Primary Science Inquiry with AI: From Topic to Record Sheet in Twenty Minutes

On a Tuesday after school in the staff room, Ms Chan, a P5 Science teacher, stares at a blank record sheet for next week's "Light and Shadows" unit. She types "torch height and shadow length" into Edor.ai; twenty minutes later she has question options, a variable frame and a record sheet — but she crosses out the AI's ready-made hypothesis and replaces it with "I predict: ______ because ______." She says, "The skeleton can be fast; the thinking stays with the pupils."

The new curriculum expects pupils to do science, not only memorise conclusions. AI shortens preparation time, but teachers must guard the parts pupils should decide — otherwise inquiry collapses into verification.

From topic to activity pack

STEAM links matter here: a shadow inquiry can connect to Art (silhouettes) or Maths (plotting height against length). Ask AI for one optional cross-curricular extension, then decide whether your week has space — the inquiry still stands if you drop it.

Turning one topic into a teachable inquiry usually means seven pieces: a measurable question, a hypothesis frame, independent and dependent variables, controls, stepped procedure, safety notes, a record sheet and an extension. A panel head might once have spent a whole afternoon; now a teacher enters the topic, grade, available materials and time limit, and AI can produce a full draft in about twenty minutes.

Speed is not the point. The draft must be investigable: the question measurable, results able to contradict the hypothesis, and pupils given blanks to fill. If AI fixes the hypothesis, controls and conclusion, rewrite immediately — that is not inquiry, only a prettified recipe experiment.

The seven-part generation sequence

Ask AI for these in order, then check each item:

Inquiry question — two or three measurable questions close to pupils' experience, not unverifiable prompts like "why is the sky blue". Hypothesis — sentence frames only, e.g. "If… then… because…", with blanks left empty. Variables — suggested independent, dependent and controlled variables, plus a column "What else would you control?" Procedure — numbered steps with rough timings for teacher trial. Safety — risks and alternatives; any heat, electricity, chemicals or sharp objects require a teacher trial. Record sheet — tables or chart frames with adjustable fields. Extension — follow-up questions, cross-curricular links or home inquiry, optional.

Variables pupils must choose

For "Which container cools water fastest?": AI might suggest material as the independent variable and temperature after ten minutes as the dependent variable. Delete any prefilled conclusion such as "metal cup wins". Leave only "prediction" and "actual result" columns. In class, ask: "What else should we control? Is one measurement enough?" — that is the soul of inquiry.

Classroom-desk inquiries

No dedicated lab is not fatal. Prompt for "classroom-feasible, everyday materials, no open flame" and AI tends toward paper-towel absorption, shadow-length measurement or phototropism with seedlings. On Edor.ai, teachers can save the seven-part draft to the Science panel library so colleagues teaching the same unit next term start from a trialled pack rather than a blank screen. P3 might ask which towel reaches the highest water line in thirty seconds, controlling water volume, towel size and time. P5 measures shadow length at torch heights 10, 20 and 30 cm and plots a line graph. P6 tracks stem tilt toward light over seven days with daily photos at the same time. None need special apparatus; each has clear variables and recordable data.

Three-minute pre-lesson trial

The floor for an inquiry is not "AI said it is safe" but "the teacher has done it once."

Primary pupils will follow steps literally. If AI writes "measure to the nearest millimetre" but your rulers show centimetres only, the lesson stalls in the first five minutes. Trial catches mismatches between AI precision and classroom reality before thirty pupils discover them together.

When trialling, check five things: are controls truly controllable; can P3–P6 pupils manage the measurements; will it finish within the lesson slot; do safety notes match your room (water near sockets, bright torches); could the result only confirm the hypothesis — if yes, change the question. Add a line at the top: "Trialled — step X changed to…" so next year's colleague saves time.

How the panel library grows

Activities that pass the check belong in the panel resource library: tag grade, topic and materials; attach trial notes, classroom photos and anonymised pupil records. When a P5 colleague opens "Light and Shadows" next year, they see Ms Chan's note — "Step 3 shortened to two minutes; pupils struggled to hold the ruler steady" — and adjust before the lesson instead of discovering the same friction on Monday morning. Next year's teacher takes it in one click and tweaks variables or columns for their class — that is panel knowledge, not a private craft. A panel head spending fifteen minutes each month curating new entries beats designing from scratch.

Common misreadings

• The claim: A complete seven-part AI output means the inquiry is done. What actually happens: Complete only means the scaffold exists. If hypothesis, controls and conclusions are filled, pupils follow steps. The teacher's job is to delete what pupils should decide and trial for feasibility.

• The claim: Inquiry requires a lab and special kit. What actually happens: Much primary science runs on a classroom desk. State constraints clearly and AI will use everyday materials; what matters is a measurable question and controllable variables.

• The claim: The more detailed the record sheet, the better. What actually happens: Too many columns and pupils never finish. Three to five core data columns plus "my explanation" is usually enough; leave the rest for discussion.

Three things to try on Monday

1. Pick one Science topic for next week, ask AI for a seven-part draft, and add to the prompt: "leave hypothesis and controls blank, classroom-feasible, no open flame".
2. Spend fifteen minutes trialling one version and write three trial notes at the top (what you changed, what to watch for).
3. Save the trialled version to the panel library with grade and materials tags, and delete any "model answer" beyond the hypothesis frame.