authorize every student, account or feature.

Put the decision with the right people

Teachers should not have to resolve contracts, consent, cybersecurity or disability rights alone. A product being available on a device is not evidence that the intended use is approved.

If rules conflict, pause the affected AI step and provide the approved learning path while the designated adults resolve it. Do not tell students to bypass filters, use personal accounts or take institutional conflicts home to solve. Use MAIN’s AI + Dual Enrollment Guide for that setting.

Complete the educator decision card

Use when: planning one activity. Completed by: its educator, with district owners resolving approval questions. Bring: the learning objective, current materials, tool approval and required supports. Copy this compact record into the school’s existing planning system.

• Activity and owner: [task, class/grade band, educator, date].
• Learning purpose: [what students must know, practice or explain themselves].
• AI role: [teacher preparation, supervised demonstration, bounded student assistance, or no AI].
• Conditions confirmed: [policy; exact feature/account/age; allowed data; accessibility/supports; family process where applicable].
• Evidence: [student work or explanation; output verification; source locations].
• Decision: [proceed within stated limits / revise / pause / do not use].
• Owner of an unresolved question: [name/role and contact route]; [next review trigger].

Complete when: the intended use and every essential condition are clear, the human learning task remains meaningful and the educator can verify the result. An unknown approval, privacy, safety or access condition means pause; a demonstration can use teacher-authored material without student accounts while the issue is resolved.

Classify the activity, not the product name

Teacher brainstorming with no student information is different from a student-facing tutor, analysis of student work or a connector that can read files. Higher reach or consequence requires stronger review. Begin without autonomous sending, posting, grading, roster changes or unrestricted browsing. A newly enabled AI feature or connection inside an approved platform still needs approval for that use.

State the AI rule for this assignment

Use the school’s actual policy language. The four levels below help translate it into steps; they do not grant permission that the district has not given.

Preserve required supports. Assignment AI directions do not remove approved IEP, Section 504 or other required accessibility supports. Speech-to-text support is not automatically authorship substitution. Resolve a conflict through the responsible team before the activity; classroom directions do not authorize an accommodation for a secure assessment. See IDEA’s assistive-technology requirement.

Assignment expectation builder and student directions

Use when: communicating a significant assignment. The teacher completes the bracketed fields and checks comprehension with an example of an allowed and a disallowed step.

“For [activity], AI is [level]. You must do [human work] yourself. You may use [approved tool/account or support] only for [steps]. Do not enter [restricted information] or use AI for [prohibited steps]. Check [facts, calculations or source locations]. Show [your reasoning/process evidence] and disclose [material help, using the school’s format]. Your required supports remain available through [route]. If you cannot access the approved path or the directions are unclear, ask [adult/contact] before that step.”

For younger learners, shorten this to the actual classroom action: “Today we will use only the material I show you. Find an answer in our source, explain your reason and ask me if something seems wrong. Do not type personal information into a chatbot.”

Complete when: a learner can identify the allowed step, their own work, the information boundary and where to ask for help. Disclosure does not make an unauthorized use acceptable.

Plan learning before generating material

Use when: designing or adapting a lesson. The teacher records the fields below; a curriculum or support specialist reviews changes that affect adopted expectations or required supports.

• Instructional basis: [current adopted standard wording and source/version], [district pacing or approved material], [prerequisite skill].
• Human performance: [what students will produce, explain, demonstrate or revise].
• Lesson sequence: [model], [guided practice], [independent evidence], [feedback], [time available].
• Optional AI contribution: [one bounded preparation or support step], [approved inputs], [verification method].
• Access plan: [required supports, language demands, device/connectivity assumptions and equivalent route].
• Release check: [reviewer/date], [changes made], [stop condition].

Supply the actual current standard and approved material. Do not rely on a model to recall the right Mississippi version or invent a standard code. Check cognitive demand, prerequisite knowledge and answer correctness. In early reading, preserve the adopted instructional sequence and decodability when that is the objective; a simpler-looking generated passage is not automatically better instruction.

AI literacy that changes classroom behavior

Explain that generative AI produces text, images, audio or code and can generate plausible errors or invented references. Reasoning, multimodal and connected systems can perform more steps, but a fluent explanation is not evidence of correctness. Students should practice matching claims to sources, testing a calculation, identifying an unsupported claim, protecting information and explaining whether they accepted or rejected assistance. Teach these within the subject, not as a test of recognizing an “AI writing style.”

Choose supervision and access for the class

These are planning prompts, not age-based permissions. Check actual product/account terms and district authorization. Do not require a personal consumer account merely because a learner is old enough to create one.

For the selected activity, record [supervision], [approved account or no-account route], [reading/language demand], [devices and connectivity], [required supports], [classroom sequence] and [stop trigger]. A shared display, printed sources, peer review or teacher feedback can preserve the learning goal when a tool is not appropriate. Do not make optional AI use a condition of a student’s grade.

Use five bounded educator workflows

Use approved educator-authored or licensed inputs, without student information unless the specific use is authorized. Prompts below are starting instructions, not safeguards that can override a tool’s access or retention behavior. The teacher verifies and approves every student-facing result.

1. Lesson options from supplied sources

“Using only [standard wording] and [approved material], suggest two lesson structures for [time/prerequisite]. Identify what students will do themselves and cite where each activity connects to the supplied material. Flag gaps rather than inventing content.” Check alignment, rigor, timing and accessibility before choosing.

2. Assignment clarity and AI boundary

“Review these educator-authored directions for unclear steps, prerequisites and AI permissions. Identify ambiguity; suggest a concrete correction without changing the learning goal.” Check that the result preserves the intended assessment, required supports and district rule.

3. Ungraded practice questions

“From [approved source], draft three practice questions, answers and exact supporting locations. Explain the skill each question checks. Do not add information absent from the source.” Independently solve each item, check distractors and remove ambiguity. Never supply secure or unreleased assessment content.

4. Family communication from verified facts

“Draft a plain-language message using only these approved facts: [purpose, activity, data conditions, local family process, supports and contact]. Flag missing facts; do not promise privacy, access or consent arrangements.” A qualified person checks translation and meaning before release; AI is not the final interpreter for consequential communication.

5. Instructional-material verification aid

“Compare this draft with [approved source]. List claims, calculations and references that need checking; identify possible missing perspectives and accessibility barriers.” Treat that list as a review aid. Verify against original sources, not another generated answer, and retain the educator’s corrections.

Material verification checklist

Use when: releasing any AI-assisted lesson material. Completed by: its educator; involve relevant specialists where needed. Record [source and location], [facts/calculations/answer key checked], [standard and cognitive demand], [licenses/quotations], [bias and inclusion], [accessibility/language], [corrections], [reviewer/date]. Release only when every consequential item is verified. If that is not feasible in the available time, use previously verified or teacher-created material.

Walk through three realistic classroom cases

Illustrative planning cases—not reported classroom trials. Replace the subject, standard, materials and timing with the school’s actual context. Each case preserves a route without a student AI account.

Elementary: teacher-prepared science practice

An educator supplies an approved short science passage and the current learning objective to an authorized teacher tool, asking for three ungraded questions with answer locations. Before class, the teacher solves the questions, checks the science and revises vocabulary. Only the checked material is printed or projected. Students explain an answer to a partner and point to evidence in the passage.

Decision: proceed as teacher preparation with no student account or student information. Stop: a question misstates the science, requires untaught background or excludes a learner. Replace it with a teacher-written question. Evidence: the reviewed questions/source locations, preparation plus checking time and students’ explanations.

Middle school: repair unsupported claims

The teacher prepares an AI-generated paragraph and two reliable short sources, screening all material before use. Students label claims supported, contradicted or not established, annotate source locations and write a corrected paragraph explaining one decision. Run it on paper or a shared display; only use direct student tool interaction if that account and task are approved.

Decision: assess evidence use and justified revision. Stop: harmful or out-of-scope output; the adult replaces it rather than extending an unsafe conversation. Evidence: annotations and the student’s correction, not whether they guessed which text “sounds like AI.”

High school: reasoning before AI feedback

Students first develop a historical claim from two teacher-approved sources and record their evidence. Where permitted, an approved tool may suggest a counterargument. Students check the challenge against the sources, explain whether they accept or reject it and write their own final response. A peer or teacher can supply the counterargument for the same learning task without an account.

Decision: limited assistance after initial reasoning; no generated final submission unless that separate role is explicitly taught and allowed. Stop: invented citations, restricted records or secure testing material. Evidence: the original claim, source locations and one explained revision. Apply the college instructor of record’s actual directions in a dual-enrollment course.

Approve tools, information and accessible participation

Use when: adopting a tool or changing its AI features, connected data or audience. Completed by: district technology/privacy/procurement and accessibility owners with the educator. Teachers supply the activity; designated owners answer the contract and legal questions.

• Approval: [exact product/feature, purpose, owner, record and review date].
• Account and age: [district-managed account, permitted users, supervision and applicable terms].
• Information: [minimum inputs; collection, retention, deletion, sharing and training conditions; approved data category].
• Permissions: [files/rosters/services reachable, minimum access, logs, disconnect/revocation and incident owner].
• Family process: [notice, inspection, consent or opt-out requirements that actually apply; responsible district reviewer].
• Accessibility: [keyboard/screen-reader/caption tests, language and device needs, required supports and effective participation route].
• Decision: [approved scope/conditions or pause], [renewal/change trigger].

Names, student work, voice, images, disability information, behavior, grades and free-text narratives can contain protected information. Prefer synthetic examples or less information. Removing names alone is not de-identification: indirect identifiers, small groups and identifying narrative can still identify a student. Ask the district privacy owner to make the appropriate determination.

District reviewers should check FERPA and the amended COPPA Rule for covered operators. The April 2025 amendments generally required compliance by April 22, 2026. School authorization for an educational purpose does not authorize unrelated commercial use. Confirm the actual consent basis, retention/security terms and applicable disclosure requirements; a teacher’s permission is not a substitute for the district process.

Sensitive questions: if a survey, analysis or evaluation asks students to disclose sensitive personal or family information, have the district privacy owner check PPRA and applicable notice, consent, inspection and opt-out requirements before use. Do not casually use a chatbot for mental-health, religious, political or family-income intake.

Accessible service, not a promise. Test actual tasks with required supports. Simplification and translation must preserve meaning and instructional expectations. An alternative route does not excuse inaccessible required services. Under DOJ’s current Title II rule, technical compliance dates are April 26, 2027 or April 26, 2028 depending on the applicable population category; existing ADA duties continue. School districts are not automatically special district governments. The district should apply DOJ’s school-district population method rather than guessing its deadline.

Complete when: the approved scope, data terms, supports, required family process and incident owner are documented and usable. Otherwise pause the affected use.

Give families accurate, useful information

Use when: explaining an approved student-facing activity. The educator and designated district owner complete and check the message. A notice is not automatically legally sufficient consent, and not every classroom AI use requires the same family process.

“Our class will [activity] to practice [learning goal]. Students will do [human work]. The approved arrangement is [teacher demonstration / supervised tool and account / no student account]. We will use [confirmed information] and will not enter [restricted information]. The educator will check material before use and supervise [steps]. Required supports and the participation route are [details]. The applicable district notice or consent process is [verified process, if required]. Please contact [school contact] with questions or an access concern.”

Do not promise “no data stored,” “fully private,” “bias-free” or “approved for all ages” without evidence for the exact configuration. Provide communication in an accessible format and needed language through the school’s established process. Record [district review], [message/version/date], [applicable response process] and [contact].

Review learning fairly and protect assessments

Use evidence that serves the learning objective: a draft, source annotation, worked calculation, demonstration, discussion, reflection or explained revision. Do not collect exhaustive process logs merely to create surveillance. Let students explain through an appropriate accessible format and follow required supports.

An AI detector score is not proof of authorship, permission or misconduct. Performance depends on the product, text, language, model and revision process. Review the communicated rule, reasonable alternative explanations, process evidence and student account, using the district’s established procedure. Distinguish unclear directions and teaching needs from intentional misconduct.

Academic-integrity evidence-review record

Use when: a concern requires review. Completed by: the teacher and designated reviewer under policy. Record [assignment/rule as communicated], [authorized supports], [specific observed evidence], [student explanation], [alternative explanations considered], [reviewer and policy route], [outcome/next instruction]. Retain it in the approved restricted system. Do not enter the student’s case in a public chatbot. Complete only through the established review and response process; no detector threshold replaces it.

Secure assessments have a separate boundary

Do not enter secure or unreleased assessment items, answer keys, testing tickets or responses from secure testing into any AI tool, including an otherwise approved one. Do not reword unreleased items to make practice. Follow the current assessment manual and school/district test coordinator process for suspected exposure. These are applications of MDE test-security requirements; a classroom AI rule does not change them.

Protect the student, then resolve the incident

AI must not make final grades, discipline, disability/eligibility/placement, accommodation, counseling, threat, health or student-safety decisions. Do not ask a chatbot to independently manage a student crisis. Stop an unsafe interaction and move responsibility to the appropriate adult and established process.

Immediate danger: call 911 and follow the school emergency plan. When there is reasonable cause to suspect child abuse or neglect, make the required immediate report to the Mississippi Department of Child Protection Services at 1-800-222-8000. An internal notification or AI review must not delay required reporting. Follow applicable written follow-up and school reporting procedures.

Harmful synthetic media: stop further exposure and use restricted safeguarding channels. Do not ask a student or teacher to download, forward, recreate or redistribute suspected child sexual abuse imagery. Record the location, time, platform and immediate actions; obtain safeguarding/law-enforcement instructions. Do not upload the material to an AI detector or chatbot.

Privacy or security exposure: stop the affected AI action or connection, notify the district incident owner promptly and preserve permitted factual records in the approved system. The district owner follows the applicable MDE incident-reporting process and other obligations; the teacher should not investigate independently or erase evidence.

Incident and escalation record

Use when: safety, privacy, security or integrity is threatened. Completed by: the responsible adult/incident owner in a restricted system. Record [what happened and when], [affected function/data category], [immediate student protection and paused access], [restricted evidence location], [required report and owner/time], [corrective action], [restart authority/conditions]. Keep unnecessary sensitive detail out of a general planning record. Complete when required reports, support, containment and the next accountable action are established; emergency action comes before paperwork.

Test a bounded activity and make the next decision

A small pilot should test feasibility, error patterns, access and workload. It does not by itself establish that AI caused better learning. Start with one planning task or supervised activity; do not expand to a whole school simply because a demonstration looked convincing.

Small-pilot charter and scorecard

Use when: the essential approvals are in place. Completed by: the educator and school/district sponsor. Record [one instructional problem], [class/planning scope and duration], [owner/reviewer], [approved inputs/tool/permissions], [baseline task], [required supports], [success measures], [stop conditions]. Choose local targets before the test, not after seeing results.

Prepare and check the material, run the bounded activity, gather appropriate feedback and review the record. Stop for unapproved access, protected or secure-test data exposure, unsafe output, ineffective required supports or a result the educator cannot verify in time.

Continue, revise, pause or stop record

Record [decision], [actual evidence and comparison limits], [remaining risk], [conditions or corrections], [owner], [next review]. Continue only within the tested approved scope. A change in model, feature, data, audience, permissions, terms, curriculum or policy triggers review. Confirm school/district authorization before expansion.

Use the method with MAIN resources

For district policy drafting, use the K–12 AI Policy and Guidance Template. For practical prompting, explore MAIN prompting guides. Educator resources and the current course catalog support professional learning; confirm current course and CEU conditions with the provider.

The Faculty Playbook serves postsecondary teaching; the Student Playbook supports student decisions; the Cybersecurity Playbook supports district technical teams. Their audience and permission rules do not automatically transfer to a K–12 class. Browse the AI Playbooks hub.

Review currency: September 5, 2026. A local owner should recheck source applicability when requirements or the activity change. The classroom cases are illustrative scenario walkthroughs, not evidence of observed learning gains or MDE endorsement.

Sources and image credit

Mississippi and federal requirements, agency guidance and voluntary risk frameworks have different roles. Consult the relevant current source and local owner; a source listing does not certify a product or a local implementation.

• MDE Digital Learning and AI guidance
• MDE Test Security
• MDE Accountability Resource Binder, January 2026, Appendix F
• MDE Information Security and Data Privacy
• MDE literacy resources for administrators
• MDCPS: Reporting Child Abuse and Neglect
• ED: FERPA
• ED: Protection of Pupil Rights Amendment
• IDEA regulation §300.105, assistive technology
• ED assistive technology guidance
• FTC COPPA final-rule amendments
• FTC COPPA guidance, including schools
• DOJ Title II web/mobile rule and school-district deadline guidance
• ED July 2025 guidance on AI use in schools
• NIST AI Risk Management Framework
• NIST Generative AI Profile
• Joint agency guidance: Careful adoption of agentic AI services, May 2026

Contextual stock photograph: Katerina Holmes / Pexels. Used under the Pexels License; cropped for composition. The photograph does not document a MAIN program or imply endorsement.