Choosing the Right Tool

The Traffic Light System

Before asking which AI tool to use, there is a more useful question: Is this particular task one you should be using AI for at all, and with what safeguards? This page walks you through tool selection in a few minutes. This guidance applies whether the tool is free, district-licensed, or built into something you already use.

This is adapted from the ARC AI Risk Assessment Checklist. It is a living document. If you run into a situation it does not clearly cover, contact the AI Resource Center at ai-resource-center@domail.maricopa.edu so we can assist you and update the list for everyone.

Every question below sorts your situation into one of three categories. Work through all seven, then act on the most restrictive result you hit β€” one red light governs the whole activity, regardless of how many greens you collected.

  • πŸ›‘ Red β€” Stop. Do not proceed without consultation and safeguards in place.
  • ⚠️ Yellow β€” Proceed with caution. Acceptable with oversight, documentation, and a plan for managing the risk.
  • 🟒 Green β€” Proceed. Low risk. Apply normal professional judgment.

The Seven Questions

πŸ›‘ Red: Student records, grades, health information, Social Security numbers, financial data, disciplinary records, anything FERPA-protected, student contact information. Employee personnel records, salary information, hiring and termination data, employee health records, confidential HR investigations. Donor information, contract negotiations, legal matters, vendor proposals with pricing, proprietary institutional data, audit materials, communications with legal counsel.

⚠️ Yellow: De-identified student work with names removed. Aggregated class performance data with no individual identifiers. Publicly available institutional data. Aggregated operational data. De-identified process metrics. General budget categories rather than line items. Anonymized survey responses. Public meeting agendas and minutes.

🟒 Green: Your own lecture notes, syllabi, and teaching materials. Publicly available resources. Generic examples. Your own work documents and drafts. Published policies and procedures. Marketing materials. Standard templates and forms. Public datasets.

Examples: Red β€” uploading a gradebook with student names to analyze grade distributions; asking AI to analyze applicant resumes with names and contact details. Yellow β€” analyzing anonymized essay patterns to improve a rubric. Green β€” generating practice problems for your math class; drafting an email template for common inquiries.

πŸ›‘ Red: Final course grades. Academic integrity findings. Student placement decisions. Recommendation letters used verbatim. Tenure or promotion decisions. Research conclusions published without verification. Hiring or termination decisions. Financial aid determinations. Student discipline. Legal or compliance decisions. Budget allocations without review. Facility access or security decisions.

⚠️ Yellow: Draft feedback you will review and personalize. Preliminary analysis for further investigation. Teaching materials you will verify. Draft communications requiring personalization. Process recommendations requiring approval. Initial screening with human review at every decision point.

🟒 Green: Brainstorming. Formatting. Grammar checking your own writing. Generating discussion questions. Organizing content. Drafting routine email. Meeting agendas. Templates. Scheduling suggestions.

Examples: Red β€” using AI-generated scores as final grades; determining plagiarism from a detection tool alone. Yellow β€” drafting initial feedback and then personalizing each response. Green β€” brainstorming ways to explain a difficult concept.

πŸ›‘ Red: No verification planned. Using output as-is for grading, feedback, or decisions. Presenting AI-generated content as fact without checking. Sending AI-written communication without reading it.

⚠️ Yellow: Spot-checking key sections. Comparing against authoritative sources. Having a colleague review. Testing with a small group first. Verifying citations and facts. Cross-referencing institutional policy. Piloting before broader use.

🟒 Green: Reviewing and fact-checking everything. Using output only as a starting point. Applying your own subject-matter expertise. Personalizing all content. Validating with the right stakeholders.

Examples: Red β€” giving students AI-written explanations without verifying accuracy. Yellow β€” generating a reading list and confirming each source exists. Green β€” using AI to check a syllabus for clarity, then deciding yourself.

πŸ›‘ Red: Unknown privacy policy. Unclear data handling. A tool that may use your input for training. Unclear third-party access. Anything requiring student or employee data to create an account. Free consumer tools with unclear terms. No vendor agreement on file.

⚠️ Yellow: A vendor with an educational privacy agreement. Clear retention policies. Limited third-party access with safeguards. A pilot with IT oversight. Approved for specific use cases only.

🟒 Green: A district-approved tool with clear protections. A FERPA-compliant vendor agreement. Data governance approval. No retention or training use. Or you are only sharing public information and your own non-sensitive work.

Examples: Red β€” using free consumer tools with student names and work; having students create accounts on unapproved platforms. Yellow β€” using a district-licensed tool still in pilot with IT guidance. Green β€” using an approved platform for your own course planning.

πŸ›‘ Red: Output affects student grades or opportunities. Could disadvantage protected groups. Perpetuates stereotypes. No consideration of bias at all. Used for high-stakes decisions. Creates unequal access. Used in hiring, promotion, or evaluation.

⚠️ Yellow: Bias is possible but your oversight reduces the impact. Used for low-stakes activities. You have considered bias and have a mitigation strategy. Applied consistently across groups with documented safeguards.

🟒 Green: No impact on student outcomes. Promotes accessibility and inclusion. Equal benefit to all students. Minimal bias risk. Supports inclusive practice.

Examples: Red β€” predicting which students will succeed without accounting for bias in training data; requiring an AI tool when some students lack reliable technology access. Yellow β€” generating diverse examples and reviewing them for representation. Green β€” using AI to draft alt text to improve accessibility.

πŸ›‘ Red: A student receives an incorrect grade or misses an opportunity. An integrity accusation based on an error. Published research containing false information. Legal liability. Irreversible academic consequences. Incorrect personnel action. Financial aid errors affecting eligibility. A compliance or regulatory breach.

⚠️ Yellow: Moderate inconvenience requiring correction. You can catch and fix errors before anyone is affected. Limited scope. Backup verification in place. Established rollback procedures.

🟒 Green: Minimal impact, easily corrected, no harm to students. Affects only your own planning or internal drafts. Errors caught during normal review.

Examples: Red β€” a detection tool wrongly flags student work and an integrity case follows; a recommendation letter contains invented achievements. Yellow β€” an AI-drafted assignment has unclear instructions you catch before posting. Green β€” a suggested activity does not quite fit your class, so you adapt it.

πŸ›‘ Red: Not vetted by IT. No security review. Unknown vendor. Requires accounts with personal or student data. Found through a web search. No privacy policy. No approved vendor agreement.

⚠️ Yellow: Pending IT approval. A pilot with oversight. Approved for limited use cases. Requires specific training. Under evaluation.

🟒 Green: On the district’s approved list. IT-vetted and security reviewed. District-wide license available. Included in professional development. Documented in district AI guidance. Gemini, NotebookLM, and Workspace Studio are licensed for Maricopa faculty and staff.

Examples: Red β€” using a grading tool found online without IT review; installing AI browser extensions without approval. Yellow β€” joining an IT-supervised pilot. Green β€” using a district-licensed platform for course development.

Four Faster Self-Checks

If you want a shortcut before working through all seven, these four catch most problems:

  • Would I share this with an outside contractor or vendor without additional approvals? If not, do not share it with AI.
  • Would this data pass the FERPA flag check? Treat AI as a third party. If it would not pass for a student’s record, it does not go in.
  • Would I be comfortable if this appeared in a public forum? Maricopa is a public institution and subject to public records requests.
  • Am I required to follow specific data handling procedures for this? If yes, those same procedures apply to AI use.

When you are building something, not just using something

The checklist above covers using an AI tool for a task. Building something others will depend on β€” an automated workflow, a configured assistant, a tool embedded in a course or a service β€” is a different scale of activity, and additional requirements apply.

Depending on what you are building, that may include data classification, contract and procurement review, vendor security review, accessibility requirements, and Arizona records retention obligations. These apply regardless of whether the tool costs anything.

None of these are ARC requirements β€” they are existing district gates that apply to any system, and AI does not exempt anything from them. If you are unsure which apply, contact the ARC and you will be pointed to the right office.

What to Do With Your Result

  1. Do not proceed with the activity as planned.
  2. Contact the AI Resource Center at ai-resource-center@domail.maricopa.edu to explore alternatives. There is frequently a way to accomplish the same goal differently.
  3. Document your intended use and the concerns identified.
  4. Wait for guidance before moving forward.

A red light is not a refusal. It means the question is bigger than one person’s judgment, and someone should look at it with you.

  1. Document your approach β€” how you will use AI and what safeguards you are putting in place.
  2. Build in oversight β€” specific verification steps and human review, named rather than assumed.
  3. Start small β€” one section, one assignment, one process before broader use.
  4. Monitor closely and be ready to adjust.
  5. Be transparent with the people affected β€” your class, your team.
  6. Ask for support. The ARC can help identify risk mitigation strategies.

  1. Apply professional judgment β€” evaluate the output with your own expertise.
  2. Keep a record of how you used AI, for your own reference.
  3. Stay current. Capabilities and limitations both change.
  4. Share what works. The ARC actively wants to hear about effective uses β€” see Case Studies.
  5. Stay alert. Even low-risk uses benefit from attention.