Why this is hard to get right
The Problem With "One-Size-Fits-All" Reading Materials
Maya is a 6th-grade ELA teacher at a Title I school in Phoenix. Her class of 28 students spans a reading range from 3rd grade to 9th grade — a gap she navigates every single day. When her district adopted a new science integration unit on renewable energy, she was handed a single-level textbook excerpt and told to "differentiate as needed."
She knew what that meant: hours of rewriting. She'd done it before — manually simplifying sentences for her struggling readers, adding sentence stems, stripping jargon, and then building a harder version for her advanced readers who finished in five minutes and stared at the ceiling. Then the questions. Then the answer key. Then the vocabulary list. For one passage.
She estimated six to eight hours of prep for a single reading set. She didn't have six to eight hours.
She turned to an AI assistant. Her first attempt: "Write a reading passage about solar energy for 6th graders with some comprehension questions." The output was a single passage, mid-level, generic. The questions were surface-level recalls — "What is solar energy?" — with no answer key. It looked like something printed from a 1990s workbook.
She tried again: "Make it easier and also make a harder version." The AI simplified by shortening sentences but didn't adjust vocabulary load or conceptual density. The "harder" version just added more paragraphs.
The real issue wasn't the AI. It was the prompt. Maya hadn't given the model the scaffolding it needed to behave like a literacy specialist. She hadn't defined lexile bands, question types, skill targets, or what "teacher-ready" actually meant.
When she restructured her request, she specified three Lexile bands, named the exact comprehension skills she was targeting (main idea, inference, evidence-citing), defined the question mix (multiple choice, short answer, evidence-based), and asked for vocabulary with student-friendly definitions and teacher notes including ELL tips.
The output was transformative. All three passages used the same topic and real-world statistics. Each passage had its own title, appropriate vocabulary load, and six aligned questions with rationale-backed answer keys. The teacher notes flagged which questions worked as discussion starters and offered a 10-minute extension activity.
Maya ran the activity the next day. Every student had an entry point. Her advanced readers weren't bored. Her struggling readers weren't lost. She spent 20 minutes reviewing the output — not six hours creating it from scratch.
The difference wasn't the AI. It was the specificity of the instruction. A prompt that mirrors how a literacy specialist actually thinks — in terms of complexity bands, skill taxonomies, question hierarchies, and classroom logistics — produces output that a real classroom can use immediately.
Common mistakes to avoid
Omitting Lexile or Grade Band Specifications
Without a defined complexity target, the AI defaults to a single mid-level passage that works for no one. Always specify Lexile ranges or grade bands (e.g., 500L–700L for Grades 3–5). Vague labels like 'easy, medium, hard' produce inconsistent results — different models interpret these differently, and outputs can cluster around the same difficulty level.
Requesting Generic Comprehension Questions
Asking for 'some questions' yields low-order recall items that don't build or assess real skills. Name the exact question types and quantities — 2 multiple choice, 2 short answer, 2 evidence-based — and tie them to specific skills like inference or main idea. This forces the model to distribute cognitive demand intentionally across Bloom's taxonomy levels.
Skipping Vocabulary and Definition Requirements
Domain vocabulary is critical for comprehension, especially in content-area reading. If you don't ask for it explicitly, the model omits it entirely or buries terms without definitions. Request a set number of vocabulary words with student-friendly definitions for each passage level to ensure the output is genuinely self-contained and usable without supplementation.
Neglecting ELL and Accessibility Constraints
Most classrooms include English language learners, yet prompts rarely account for them. Without explicit guidance, AI-generated passages may use idiomatic language, cultural references, or assumed background knowledge that exclude ELL students. Add a constraint requiring neutral tone, no idioms, and teacher notes with ELL differentiation tips to make every passage inclusive by design.
Forgetting to Request an Answer Key With Rationales
An answer key without rationales is only half-useful. Teachers need to understand why an answer is correct — especially for inference and evidence-based questions — to guide discussion and address misconceptions. Specify that you want rationales for every answer, not just the correct letter or phrase. This also helps catch AI errors before the material reaches students.
Using Broad or Abstract Topics Without Constraints
Prompting for a passage on 'science' or 'history' produces unfocused output with unpredictable scope. Name a specific topic, include a note about prior knowledge requirements, and flag any content boundaries (e.g., 'avoid political framing'). Tight topic scoping also lets the AI maintain consistency across all three reading levels — which is essential for differentiated sets.
The transformation
Write a reading passage with some questions for my class.
You are a K–12 literacy specialist. Create 3 differentiated nonfiction passages on the same topic (renewable energy). 1: Grades 3–4 (~500L). 2: Grades 5–6 (~800L). 3: Grades 7–8 (~1000L). 1) Focus skills: main idea, inference, citing evidence, domain vocabulary. 2) For each passage, provide: title, 350–550 words, 6 questions (2 multiple choice, 2 short answer, 2 evidence-based), answer key with rationales, 5 vocabulary terms with student-friendly definitions. 3) Constraints: neutral, engaging tone; no prior knowledge required; avoid technical jargon; include one real statistic with source. 4) Add teacher notes: extension activity, differentiation tips for ELLs, and suggested timing.
Why this works
Lexile Anchors Control Complexity
The After Prompt specifies three distinct Lexile bands — ~500L, ~800L, and ~1000L — mapped to grade ranges. This eliminates the model's tendency to default to a single difficulty level. Each passage is calibrated to a real reading development stage, which means the output mirrors what a trained literacy specialist would produce for a tiered classroom resource.
Structured Question Mix Targets Skill Depth
The After Prompt mandates 6 questions per passage in a specific distribution: 2 multiple choice, 2 short answer, 2 evidence-based. This distributes cognitive demand across comprehension levels — from literal recall to inferential reasoning to textual evidence. Without this structure, AI-generated questions cluster at low Bloom's levels and fail to assess higher-order skills.
Topic Constraints Ensure Cross-Level Consistency
By fixing the topic (renewable energy) and banning prior knowledge requirements or technical jargon, the After Prompt ensures all three passages are comparable and fair. Students at different levels read about the same content, enabling whole-class discussion. Shared topic coherence is something AI skips entirely without explicit instruction to maintain it.
Answer Key With Rationales Builds Teacher Confidence
The After Prompt requests an answer key with rationales for every question. This transforms the output from a student-facing worksheet into a complete instructional resource. Rationales help teachers anticipate misconceptions, facilitate evidence-based discussion, and verify the accuracy of AI-generated answers before use.
Teacher Notes Extend Classroom Utility
The After Prompt asks for extension activities, ELL differentiation tips, and suggested timing. These details convert a reading passage into a deployment-ready lesson component. Without them, teachers still face preparation work after receiving AI output — eliminating the time savings that motivated the prompt in the first place.
The framework behind the prompt
The Theory Behind Differentiated Reading Instruction
Differentiated instruction traces its formal framework to Carol Ann Tomlinson's 1990s work at the University of Virginia, which established that effective teaching adjusts content, process, and product based on student readiness, interest, and learning profile. In reading specifically, this means providing texts at appropriate challenge levels — not dumbing content down, but calibrating complexity so every learner has a productive entry point.
The Lexile Framework, developed by MetaMetrics, operationalizes text complexity as a single score combining word frequency and sentence length. It gives educators a measurable target rather than a subjective judgment. The Common Core State Standards (CCSS) formalized Lexile bands by grade in 2010, connecting text complexity directly to college and career readiness expectations. This framework underpins the grade-band structure used in the After Prompt on this page.
Bloom's Taxonomy explains why question design matters as much as passage design. Benjamin Bloom's 1956 cognitive hierarchy — remember, understand, apply, analyze, evaluate, create — maps directly to question types. Recall questions (multiple choice, literal comprehension) sit at the lower end. Inference questions and evidence-based written responses demand analysis and evaluation. A well-designed reading set deliberately distributes questions across these levels to build — not just assess — higher-order thinking.
The Three-Tier Vocabulary Model (Beck, McKeown, and Kucan) further informs word selection across levels. Tier 1 words are everyday language. Tier 2 words are high-utility academic terms that appear across content areas (analyze, contrast, significant). Tier 3 words are domain-specific (photosynthesis, legislative, elasticity). Effective differentiated passages calibrate vocabulary density and tier by reading level.
Finally, Universal Design for Learning (UDL) principles — particularly the CAST framework — argue that accessible materials benefit all learners, not just those with identified needs. Requesting ELL notes, avoiding assumed background knowledge, and providing student-friendly definitions are all UDL-aligned practices. A well-structured AI prompt can encode these principles systematically, producing materials that meet accessibility standards by default rather than by afterthought.
Prompt variations
You are an instructional designer specializing in workplace learning. Create 3 differentiated reading passages on the topic of cybersecurity awareness for corporate employees.
Levels:
- Entry-level employees (8th-grade reading level, ~600L)
- Mid-level staff (11th-grade reading level, ~1000L)
- Technical roles (college-level, ~1300L)
For each passage, provide:
- Title and 350–500 words
- 5 comprehension questions: 2 multiple choice, 2 scenario-based short answer, 1 application question
- Answer key with rationales
- 4 key terms with plain-language definitions
Constraints: Use workplace scenarios, avoid vendor-specific language, assume no prior IT knowledge for levels 1 and 2, use neutral professional tone.
Facilitator notes: Include one discussion prompt per passage, a suggested debrief activity, and a note on which questions to prioritize for 30-minute training sessions.
You are a K–5 science and literacy integration specialist. Create 3 differentiated nonfiction passages about the water cycle for an elementary classroom.
Levels:
- Grade 1–2 (~200L, 150–200 words)
- Grade 3–4 (~500L, 300–400 words)
- Grade 5 (~750L, 450–550 words)
For each passage, provide:
- An engaging title
- 4 questions: 1 picture-based (describe what a diagram would show), 1 literal recall, 1 vocabulary in context, 1 making connections
- Answer key
- 3 vocabulary words with kid-friendly definitions using everyday language
Constraints: No prior science knowledge assumed, no technical Latin-root terms for levels 1 and 2, include one real-world example (rain, rivers, drinking water), maintain consistent characters or setting across all three levels to support whole-class discussion.
Teacher notes: Suggest one hands-on activity that pairs with each level and a whole-group anchor question.
You are a high school literacy coach. Create 3 differentiated argumentative reading passages on the same policy topic: whether cities should ban single-use plastics.
Levels:
- Grades 9–10 (~900L, 400–500 words) — one-sided argument
- Grades 11–12 (~1100L, 500–600 words) — balanced argument with counterargument
- AP/Advanced (~1300L+, 600–700 words) — nuanced policy analysis with competing evidence
For each passage, provide:
- 6 questions: 2 identifying claims and evidence, 2 evaluating author reasoning, 2 written response (short paragraph)
- Answer key with scoring criteria for written responses
- 5 rhetorical or academic vocabulary terms with definitions
Constraints: Present both sides fairly, cite one real statistic per passage, avoid partisan framing, use formal academic register.
Teacher notes: Include a cross-level synthesis question for whole-class Socratic seminar, suggested timing per level, and tips for using passages in argument mapping activities.
You are an adult ESL curriculum developer. Create 3 differentiated reading passages on the topic of healthy eating for adult learners in a community English program.
Levels:
- Beginner (CEFR A1–A2, ~200–250 words, simple present tense, short sentences)
- Intermediate (CEFR B1, ~350–450 words, varied sentence structure)
- Upper-Intermediate (CEFR B2, ~500–600 words, complex sentences, passive voice used sparingly)
For each passage, provide:
- 5 questions: 1 true/false, 2 multiple choice, 1 fill-in-the-blank from passage, 1 personal connection question
- Answer key
- 4 vocabulary items with definitions and a sample sentence for each
Constraints: Use culturally neutral food examples, avoid idiomatic expressions at levels 1 and 2, use relatable everyday contexts (grocery shopping, cooking at home), avoid Western food assumptions.
Instructor notes: Include a speaking prompt for pair practice, one grammar focus point per level tied to the text, and a suggested warm-up activity.
When to use this prompt
Marketing Managers
Create content literacy passages about your product category for internal enablement training, tailored by role seniority.
Product Managers
Generate user education passages that explain complex features at beginner, intermediate, and advanced reading levels.
Customer Success Teams
Build help center reading practice with comprehension checks for onboarding, localization, and accessibility needs.
Researchers
Summarize study topics into leveled passages to prep stakeholders with varying expertise before briefings.
Educators and Curriculum Leads
Produce standards-aware, leveled reading sets for mixed-ability classrooms with aligned questions and notes.
Pro tips
- 1
Specify lexile or grade bands to control complexity and ensure accessibility.
- 2
Define exact question types and counts to align with assessment goals.
- 3
Include topic constraints and tone guidance to prevent misconceptions and bias.
- 4
Add teacher or facilitator notes to make the output immediately classroom- or training-ready.
Once you've mastered basic three-level differentiation, you can add instructional scaffolding directly into each passage to deepen accessibility without separate handouts.
For the lowest level, request embedded glossary callouts — short parenthetical definitions immediately following unfamiliar terms in the text itself. This reduces cognitive load for struggling readers and eliminates the need to flip between glossary and passage.
For the middle level, ask for text features like a bolded summary sentence at the end of each paragraph. This mirrors how real informational texts are structured and teaches students to identify main ideas in context.
For the highest level, request deliberate use of implicit evidence — statements where the answer requires synthesizing two non-adjacent sentences. This trains students for standardized test formats and close reading.
You can also request margin annotation prompts — short questions printed in a sidebar next to each paragraph (e.g., 'What is the author's purpose here?'). These turn each passage into an active reading exercise without additional materials.
Finally, if you're building a unit rather than a one-off activity, ask the AI to include a cross-level bridge question at the end of each passage — a single question that every student, regardless of level, can answer using their respective passage. This enables genuine whole-class discussion without exposing reading level differences.
The structure of your passage prompt should change depending on whether the output is for formative practice or summative assessment.
For classroom practice, prioritize open-ended short answer questions, discussion prompts, and vocabulary application tasks. Include teacher notes with suggested discussion facilitation moves. Accuracy matters, but the goal is building skills — so you can tolerate some ambiguity in answer keys.
For formal assessment, tighten every constraint. Specify that questions must have a single defensible answer. Request that each multiple-choice option includes a distractor rationale — an explanation of why each wrong answer is plausible but incorrect. This dramatically improves the validity of your assessment and helps you identify misconceptions from incorrect responses.
For benchmark or placement testing, add a Lexile ceiling and floor for each level, and request that the passage avoids topic-specific background knowledge that might advantage students from certain geographic or cultural contexts. Ask the AI to use a 'cold read' structure — no context-setting title or introduction that might scaffold comprehension artificially.
A practical tip: label your prompt with its purpose from the start. 'You are designing a formal assessment passage, not a practice worksheet' meaningfully changes the rigor and precision of the AI's output.
If you generate reading sets regularly — for a curriculum team, a test prep company, or a multi-grade department — you can use this prompt structure to build a reusable template library rather than starting from scratch each time.
Start by creating topic-agnostic shell prompts that contain all your fixed constraints (reading levels, question types, vocabulary format, teacher notes structure) but leave the topic variable. Save these as named templates. When a new topic arises, you slot it in and run the prompt — no structural rework required.
Next, build a content constraint bank — a list of pre-approved topic scoping notes for common subject areas. For example: 'Science passages: avoid speculation, cite peer-reviewed data, no medical advice.' Having these constraints pre-written reduces the risk of forgetting critical guardrails.
Finally, maintain a quality review checklist that runs alongside every AI output: Lexile match (spot-checked), question type distribution, vocabulary appropriateness per level, factual accuracy of statistics, answer key rationale completeness, and ELL accessibility flags. This checklist catches the most common failure modes before any passage reaches a student.
Teams that combine structured prompts with systematic review reduce revision time by 60–70% compared to unstructured AI use — and produce more consistent outputs across contributors.
When not to use this prompt
When This Prompt Pattern Is Not the Right Tool
This prompt structure produces its best results for informational, nonfiction reading practice aligned to comprehension skill-building. It's not always the right approach.
Avoid it for literary or narrative fiction. Differentiating a short story by Lexile level distorts voice, pacing, and author intent. For fiction, differentiation usually works better through scaffolded questions or audiovisual supports — not rewritten text.
Don't use it for high-stakes standardized test passages. Assessments like the SAT, ACT, or state summative exams require human-authored, professionally validated passages with documented psychometric properties. AI-generated passages may contain subtle factual errors or ambiguous answer choices that invalidate their use as assessment instruments in formal testing contexts.
Skip it when your topic requires scientific precision. Medical, legal, or highly technical content carries risk if AI-generated passages contain inaccuracies. In these domains, a subject-matter expert must review every output before any learner sees it — which may negate the time savings.
Consider simpler prompts when you only need one level. This three-passage structure is powerful for differentiation but adds complexity. If your audience is homogeneous, a single targeted passage with aligned questions is faster and easier to quality-check.
Troubleshooting
All three passages read at approximately the same difficulty level despite different Lexile targets
Add explicit linguistic differentiation markers for each level. Specify: 'Level 1 must use only simple sentences under 12 words and Tier 1 vocabulary. Level 3 must include subordinate clauses, Tier 3 domain terms, and at least one complex sentence with embedded information.' Abstract Lexile numbers alone don't consistently trigger structural changes — linguistic rules do.
Questions are all low-order recall items with no inference or evidence-based tasks
Replace the generic 'comprehension questions' request with a named taxonomy. Add: 'At least 2 questions must require students to infer meaning not stated directly in the text. At least 2 must require students to cite specific evidence with a sentence starter like "According to the text..."' Naming the cognitive operation forces the model to distribute question difficulty intentionally.
The AI changes the topic or introduces new subtopics across the three passage levels
Add a cross-level coherence constraint: 'All three passages must address the same central topic from three different angles — what it is, how it works, and why it matters — without introducing new subject matter not relevant to the core topic.' Also specify: 'A student who reads all three passages should be able to participate in a shared discussion about one central idea.'
Vocabulary terms are too advanced for the lower level or too simple for the higher level
Link vocabulary selection directly to each level's Tier structure. Specify: 'Level 1 vocabulary must be Tier 1 or high-frequency Tier 2 words. Level 3 vocabulary must include Tier 3 domain-specific terms used in the passage context.' If the model still miscalibrates, provide 2–3 example vocabulary words for the hardest level to anchor its selection logic.
Answer key lacks rationales and only provides correct answers
Restate the rationale requirement explicitly in a separate instruction line: 'For every answer key entry, explain in 1–2 sentences why the answer is correct and what skill or text evidence supports it. For multiple-choice items, briefly explain why each distractor is incorrect.' Burying this requirement in a list often causes the model to skip it — a standalone instruction line increases compliance significantly.
How to measure success
How to Evaluate the Quality of Your AI Output
Use these criteria to assess whether the generated reading set is classroom-ready.
Passage quality:
- Each passage falls within 75–100L of its specified target band (verify with a Lexile estimator)
- Sentence length and vocabulary density visibly differ across levels
- The topic stays consistent — a student could recognize all three passages as being about the same subject
- No passage requires prior background knowledge not provided in the text itself
Question quality:
- Questions are distributed across at least 3 cognitive levels (recall, inference, evidence-citing)
- Every multiple-choice item has 3–4 plausible options — not one obvious answer and three nonsensical distractors
- Evidence-based questions require citing specific text, not just agreeing or disagreeing
Answer key quality:
- Every answer includes a rationale, not just the correct response
- Rationales reference specific text evidence or reasoning steps
- Written response rubrics (if requested) include at least 2 scoring levels
Teacher utility:
- Vocabulary terms are calibrated to each level's appropriate Tier
- ELL differentiation notes are specific, not generic
- Suggested timing is realistic for a standard class period
Now try it on something of your own
Reading about the framework is one thing. Watching it sharpen your own prompt is another — takes 90 seconds, no signup.
Turn your topic and grade bands into a complete, classroom-ready differentiated reading set in one structured prompt.
Try one of these
Frequently asked questions
Yes — this prompt structure works across any content area. Science, social studies, health, and professional development all benefit from leveled reading sets. The key is keeping the topic consistent across all levels and specifying domain constraints (e.g., 'avoid technical jargon' or 'assume no prior chemistry knowledge'). The reading skills you target may shift — summarizing in social studies, comparing in science — but the structure stays the same.
AI models estimate complexity but don't calculate official Lexile scores. Treat the output as a starting approximation. For formal assessment use, run the final text through a free Lexile or Flesch-Kincaid analyzer tool to verify. For classroom practice, review sentence length, vocabulary density, and conceptual load manually. Most AI outputs hit the target range within 100–150L when Lexile bands are clearly specified.
Add a standards clause to the constraints section. For example: 'Align all questions to CCSS ELA standards RI.6.1, RI.6.3, and RI.6.6' or reference your state's literacy framework by name. The AI will map question types and skill descriptors to those standards. Always verify alignment manually — AI familiarity with standards is generally strong for CCSS but less reliable for state-specific variations.
This happens when Lexile targets are present but the model doesn't adjust vocabulary and sentence structure independently. Add explicit differentiation instructions — for example: 'Level 1 should use only single-clause sentences and Tier 1 vocabulary. Level 3 should include multi-clause sentences, Tier 3 domain terms, and embedded parenthetical phrases.' Explicit linguistic markers force meaningful differentiation beyond just word count.
Three levels works for most mixed-ability contexts, but you can request two or even four. Match the number of levels to your actual student groupings. If your class has only two reading clusters, two passages reduce preparation overhead without sacrificing utility. For pull-out or gifted programs, a single advanced-level passage with enriched questions may be all you need.
Specify in the prompt that each passage should approach the same topic from a different angle or time scope. For example: Level 1 covers what renewable energy is, Level 2 covers how it works, Level 3 covers policy debates around it. This keeps the topic coherent without identical sentences appearing across levels — a common failure mode when topic consistency is requested without angle differentiation.
Well-structured prompts produce output that requires light editing rather than heavy revision — typically 10–20 minutes of review. Check for: factual accuracy of any statistics, grade-appropriateness of vocabulary at each level, and question alignment to your specific assessment goals. Never use AI-generated answer keys without reviewing rationales — occasional inference errors occur, especially on ambiguous short-answer questions.
Add an explicit neutrality constraint to the prompt: 'Present all claims factually, avoid political framing, and do not express a preferred outcome.' For topics with genuine complexity (climate policy, historical events), request that the passage names multiple perspectives without ranking them. You can also ask the model to flag any sections where bias risk is highest so you can review those first.