Professional learningfor Teachers, Librarians, Coaches
Bias Detectives
Teams compare paired AI outputs where only a name, a language, or a pronoun changed, then run their own fair tests and propose fixes before AI output reaches students.
All activities Professional learning
Personalization with Guardrails
Teachers design AI-supported versions of one reading task for different learners, then judge every version against the same learning target, a fact check, and a privacy line.
Overview
AI tools can quickly produce leveled texts, bilingual glossaries, chunked questions, and read-aloud audio, and paired with long-standing assistive technology like text-to-speech and speech-to-text, they can open grade-level content to more students. They can also quietly lower expectations, introduce factual errors, or expose student information. In this session teams first catch an error in a mock AI-leveled passage, then design supported versions of one science reading for fictional learner profiles and judge every design against four guardrails: same learning target, verified accuracy, protected privacy, and evidence of learning.
Objectives
- Participants will design AI-supported versions of a reading task that keep the same grade-level learning target for different learners.
- Participants will verify an AI-modified text against its original and correct any meaning or factual changes.
- Participants will match supports (text-to-speech, speech-to-text, glossaries, chunking, extension) to specific learner needs rather than labels or "learning styles."
- Participants will protect student privacy by keeping names and disability or IEP details out of AI prompts.
Materials
On paper
- Handout A: The Monarch Waystation, original passage and a mock AI-leveled version (1 per person)
- Handout B: Learner Profile Cards (1 set per team, cut apart)
- Handout C: Guardrails Rubric (2 per team)
- Chart paper for design posters, markers, sticky notes for gallery feedback
On screen
- A generative AI chatbot your district approves, projected by the facilitator or used by teams with approved staff accounts
- Any text-to-speech tool available on district devices (many operating systems and browsers include one), plus speech-to-text dictation
- Optional: headphones for trying text-to-speech
Before you start
- Print Handouts A and C; print and cut Handout B. Each team needs all eight profile cards.
- Test the text-to-speech and dictation features on your district's devices so you can demo them in under two minutes.
- Run one leveling prompt on your approved AI tool with the Handout A original so you know what it currently produces. Check its output against the original yourself first.
- If your campus has a special education or dyslexia specialist, invite them to co-facilitate the Model step.
Step by step
-
10–5 min
Hook
Easier, or different?
Project the two versions on Handout A side by side. Ask: "The simplified version is easier to read. Is it teaching the same thing?" Give 90 seconds of silent reading, then collect first reactions without confirming anything yet.
Facilitator noteMost people notice the shorter sentences first. Fewer notice the content changes. That gap is the session.
-
25–15 min
Explore
Verify the leveled text
Individually, compare the mock AI-leveled version line by line with the original. Mark (1) any factual error, (2) any key idea dropped, and (3) any vocabulary the learning target needs that was removed. Teams compare and agree. Then reveal: the mock version says monarchs lay eggs on "many kinds of plants" (false: only milkweed), drops the reason the waystation matters, and removes the word migration, which is the lesson's target vocabulary.
Facilitator noteKey message: an AI-leveled text is a draft, not a finished material. Simplifying can change meaning, and a student reading the easier version would learn something false.
-
315–25 min
Model
Supports that keep the target
Demonstrate three moves on the projector. (1) Text-to-speech on the original passage: a student with dyslexia can access the grade-level text itself. (2) A privacy-safe prompt to the approved AI tool: "Create a glossary of five key terms from this passage with student-friendly definitions and a Spanish translation. Keep the terms migration, milkweed, and waystation." No student names, no diagnosis. (3) Checking the output against the original before it goes to anyone. Name the principle: change the access, keep the thinking.
Facilitator noteAddress "learning styles" directly and kindly: research doesn't support matching instruction to a preferred style like "visual learner." Design for specific needs (decoding, language, attention, vision) and offer choices to everyone.
-
425–45 min
Create
Design for two learners
Each team draws two Learner Profile Cards and designs a supported version of the Monarch Waystation task for each: the supports, whether AI is involved (and the exact prompt, written without identifying details), how they'll verify any AI output, and what the student will produce to show they met the target: "Explain why a school garden with milkweed helps monarchs during migration." Put both designs on a chart poster. At least one support must be non-AI.
Facilitator noteWatch for designs that swap the task for an easier one ("label the parts of a butterfly"). Ask: "Is this student still explaining why the waystation matters?"
-
545–60 min
Practice
Guardrail gallery
Posters go on the wall. Teams rotate and score two other posters with Handout C, leaving sticky notes: one "keeps the target because…" and one "guardrail to add…". Back at home posters, teams revise one design based on the notes and mark the change in a different color.
Facilitator noteThe privacy guardrail gets missed most often. Look for prompts that include phrases like "a 7th grader with dyslexia named…".
-
660–70 min
Transfer
Your real text
Each person picks a reading they'll assign in the next two weeks and one student need in their class. They write the learning target, one support (AI or not), the verification step, and the evidence of learning they'll collect. Partners ask: "How will you know the support helped this student learn, not just finish?"
Facilitator noteCoaches: offer to co-plan the verification step and look at the student work together afterward. That's the measure-and-adjust part of the cycle.
Paper or screen
Unplugged
All the core thinking works on paper: verifying the mock leveled passage, designing supports from profile cards, and scoring with the rubric. Teams can describe any AI or assistive-technology support in writing, and the facilitator can read the original passage aloud to simulate text-to-speech. Non-AI supports such as printed glossaries, chunked question strips, and partner reading make excellent unplugged designs.
Digital
Teams with approved staff access run their privacy-safe prompts on the district's AI tool, try text-to-speech and dictation on the original passage, and verify the outputs against Handout A before adding them to their posters. In an LMS, teams post their designs to a discussion board for asynchronous rubric scoring, with the AI prompt and the verification notes included.
Does it need a screen? Actually hearing text-to-speech read the grade-level passage, and seeing an AI tool generate a glossary in seconds, shows what these supports make possible. Verifying the live output shows why teacher judgment still has to sit between the tool and the student. The evidence is a student producing the same grade-level explanation through a different path.
Evidence of learning
What you should be able to see or collect if it worked.
- Teams identify the factual error, the dropped key idea, and the removed target vocabulary in the mock leveled text.
- Every design keeps the same learning target and names the student product that will show it.
- AI prompts on posters contain no names, diagnoses, or other identifying details.
- Each design includes a verification step for any AI output and at least one non-AI support.
Adaptations
- Grades 3–5 teachers
- Use only the first two paragraphs of the original, and have students show understanding with a labeled drawing plus one spoken or dictated sentence explaining why milkweed matters.
- Special education and dyslexia specialists
- Lead the Model step. Add how supports connect to students' documented accommodations, and how students can learn to choose and turn on their own tools.
- Higher Ed faculty
- Swap the passage for a dense course reading and discuss how AI summaries, text-to-speech, and glossaries can support access while students still engage with the primary text.
Standards connections
Teachers verify an AI-leveled text and match supports like text-to-speech to learners, so students reach the same grade-level reading target.
- TEKS
- comprehension skillspractical technology conceptsTEKS sections: Technology Applications §126.8–§126.10, §126.17–§126.19, high school Technology Applications courses (19 TAC Chapter 126); ELAR §110.5–§110.7, §110.22–§110.24, §110.36–§110.39
Reflect
- How am I leveraging AI to personalize and enhance learning experiences without lowering expectations?
- When I simplify a text, how do I check that I haven't changed what students learn?
- How can students learn to choose and use their own supports, so personalization isn't something done to them?
Take it to your students
Use your Transfer plan with one real reading this week. Save the student work from students who used the support and from a few who didn't, and bring both to your PLC or coach to ask: did the support help students reach the target?
Pairs well with
- Digital Accessibility Readiness AssessmentChecks whether the materials and tools you use are accessible in the first place.
- AI Essentials Prompt LibraryStarter prompts for differentiation. Apply today's privacy and verification guardrails to each one.
