Win Elements · Teacher Guide · AI Tools
Turn Any Reading into a 40-Question Sleedu Quiz Bank with AI
Teacher GuideSleedu LMSClaude · ChatGPT · GeminiAbout 10 minutes per reading
Paste one prompt, attach one reading, import one file. Wrong answers give students a hint that sends them back to the reading — never the answer.
I teach full-time, so I don’t have hours to hand-write quiz questions for every reading. This page gives you the exact prompt and workflow I use to turn any reading — my free Skill Hub notes or your own materials — into a complete Sleedu question bank in about ten minutes.
1. What you’ll end up with
One .xml file you import into Sleedu in two clicks. Inside:
- 40 Cloze (Embedded answers) questions — dropdowns, short answers, and number blanks mixed inside sentences. All editable later in Sleedu’s normal question editor.
- Three rigor levels, labeled in every question name: Q1–Q14 are DOK 1 recall, Q15–Q30 are DOK 2 apply-and-explain, and Q31–Q40 are DOK 3 items built on new scenarios and numbers that are not in the reading — so answers can’t be found by searching the page.
- 10 visual questions with original diagrams built right into the question: charts to read, diagrams to label, and “spot the error” pictures drawn to match a common misconception. The picture carries the information — a student who skipped the reading can’t fake it.
- A hint behind every wrong answer. A student who misses sees something like “Re-read Section 2. Hotter air goes with which kind of particle motion?” — a pointer and a guiding question, never the answer.
Figure 1. The hint-not-answer loop. In Sleedu’s “Interactive with multiple tries” mode, a second-try success still earns two-thirds credit.
2. What you need
- A reading in text form. Any free notes page from the Win Elements Skill Hub works as-is — just download and attach it. Your own materials work too: a textbook section, an article, or notes exported to a text or HTML file. Richer readings (specific numbers, labeled figures, a “common mistake” callout) produce stronger banks.
- An AI assistant — Claude, ChatGPT, or Gemini all work. See the tool cards below.
- A Sleedu course where you can reach Question bank → Import. One label to know: Sleedu is built on the Moodle engine, so the import screen’s file-format option is called “Moodle XML format” — that’s the one you want. (Bonus: the same file and steps work on any Moodle site too.)
3. Pick your AI tool
The prompt is the same for all three. Only the buttons differ. One shared rule: pick the most capable model your plan offers — the ten diagrams are hand-written code, and smaller models sometimes draw pictures that are decorative instead of necessary.
Claude (claude.ai)
- Start a new chat and attach your reading with the paperclip.
- Paste the prompt, fill in your skill code, and send.
- Claude can hand back the finished
.xmlas a downloadable file — ask for it as a file if it prints code instead. - If the output ever cuts off, reply: “Continue from the last complete question, same format.”
ChatGPT (chatgpt.com)
- New chat, attach the reading with the paperclip or plus button.
- Paste the prompt and send.
- Ask for the result “as a single downloadable .xml file”; if it answers in a code block instead, use the block’s Copy button.
- Paste into Notepad or TextEdit (plain text) and save as
bankname.xml.
Gemini (gemini.google.com)
- New chat, upload the reading with the plus button.
- Paste the prompt and send.
- Gemini usually answers in a code block — use its Copy button.
- Paste into a plain-text editor and save as
bankname.xml(UTF-8). Don’t save from Word or Docs — they add formatting that breaks the import.
4. The six steps
Copy the prompt. Scroll to the prompt box at the bottom of this page and hit Copy prompt. Paste it into a new AI chat.
Fill in two blanks. Give the lesson a short code (
BIO9-2.03, USHIST-1.1 — anything that sorts nicely) and attach your reading file. That’s the only customization.Generate. The AI writes the full XML bank, then lists the questions with their DOK labels and which ten are visual. Skim that list.
Test-import before reviewing. In Sleedu: Question bank → Import → Moodle XML format → upload. A syntax slip shows up instantly at import — far cheaper to catch now than after reading all 40 questions. Then Preview one visual question to confirm the diagram renders in your theme.
Flip two quiz settings. These make the hint system work — see the box below.
Build quizzes from random draws. Don’t assign all 40. Add random questions from the category — I use about 13 per quiz (5 DOK 1 + 5 DOK 2 + 3 DOK 3). Every student gets a different mix, and the same bank serves the practice quiz, the retake, and the review.
The two Sleedu quiz settings that matter
1. Show the hints: Quiz settings → Review options → turn ON “Specific feedback” in the “During the attempt” column.
2. Allow the retry loop: Quiz settings → Question behaviour → choose “Interactive with multiple tries.” Students who miss see the hint, re-read, and try again in the same attempt. The bank’s built-in penalty is one-third per retry, so a second-try success still earns two-thirds credit — effort pays, guessing doesn’t.
5. Common pitfalls
Watch out
- One reading per run. 40 questions with 10 diagrams is near the ceiling of a single AI response. Don’t ask for three lessons at once.
- Re-importing duplicates, it doesn’t replace. If you regenerate a bank, delete the old question category first.
- Spot-check the DOK 3 items. They use scenarios not in the reading (that’s the point), so give those ten a quick sanity read — it’s where an AI is most likely to drift from your content.
- Save as plain text. If you copy code out of a chat, paste into Notepad or TextEdit — never Word or Google Docs — and save with the
.xmlextension. - Edit in Sleedu, not by regenerating. Every question is a normal Cloze question: reword it, swap a diagram for your own photo, or add an option right in the question editor.
6. The prompt
Copy everything in the box, fill in the skill code, attach your reading, and send. Prefer a file you can keep? Download the prompt as a .zip — unzip it and open the file inside in any text editor.
THE PROMPT · works in Claude, ChatGPT, and Gemini
You are an expert assessment writer. I will give you a reading (a notes page, textbook section, or article). Create a question bank for my Sleedu LMS. Sleedu is built on the Moodle engine, so deliver it as a single standard Moodle XML file, using Moodle Cloze (Embedded answers) syntax, that I can import via Question bank > Import > Moodle XML format.
Skill code: [e.g., SCI6-1.02]
Reading: [attached]
BANK STRUCTURE
- Exactly 40 questions, all type=cloze (Embedded answers) so I can edit text and swap images in the question editor.
- One $course$/top/... category question at the top, named after the skill.
- Question names: [SKILL] Q## - DOK# - short topic, so I can sort and filter.
- DOK split: Q1-Q14 are DOK 1 (recall of specific facts, numbers, and figure values), Q15-Q30 are DOK 2 (classify, explain, read figures, apply the reading's routine), Q31-Q40 are DOK 3 (evaluate misconceptions, transfer to NEW scenarios and NEW numbers not found in the reading).
- Every question must require having read the reading. Use its exact numbers, figure labels, section names, key terms, and common-mistake callouts. No generic trivia answerable without the reading.
VISUAL REQUIREMENT (most important)
Exactly 10 of the 40 questions must contain an original inline SVG visual inside the question text, and the visual must carry information the student needs to answer, never decoration. Text-only versions of these questions should be unanswerable. The other 30 questions are text-only.
Technical rules for the visuals:
- Write each visual as inline SVG directly in the question's CDATA (no external image files, no base64, no img src links) so the bank imports with zero file uploads and I can later replace any SVG with my own image in the editor.
- viewBox="0 0 540 H" with height as needed. Keep each SVG under about 8 KB.
- Clean style: white panels, dark green #0f3d2e, gold #8a6d00, blue #3a6ea5, red #c2452d, ink #22302a; Arial labels 10-12px. Visuals must stay readable printed in black and white (vary position, length, and count, never color alone).
- Label every element the question refers to (A/B/C tags, axis labels, arrows). Never put the answer itself in the visual's text.
Use a mix of these visual question patterns (at least 4 of the 5):
1. Read the figure: a chart or diagram like the reading's figures but with different values; blanks ask students to read values off it.
2. Which diagram is correct: two or three small labeled panels (A, B, C) showing competing models; a dropdown asks which panel matches the reading and why.
3. Label the diagram: a diagram with lettered callouts; each blank is a dropdown assigning the right term to each letter.
4. Interpret new data: a bar chart, table, or arrow diagram with fresh numbers; NUMERICAL blanks require computing from the visual using the reading's rule.
5. Spot the error: a diagram deliberately drawn to match a misconception from the reading; blanks ask what is wrong with it.
Spread the 10 visuals across the DOK bands: 2 in DOK 1, 4 in DOK 2, 4 in DOK 3.
ANSWER-BLANK RULES (Cloze syntax)
- Multiple choice: {1:MULTICHOICE_S:...} (shuffled dropdown). 3-4 options; distractors drawn from real misconceptions or misread figures.
- Exact words: {1:SHORTANSWER:=answer~=accepted variant~*#hint} (case-insensitive, with accepted variants).
- Numbers: {1:NUMERICAL:=value:tolerance~*#hint} with a sensible tolerance on decimals.
- Never use the characters } ~ # = inside any answer or feedback text (they break Cloze syntax). Write comparisons in words instead of using an equals sign.
- Wrap all question text in CDATA; HTML (subscripts, degree and times symbols, and the inline SVG) is allowed inside.
- Include defaultgrade equal to the number of blanks, penalty 0.3333333, hidden 0.
FEEDBACK RULES: HINTS, NEVER ANSWERS
- Every wrong option gets #Hint: feedback that (a) points to the exact place to re-read ("Re-read Section 2", "Find the second row of Figure 3"), (b) asks a guiding question, and (c) targets that specific mistake ("You read the wrong bar", "Right word, wrong reason"). For visual questions, the hint directs the student back to a specific labeled part of the SVG or the matching figure in the reading.
- Hints must never state or strongly imply the correct answer.
- SHORTANSWER and NUMERICAL blanks end with a ~*#Hint: catch-all for any wrong entry.
- The generalfeedback element is a neutral one-line "Where to look:" pointer (section or figure name only, no answers).
- Correct options get no feedback text.
QUALITY CHECKS BEFORE YOU DELIVER
- Verify every Cloze blank has exactly one correct option marked with = and parses cleanly (no stray } ~ # = inside options or hints).
- Verify the 40-question count, the DOK labeling, and that exactly 10 questions contain a load-bearing SVG (2 DOK 1, 4 DOK 2, 4 DOK 3).
- Verify the DOK 3 items use scenarios and numbers NOT present in the reading.
- After writing all 40 questions, re-scan every blank and hint for stray } ~ # = characters and fix them before outputting.
- Deliver one downloadable .xml file named [skill]-visual-cloze-bank.xml, then summarize: the DOK breakdown, which questions are visual and what pattern each uses, and remind me of the two quiz settings (turn ON "Specific feedback" in Review options; set Question behaviour to "Interactive with multiple tries").
Why hints instead of answers?
When a wrong answer reveals the correct one, the loop is guess → copy → move on — nothing was read. When a wrong answer reveals a location and a question, the loop becomes guess → re-read → reason → retry. Same quiz, completely different cognitive work. In my classes, a 13-question quiz in this mode fills a period with students arguing with the text — which is the point.
Free lessons, assessments & worksheets at WinElements.com · Unlimited online practice at Sleedu.com
