Generating Differentiated Word Problems Problems With AI
AI generates differentiated word problems most effectively when the differentiation is specified as problem structure differences, not just number size differences. Giving struggling students problems with smaller numbers is a limited form of differentiation — it changes the arithmetic difficulty but not the problem-solving complexity. Genuine differentiation changes the number of steps, the amount of given information, the explicitness of the question, and the level of context interpretation required.
Quick Answer: Specify differentiation at three levels: Level 1 (all information given explicitly, one operation, numbers within student comfort range), Level 2 (standard problem — two operations, grade-level numbers, requires identifying which operation), Level 3 (extended problem — three or more operations, some implicit information, requires multi-step planning and real-world interpretation). These three levels address three different problem-solving competency bands, not just three arithmetic difficulty levels.
Why "Same Problem, Smaller Numbers" Is Not Differentiation
The most common and least effective approach to word problem differentiation is reducing the numbers in a standard problem for struggling students. A Grade 5 student who can't solve "48 boxes with 6 bags each — how many bags in total?" is unlikely to be helped by changing the problem to "12 boxes with 2 bags each." If the difficulty was identifying the required operation (multiplication, not addition), smaller numbers don't address it.
This matters for AI prompting because it's the implicit differentiation strategy AI uses when asked to "write easy, medium, and hard versions of a word problem" without further specification. The easy version has small numbers; the hard version has large numbers; the medium version is somewhere between. This is stratification by arithmetic demand, not by problem-solving complexity.
Genuine word problem differentiation, as defined by ASCD (2025), modifies the cognitive demand of the problem-solving process — the number of steps required, the amount of inferential reasoning, the explicitness of the question, and the quantity of given information. These are the parameters that separate students who can solve word problems from those who cannot, and they are all specifiable in AI prompts.
The Four Differentiation Parameters for Word Problems
Parameter 1: Number of Operations Required
A one-step word problem (find the total cost if one item costs $6 and you buy 8 items) tests whether students can identify the operation. A two-step problem (total cost after a 10% discount) tests whether they can sequence two operations. A three-step problem (total cost after discount, then split among three people, then express as a percentage of a budget) tests multi-step planning.
Parameter 2: Explicitness of the Question
An explicit question directly states what to find: "How many bags are there in total?" An implicit question requires interpretation: "Is this enough for the whole class?" Students at Level 1 need explicit questions; students at Level 3 benefit from implicit questions that require determining what they actually need to find.
Parameter 3: Amount of Given Information
A scaffolded problem gives all required information and nothing else. A standard problem gives all required information with one or two additional pieces that aren't needed (students must identify which information is relevant). An extended problem has some required information implicit in the context — students must infer it from stated facts.
Parameter 4: Context Interpretation Required
A low-interpretation problem has a familiar context where the relationship between quantities is obvious (buying items at a fixed price — students have done this in shops). A high-interpretation problem has an unfamiliar context where students must model the relationship before they can calculate (comparing phone plan pricing structures with different rate models).
The Three-Level Differentiation Framework
These four parameters combine to produce three distinct problem levels that represent genuine differentiation — not just number resizing.
| Level | Operations | Question Type | Given Information | Context |
|---|---|---|---|---|
| Level 1 (Supported) | 1 operation | Explicit | All given, nothing extra | Familiar, direct relationship |
| Level 2 (On-track) | 2 operations | Explicit or mildly implicit | All given + 1-2 irrelevant pieces | Familiar, requires operation selection |
| Level 3 (Extended) | 3+ operations | Implicit (interpret before solving) | Some information must be inferred | Less familiar, requires relationship modelling |
A prompt that produces all three levels for the same mathematical topic:
"Write a differentiated set of word problems on percentage calculation for Grade 7. Same mathematical topic (percentage increase and decrease), three levels of cognitive demand.
Level 1 (Supported): A shirt costs $40. It is on sale for 25% off. How much does the shirt cost after the discount? [All information given, one operation, familiar shopping context, explicit question.]
Level 2 (On-track): A shop is having a sale. A jacket originally priced at $120 is reduced by 30%. The shop also sells a pair of trousers for $85, reduced by 20%. How much more does the jacket cost after the discount than the trousers after the discount? [Two operations, requires sequencing the discount calculation for each item, one extra piece of information — the trouser price — which students must use correctly.]
Level 3 (Extended): A mobile phone's price increases by 15% in January. In March, the same phone is on sale for 20% off the January price. Is the March sale price higher or lower than the original price? By what percentage? [Three operations — calculate January price, calculate March price, compare to original and express as percentage; requires multi-step planning; question asks students to determine what they need to find rather than stating it directly.]"
The same mathematical content (percentage change) appears at all three levels. The number range is comparable across levels. The difference is entirely in the problem structure — the number of steps, the explicitness of the question, and the amount of inference required.
A Classroom Scenario: A Three-Level Friday Task Set for Grade 6
Say you teach Grade 6 mathematics to a class with a wide ability range — some students are still consolidating multiplication and division of whole numbers, while others are ready for multi-step problems involving fractions. You could use a three-level word problem set every Friday as an end-of-week practice task.
A preparation workflow (15 minutes, generates a full Friday task set):
You identify the week's mathematical focus — say this Friday it's ratio problems. Your prompt:
"Write a differentiated three-level ratio word problem set for Grade 6. All problems use familiar local contexts: market pricing, food preparation, school supplies.
Level 1 (3 problems): A trader sells 5 oranges for 3 cedis. Find the cost of 15 oranges. [Direct ratio scaling, all information given, one step, explicit question.] Write 3 similar problems.
Level 2 (3 problems): A recipe uses 2 cups of flour to 3 cups of water. A baker has 10 cups of flour. How much water does she need, and how many 'portions' of the recipe can she make? [Two-step, requires finding the unit ratio then scaling; 'portions' question adds an additional layer.] Write 3 similar problems.
Level 3 (2 problems): A school tuck shop sells juice, water, and biscuits in the ratio 5:3:2. On Monday, the shop sold 120 items in total. On Tuesday, demand for juice increased — the shop sold the same total number of items but the ratio changed to 6:3:1. How many more juice items were sold on Tuesday than on Monday? [Three steps: find each day's juice sales using the ratio, then find the difference; requires interpreting ratios as proportional parts before any calculation.] Write 2 similar problems.
Provide full answer keys for all three levels, with all working steps shown."
The three-level set serves your mixed-ability class entirely: Level 1 students complete problems they can access successfully, building confidence; Level 2 students practise the standard curriculum content; Level 3 students extend to the problem structure they'll encounter at Grade 7. All students work on the same mathematical topic (ratio), none experiences the stigma of "the easy worksheet."
According to RAND Corporation (2025), differentiated task design — where all students access the same mathematical content at different entry points — is more equitable and more effective for long-term mathematics achievement than ability grouping with different content, because it maintains high expectations across all groups while varying the support level.
Differentiation Beyond the Three-Level Framework: Context and Readability
Two additional differentiation parameters that are easy to specify in AI prompts but often overlooked:
Sentence Length and Vocabulary Complexity
For students with reading difficulties or English Language Learners, word problem accessibility is often constrained by sentence length and vocabulary, not mathematical complexity. A student who could solve the mathematics can't access the problem because of unfamiliar vocabulary or sentence structures that are too long to parse.
"Write Level 1 ratio word problems for Grade 6 with accessibility modifications: maximum 20 words per sentence; avoid words with more than two syllables where possible; use only concrete, familiar vocabulary. Do not simplify the mathematics — simplify the language only."
This prompt produces mathematically appropriate problems that are linguistically accessible — which is the right form of language-based differentiation.
Context Relevance
Students who don't recognise the context of a word problem are solving a knowledge problem alongside a mathematics problem — understanding what a "rail fare" is adds a comprehension burden to a percentage calculation. Specifying familiar local contexts removes this unnecessary barrier.
"Write Grade 7 percentage word problems set only in contexts familiar to secondary school students in urban Nigeria: mobile data recharging, bus fare pricing, hawker pricing, school fee calculations. Avoid: banking, property, investment, international trade."
Using AI to Generate Scaffolded Versions of Existing Problems
Sometimes the teacher has a good word problem from a textbook or exam paper and needs differentiated versions — not all-new problems. AI can generate scaffolded adaptations of an existing problem efficiently.
"Here is a Grade 7 word problem [paste problem]. Create three versions:
Version A (supported): Add visual scaffolding — break the problem into numbered steps with blanks: 'Step 1: Find the ratio of boys to girls: ___ : ___. Step 2: Find the total parts: ___. Step 3: Find how many students are in each part: ___ ÷ ___ = ___. Step 4: Find the number of boys: ___.' Same numbers as the original.
Version B (standard): The original problem, unmodified.
Version C (extended): Add one additional question that requires using the answer to the original problem: 'What percentage of the total students are girls? Is this percentage greater than or less than 60%? By how much?'"
This three-version approach from a single existing problem takes 5-7 minutes and provides a complete differentiated set without requiring the teacher to design new problems from scratch.
Strand-Specific Differentiation Prompts
Fraction Word Problems (Grades 4-6)
Fraction word problems have specific structural features that determine their difficulty level — the problem type (part-of-a-set, part-of-a-whole, fraction of a fraction) matters as much as the size of the fractions.
"Write 9 differentiated fraction word problems in 3 levels (3 per level) for Grade 5. Level 1: part-of-a-set problems with unit fractions (1/3 of 12 books). Level 2: non-unit fraction of a set, answer not a whole number (3/4 of 10 students — answer is 7.5 — students must interpret this in context). Level 3: fraction of a fraction applied to a real scenario (2/3 of the students who brought lunch boxes are vegetarian — what fraction of the total class is vegetarian if 3/5 of the class brought lunch boxes?)."
Geometry Word Problems (Grades 6-8)
Geometry word problems differentiate naturally by the number of shapes involved and the number of properties students must recall or derive.
"Write 9 differentiated area word problems in 3 levels for Grade 6. Level 1: find the area of one rectangle given length and width. Level 2: find the area of an L-shaped composite figure (students must split into two rectangles, find each area, add). Level 3: a rectangular swimming pool with a 1.5-metre path around its perimeter — find the area of the path only. Given: outer dimensions. Students must find the inner dimensions, calculate both areas, subtract."
Statistics Word Problems (Grades 7-8)
Statistics differentiation focuses on how much statistical reasoning is required beyond calculation.
"Write 9 differentiated statistics word problems for Grade 7. Level 1: calculate the mean of 6 given values. Level 2: a student's test scores are given; the student needs a mean of 75 to pass — what score does she need on the 6th test? Level 3: two students' test scores are given, with one score missing for each. Both claim to have the same mean. Is this possible? Find a possible missing score for each student that makes their claim true, and explain whether there is a unique answer or multiple possibilities."
Pro Tips for AI Differentiated Word Problem Generation
- Always write Level 2 (standard) first, then derive Levels 1 and 3. The on-track problem is the curriculum-expected problem. Generate it first, then ask AI to scaffold it into Level 1 (breaking it into steps, removing unnecessary information demands) and extend it into Level 3 (adding steps, making the question implicit, requiring an additional layer of reasoning). This ensures all three levels address the same mathematical objective.
- Specify that all three levels should use the same contextual scenario. "Three students all read a problem about the same market scenario" is less cognitively disruptive than three students reading three completely different scenarios. Same context, different cognitive demand. This also makes whole-class discussion more accessible — all students can discuss the scenario even if they're solving different versions.
- Include a "starting sentence" scaffold for Level 1 problems. "Begin: 'First, I need to find...'" gives struggling students the direction they need without giving the answer. This is a lighter scaffold than breaking the problem into numbered steps, and is more instructionally independent.
- Generate the answer key for all three levels in the same prompt. A differentiated problem set with an incomplete answer key is difficult to use. Specify: "provide full worked solutions for all three levels."
- For Level 3 problems, verify the mathematics carefully before distributing. Extended, multi-step problems are the most complex AI outputs for mathematics — they involve the most calculations and the most reasoning steps, which means they have the highest error rate. Read through Level 3 problems specifically, working through the solution yourself.
What to Avoid
Avoid Differentiating Only by Number Size
Giving the Level 1 group problems with single-digit numbers and the Level 3 group problems with four-digit numbers is not meaningful differentiation if the problem structure is identical. A student who can't identify that a word problem requires multiplication will not benefit from smaller numbers — they'll make the same operation error whether the numbers are 4 × 7 or 48 × 73. Always differentiate by problem structure first; number size is a secondary adjustment.
Avoid Obvious Level Labels on Student Materials
A worksheet with sections labelled "Easy / Medium / Hard" or "Level 1 / Level 2 / Level 3" communicates ability grouping to students and can stigmatise students at Level 1. Use colour-coding, icons, or unlabelled differentiation — all students receive the worksheet with their level's problems presented as the task, with no visible indication of it being a differentiated version. Teachers manage the distribution; students see only the problems.
Avoid Differentiation That Changes the Mathematical Topic
A Level 1 problem that uses addition while a Level 2 problem uses multiplication is not a differentiated version of the same problem — it's a different problem. The mathematical topic (the specific skill being practised) should be identical across all three levels. Only the structural features (steps, explicitness, inference required) change. A student completing Level 1 should be building toward Level 2; if the mathematical operation is different, there's no developmental pathway.
Avoid Using Differentiated Problems Without Knowing Which Level Each Student Receives
Differentiated materials require the teacher to know which version of the problem each student is working on during marking. Anonymous submission makes differentiated marking impossible. Establish a simple tracking system: coloured paper, a letter code at the top of the page, or a seating-based distribution system. The marking of a Level 1 answer must be held to Level 1 expectations; a Level 3 answer to Level 3 expectations.
Key Takeaways
- Genuine word problem differentiation changes problem structure — number of operations, explicitness of question, amount of given information, context interpretation required — not just the size of numbers.
- The three-level framework (Supported / On-track / Extended) produces three structurally different problems on the same mathematical topic, enabling all students to engage with the same content at different entry points.
- Always generate the Level 2 (standard) problem first, then derive Levels 1 and 3 from it — this ensures all three levels address the same mathematical objective.
- Using the same contextual scenario for all three levels enables whole-class discussion across ability groups and reduces the stigma of differentiation.
- Level 3 problems require careful teacher verification — multi-step extended problems have the highest AI error rate in mathematics word problem generation.
- Avoid labelling differentiation levels on student materials — distribute by colour, icon, or seating without visible level indicators.
FAQ
What is the difference between tiered and differentiated word problems?
Tiered word problems are a specific type of differentiated task where all students access the same mathematical concept at different levels of cognitive complexity — three versions, three structural levels, same mathematical topic. Differentiated word problems is a broader term that includes tiering, readability modifications, language scaffolding, and context variation. In practice, most teachers use "tiered problems" and "differentiated problems" interchangeably for the three-level framework described here. For middle school tools that support differentiation across all strands, see AI Math Tools for Middle School Teachers.
How do I differentiate word problems for students with reading difficulties?
Specify readability constraints in the Level 1 prompt: maximum sentence length, familiar vocabulary only, no dependent clauses. Avoid simplifying the mathematics — the reading should be accessible, the thinking should still develop. Use sentence-level scaffolds for Level 1: "First, find ___. Then, use ___ to find ___." These sentence starters guide the problem-solving sequence without giving the calculation. For estimation problems that reduce reading demand while maintaining reasoning, see Using AI to Create Estimation Practice Problems.
Can I use AI to differentiate a textbook word problem I already have?
Yes — paste the existing problem and specify: "Create three versions of this problem: Version A: add numbered step scaffolds and remove any irrelevant given information. Version B: the original problem. Version C: add one extension question that requires students to use the answer from the original problem in a new context." This takes 5-7 minutes and works for any existing problem from any source. For more on AI-assisted assessment design, see the AI for Math Education: The Complete 2026 Guide.
How many levels of differentiation should I create for a typical class?
Three levels suits most mixed-ability classes — Supported, On-track, and Extended. A fourth level (Advanced/Gifted) is worthwhile if you have students who regularly complete the Extended level before other students finish On-track. Avoid five or more levels — the preparation overhead outweighs the differentiation benefit, and narrow level groupings create more stigma than broader ones. For comprehensive study materials that serve all levels, see Best AI Study Guide Generators in 2026.
For the complete AI in mathematics education overview, see the AI for Math Education: The Complete 2026 Guide. For middle school-specific AI tools that support differentiated instruction, see AI Math Tools for Middle School Teachers. For estimation problems that serve as natural Level 1 entry points, see Using AI to Create Estimation Practice Problems. For foundational place value differentiation at primary level, see Best AI for Place Value in 2026-2027. For comprehensive study guide generation that complements differentiated practice, see Best AI Study Guide Generators in 2026.