ai assessment

Generating Question Banks for ESL With AI

EduGenius Team··20 min read

Watch the EduGenius tutorials playlist

Feature walkthroughs, setup help, and practical learning workflows connected to this article.

Open Tutorials

Generating Question Banks for ESL With AI

An ESL question bank needs one content-standard wrinkle and two real additions beyond the standard five tags: a WIDA/CEFR level tag that never becomes optional, and a home-language (L1) tag scoped only to selected-response items built on a named transfer pattern. Neither new tag works like music's asset-dependency tag. Nothing in an ESL item depends on a file that can go permanently offline the way a recording can.

Quick Answer: An ESL question bank inherits the standard five tags — content standard, Bloom's level, format, difficulty, last-used — with one wrinkle (content standard has to separate a language-focused item from a content item using linguistic modification) and two real additions: a WIDA/CEFR level tag, which is universal and can't be folded into difficulty without recreating the validity trap inside the tag itself, and a home-language tag, which applies only to selected-response grammar items carrying an L1-transfer distractor. The home-language tag isn't a durability risk like music's audio-source tag. A Korean-L1 article-omission item doesn't go stale when this year's roster has no Korean-L1 students — it goes dormant, fully valid, ready for the next Korean-L1 student a future roster brings.

The origin guide in this series built its case on L1-transfer distractors, WIDA/CEFR calibration, and the validity trap — a content quiz that accidentally tests English instead of knowledge. The series' own closing synthesis found those three parts collapse to one true constant (calibration), one format-gated technique (distractor design), and one derived risk (the trap). Neither guide asked what a standing bank, assembled across a whole school year, should tag by. This guide does.

This guide covers:

  • What an ESL bank inherits from general bank architecture, and the one wrinkle content standard picks up
  • Why music's external-asset tag doesn't transfer to ESL, tested directly rather than assumed
  • Whether a roster's changing L1 composition creates a durability risk the way a dead recording link does — the real, non-obvious question this guide exists to answer
  • Why WIDA/CEFR level can't be folded into a difficulty tag without recreating the validity trap
  • A full tag set, a step-by-step build workflow, and a tools comparison

This entry sits inside the same subject-by-subject series as How to Create Spanish Quizzes With AI, where each guide tests a shared format question against one subject's real structure.

What an ESL Question Bank Inherits From General Bank Architecture

An ESL bank starts from the same five tags any Bloom's-aligned bank needs, and most carry over with no ESL-specific rework. The Bloom's-aligned bank pillar set that baseline against a living table-of-specifications blueprint: content standard, Bloom's level, item format, difficulty, and last-used or exposure (Notar, Zuelke, Wilson, & Yunker, 2004).

Four Tags That Transfer Without Real Modification

Bloom's level, item format, difficulty, and last-used behave the same for an ESL item as for any subject already tested in this series.

  • Bloom's level doesn't discriminate an L1-transfer item from a linguistically modified one any more than it discriminates a biology recall item from an inference item.
  • Item format (MCQ, true/false, short-answer) carries over cleanly, though — as the next section shows — a bank can't sort by format alone.
  • Difficulty tracks content difficulty or pass rate, unchanged in principle, though it needs a firm boundary from proficiency level, covered later in this guide.
  • Last-used/exposure works identically to any subject: it tracks whether a student has already seen an item, nothing ESL-specific about it.

The Wrinkle: Content Standard Has to Separate a Language Item From a Content Item

A concept tag like "article usage" or "past tense" only makes sense for an item that's actually testing grammar. A science or social studies item written for the same students isn't testing a grammar rule at all — the origin guide's linguistic-modification section covers rewriting a content item's sentence structure while holding its subject difficulty constant (Abedi & Lord, 2001).

That's a real wrinkle on content standard, not a new tag: it needs to record which of two lanes an item sits in — a language-proficiency standard (an ELD strand or a CEFR can-do descriptor) for grammar and vocabulary items, or a subject standard plus a "linguistically modified" flag for content items.

Two items can share a grade level and a topic — both about the water cycle, say — while one is a language-focused vocabulary check and the other a linguistically modified content item. Without this wrinkle, a bank can't tell a teacher which lane either one belongs to.

Testing the Obvious Candidate: Does ESL Need an Asset-Dependency Tag Like Music's?

The obvious next move copies music's bank guide directly: music needed a stimulus-asset tag for notation images and an audio-source tag for recordings, because neither dependency was nameable from content standard alone. Before assuming ESL needs the same shape, it's worth running music's own test against an ESL item instead of borrowing the conclusion.

Does an ESL Item Depend on Something That Can Go Permanently Offline?

Music's tags exist because a notation item's correctness is inseparable from one specific, sourced image, and an ear-training item's correctness is inseparable from one specific recording. An ESL item fails that test on the same grounds the coding bank guide already used for a code snippet.

  • An L1-transfer distractor is plain text, fully contained inside the item stem and answer choices, exactly like the rest of the item.
  • Nothing about it points outside the document the way a "see attached staff excerpt" instruction does.
  • There is no equivalent of a lost recording. An article-omission distractor doesn't reference a file that a subscription or a shared drive could stop serving.

So the direct parallel fails the same way it failed for coding: ESL has no analog to a staff image or an audio clip. That's a genuine answer, not a shortcut past the question. But it doesn't close the case the way it did for coding, because ESL has one property coding's snippets never had.

PropertyMusic's Audio DependencyESL's L1-Transfer Distractor
Is the item self-contained text?No — depends on an external recordingYes — the distractor is written directly into the item
Can it be re-verified for free?No — a lost recording can be permanently unrecoverableYes — the transfer pattern is checkable against the linguistics literature at any time
Does something about the item change year to year?The access point (subscription, shared drive) can lapseThe item never changes, but the roster it's relevant to does

That third row is the real question this guide has to answer, and it's not one music or coding's precedent settles on its own.

The Real Question: Is Home-Language a Hidden Version of Music's Durability Risk?

A class roster's L1 composition changes every year, so a "Korean L1, article-omission" distractor is only immediately useful for a future class that also has Korean-L1 students — and that dependency on roster composition is the one property that makes ESL genuinely different from coding's fully portable snippet. The question is whether that counts as a form of staleness the way a dead audio link does, or something structurally different.

What Actually Changes When a Roster Changes

Say a bank holds a Korean-L1-tagged article-omission item, built and verified against a Korean-Entering-level roster. Two years later, the roster shifts to majority Spanish-L1 students. Spanish has a full article system, so the same item's distractor no longer traces to any real transfer pattern for this year's class.

That sounds, at first read, like the item just broke. It didn't.

  • The item's linguistic claim is still true. Korean has no grammatical articles (Odlin, 1989), and that fact doesn't expire.
  • The item still works perfectly for any Korean-L1 student on any roster, this year, next year, or a decade from now, at any school.
  • What changed is eligibility, not correctness. The item didn't degrade; it simply isn't the right tool for this particular class this particular year.

The Difference Between "Gone" and "Not Currently Eligible"

A lost recording is the whole failure mode music's audio-source tag exists to catch — there's no replacement it can regenerate on its own, and a subscription lapsing takes the item down with it, permanently, for every future class regardless of who's on the roster. An L1-tagged distractor never enters that state.

A dead recording is gone for every future class, forever, no matter who's in the room. A dormant Korean-L1 item is gone for exactly nobody — it's sitting in the bank, fully correct, waiting for the next Korean-L1 student a roster brings, whenever that happens to be.

That's the structural difference the roster-composition question was really testing. Music's risk is about an item's own survival. ESL's L1 dependency is about an item's targeting, which is a job content standard already does for every other subject — a fractions item isn't "stale" the month a class is studying geometry instead, it's just filed for when fractions comes back around.

Why It Still Needs Its Own Tag

Content standard can't absorb this targeting job on its own, though, because L1 isn't a curriculum concept the way "fractions" or "article usage" is — it's a demographic fact about the students an item is built for, and two items can share an identical content-standard tag (both "article usage") while serving completely different, non-overlapping L1 populations.

Skipping the tag doesn't make an item wrong. It makes it unfindable when it's finally relevant again, and, more importantly, it recreates the origin guide's validity trap one level up: an untagged article-omission item can get pulled for a Spanish-L1 student, whose wrong answer then gets misread as a content gap or a careless mistake instead of what it actually is — an item that was never diagnostic for that student's L1 in the first place.

WIDA/CEFR Level: The Tag That Can't Be Folded Into Difficulty

WIDA or CEFR level looks, at a glance, like it might just be a form of difficulty, but folding the two together reproduces the exact conflation the origin guide's validity-trap section warned against, now baked into the tag itself. Difficulty in a bank context usually means content difficulty or a tracked pass rate. WIDA/CEFR level means linguistic-load calibration, an orthogonal axis (WIDA, 2020).

The Retrieval Test

Say a bank holds 220 items spanning grammar, vocabulary, and linguistically modified content across a school year. A teacher needs every item safe for a newly arrived WIDA Entering-level student, regardless of subject or content difficulty.

A difficulty tag can't answer that. An "easy" content item can still carry an embedded relative clause a WIDA Entering student can't parse, the exact validity-trap scenario the origin guide opens with. Only a level tag, held separately from difficulty, lets a teacher filter by linguistic load without accidentally filtering by content difficulty instead.

  • Content difficulty asks: how hard is the underlying skill or fact?
  • WIDA/CEFR level asks: how much English does the sentence carrying that skill or fact demand?
  • Collapsing the two hides exactly the gap the closing synthesis named as the framework's one true constant — present in every format, never optional, and consistently found to matter more under constraint, not less.

That's why WIDA/CEFR level earns its own tag rather than a value inside an existing one: it's the single dimension every sibling guide in this series found could never be safely skipped, and a bank that lets it hide inside "difficulty" loses the ability to check for that safely.

The Tag Set an ESL Question Bank Actually Needs

A workable ESL bank keeps the five general tags, splits content standard into a language-item/content-item wrinkle, and adds two new tags scoped to different jobs. That's a genuinely different shape from music's two tags, even though the count matches.

TagCarries Over Unmodified?Notes for ESL
Content standardWith a wrinkleRecords whether an item is a language-proficiency item (ELD strand/CEFR descriptor) or a content item with a "linguistically modified" flag
Bloom's levelYesDoesn't discriminate item lane; tag it anyway for the reasons any bank needs it
Item formatYesMCQ, true/false, short-answer; format alone doesn't say whether an L1-transfer distractor applies
DifficultyYes, kept strictly separate from levelContent difficulty or pass rate — never a substitute for WIDA/CEFR level
Last used/exposureYesSame purpose as any bank, any subject
WIDA/CEFR level (new)NoUniversal; applies to every item, language and content alike, never optional
Home-language/L1 (new)NoConditional; applies only to selected-response items carrying a named L1-transfer distractor

Constructed-response items — short-answer, oral checks — carry no home-language tag at all, since the closing synthesis found transfer patterns show up unprompted in a student's own writing rather than as a pre-built distractor. There's nothing to tag before the fact.

Building an ESL Question Bank: A Step-by-Step Workflow

Building the bank starts with sorting each new item into its content-standard lane before writing it, since that sort decides whether a home-language tag applies at all.

  1. Decide the item's lane first: a language-proficiency item (grammar, vocabulary) or a content item that will use linguistic modification.
  2. Name the WIDA or CEFR level before drafting, regardless of lane — this step never becomes optional, per every sibling guide in this series.
  3. For a language-proficiency item in a selected-response format, name the target L1(s) before building distractors, so each wrong answer traces to a real, checkable transfer pattern (Lado, 1957).
  4. For a content item, request linguistic modification explicitly — short sentences, no embedded clauses, a word bank — and skip the home-language tag entirely; it doesn't apply.
  5. For a short-answer or oral item, skip upfront distractor design altogether. Tag the WIDA/CEFR level and expected response length instead, then read the student's own output for a transfer pattern afterward.
  6. Tag every item on all seven fields at creation. Retagging a backlog once a bank reaches a few hundred items costs far more than tagging while the decision is still fresh.
  7. Re-check the home-language tag's relevance whenever a roster's L1 composition shifts, not because the item changed, but because a fresh filter now needs to surface a different dormant set.

Say a bank holds a dozen Korean-L1 article-omission items filed from a prior grade 5 roster. A new grade 5 class arrives with mostly Spanish-L1 and Arabic-L1 students. Those Korean items stay exactly as correct as the day they were written — they just don't surface in this year's pull, while a Spanish-L1 definite/indefinite-confusion set and an Arabic-L1 word-order set do.

You could use EduGenius to hold a class profile's WIDA or CEFR range and current roster's home-language notes, then draft a batch of L1-transfer items and linguistically modified content items in one pass, tagged by lane and level, before exporting the set to PDF, DOCX, or Google Forms.

Can AI Tag These Fields Accurately?

No dedicated accuracy research exists yet for classifying ESL items by home-language target, the same honest gap the science bank guide and the music bank guide found for their own new tags. The tagging-speed guide reports real figures for automated Bloom's-level classification specifically — roughly 72-73% for general-purpose AI tools, rising to the low-to-mid 90s for purpose-built classifiers (Kumar, Gulwani, & Singh, 2025) — but that evidence doesn't extend to L1-transfer tagging.

That gap matters less than it sounds, for the same reason the coding bank guide found it did: a writer already decides which L1 a distractor targets before drafting it. The tag is a byproduct of a decision already made, not a separate research problem waiting on a classifier.

Tools for Building an ESL Question Bank

No mainstream AI content tool ships with a home-language or WIDA/CEFR level field built in by default; the real difference is whether a tool holds those values across a whole drafting session or needs them retyped into every prompt.

ToolHolds WIDA/CEFR Level Across ItemsTags Home-Language for DistractorsSupports Linguistic ModificationStarting Price
EduGeniusYes, via a saved class profileYes, if the L1 is named in the promptYes, if requested explicitly25 welcome credits free; Starter $7.99/month (500 credits); Professional $15.99/month (1,000 credits)
General-purpose AI chat toolsNo, unless retyped each timeOnly if explicitly requestedOnly if explicitly requestedVaries by provider
WIDA's released ACCESS item samplesYes, by designNot applicable; not distractor-focusedYes, by designIncluded for WIDA Consortium member states
Spreadsheet or LMS native item banksFully manual; teacher enters every fieldDepends on platformDepends on how it's builtOften bundled with an existing LMS, or free

For the wider bank-tagging landscape across subjects, see How to Write Better Question Banks Faster With AI. For single-item ESL quiz design that this bank's tags get built from, see Generating Exams for ESL With AI and How to Create ESL Quizzes With AI.

Pro Tips for an ESL Question Bank

  • Tag the lane (language vs. content) before writing the item, not after — it determines whether a home-language tag applies at all.
  • Never let WIDA/CEFR level hide inside a difficulty field. They answer different questions, and folding them together reproduces the validity trap inside the tag itself.
  • Treat a dormant L1-tagged item as inventory, not dead weight. It's not costing anything sitting unused; it's waiting for the roster that makes it relevant again.
  • Re-run the home-language filter every time a roster changes, not just once a year, since a mid-year transfer student can shift which items are suddenly relevant.
  • Skip the home-language tag on constructed-response items entirely. The closing synthesis confirms there's no upfront distractor to name an L1 against in short-answer or oral formats.

What to Avoid When Building an ESL Question Bank

Copying Music's Durability Field Onto the Home-Language Tag

Music's audio-source tag needs an owned-versus-streamed field because a lost recording can be unrecoverable. A home-language tag doesn't need that field — an L1-tagged item never goes permanently unusable, it only goes temporarily irrelevant to a specific roster.

Folding WIDA/CEFR Level Into a Difficulty Tag

An easy-content, high-language-load item and a hard-content, low-language-load item can end up with the same difficulty value, hiding exactly the distinction a level tag exists to preserve.

Tagging Home-Language on Content Items

Linguistic modification, not a named transfer pattern, is the right tool for a content-area item's language load. A home-language tag on a science or social studies item records a dependency the item doesn't actually have.

Deleting or Retiring Dormant L1-Tagged Items

An item with no current roster match isn't broken. Retiring a Korean-L1 item because this year's class has no Korean-L1 students throws away work that's exactly as valid the next time a Korean-L1 student appears, on any future roster.

Key Takeaways

  • An ESL question bank needs one wrinkle and two real new tags: content standard splits into a language-item/content-item lane, WIDA/CEFR level is universal, and home-language is conditional on selected-response distractor items.
  • The direct parallel to music's asset-dependency tag fails on its own terms. An L1-transfer distractor is self-contained text with no equivalent of a staff image or a recording that can go permanently offline.
  • A roster's changing L1 composition is a targeting question, not a durability risk. A dormant L1-tagged item stays fully correct; it just isn't relevant to this particular class this particular year.
  • The gap between "gone" and "not currently eligible" is what actually settles the comparison: a dead recording is unrecoverable for every future class, while a dormant L1 item reactivates correctly the next time a matching roster appears.
  • WIDA/CEFR level can't be folded into difficulty without recreating the validity trap inside the tag itself — they measure linguistic load and content difficulty, two different things.
  • Skipping the home-language tag reproduces the validity trap at bank-retrieval scale: a mismatched item can get pulled for the wrong student, and a wrong answer gets misread as a content gap instead of an irrelevant distractor.
  • No published accuracy research yet covers automated home-language tagging, the same honest gap this series found for music's and science's own new tags.

Frequently Asked Questions

Does an ESL question bank need a tag for external files, the way a music question bank needs one for audio and notation?

No. An L1-transfer distractor is self-contained text with no equivalent of a staff image or an audio recording — nothing about it points to an external file that could go offline. The real additions an ESL bank needs are a WIDA/CEFR level tag and a home-language tag, neither tied to an external asset.

Does an ESL question bank item become outdated when a class roster's home languages change?

No, not in the way a stale audio link does. An L1-tagged distractor stays linguistically accurate regardless of who's currently enrolled; it just stops surfacing in a retrieval pull until a future roster includes a student who shares that L1, at which point it's exactly as usable as the day it was written.

Should WIDA/CEFR level be tracked as part of a difficulty tag?

No. Difficulty tracks content difficulty or pass rate; WIDA/CEFR level tracks linguistic load. An easy-content item can still carry language too complex for a lower-level student, the exact validity-trap scenario a merged tag would hide instead of catch.

Which ESL items need a home-language tag, and which don't?

Only selected-response items — multiple-choice or true/false — built on a named L1-transfer pattern like article omission or word-order transfer need the tag. Linguistically modified content items and constructed-response formats like short-answer or oral checks don't, since neither builds an upfront, L1-specific distractor.

Sources

  • Selinker, L. (1972). Interlanguage. International Review of Applied Linguistics in Language Teaching, 10(1-4), 209-231.
  • Odlin, T. (1989). Language Transfer: Cross-Linguistic Influence in Language Learning. Cambridge University Press.
  • Lado, R. (1957). Linguistics Across Cultures. University of Michigan Press.
  • WIDA (2020). WIDA English Language Development Standards Framework, 2020 Edition. Board of Regents of the University of Wisconsin System.
  • Council of Europe (2001). Common European Framework of Reference for Languages. Cambridge University Press.
  • Abedi, J., & Lord, C. (2001). The language factor in mathematics tests. Applied Measurement in Education, 14(3), 219-234.
  • Notar, C. E., Zuelke, D. C., Wilson, J. D., & Yunker, B. D. (2004). The table of specification: Insuring accountability in teacher made tests. Journal of Instructional Psychology, 31(2), 115-129.
  • Kumar, R., Gulwani, D., & Singh, S. (2025). Automated analysis of learning outcomes and exam questions based on Bloom's taxonomy. arXiv preprint arXiv:2511.10903.
  • NCES (2024). English Language Learners in Public Schools. National Center for Education Statistics, Condition of Education.
#teachers#assessment#quiz