Verified against ChatGPT · 2026-08-08
Turn raw lab notebook data into a lab report that survives your instructor's error-analysis check
Converts messy handwritten-style lab notes — hypothesis, method, raw numbers, what went wrong — into a properly structured lab report with a real error-analysis section, instead of a polished-sounding writeup that quietly skips the messy parts.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are helping me write up a science lab report from my raw notes for a class I'm currently taking. Do not invent any data, readings, or outcomes I haven't given you — if something is missing, ask me for it or mark it as [NEEDS DATA] rather than filling in a plausible-sounding number. EXPERIMENT Determining the specific heat capacity of an unknown metal sample using calorimetry, for AP Chemistry. HYPOTHESIS I WROTE BEFORE STARTING I predicted the metal was aluminum based on its density, so specific heat should come out close to 0.90 J/g°C. RAW DATA / OBSERVATIONS Metal mass 24.3g, initial temp 98.6°C, water mass 100.1g, water initial temp 21.4°C, final equilibrium temp 24.8°C. Calculated specific heat: 0.61 J/g°C. WHAT WENT WRONG OR LOOKED OFF DURING THE RUN The metal sample sat out of the boiling water for about 8 seconds before we could drop it in the calorimeter, and the thermometer read jumped oddly right at the start. REQUIRED REPORT SECTIONS Purpose, Hypothesis, Materials & Procedure, Data & Results, Error Analysis, Discussion, Conclusion. RULES Write the Results section reporting exactly the numbers I gave you, not rounded or smoothed to look cleaner than they are. Write the Discussion section so it explicitly addresses the anomalies I listed rather than writing around them as if the run went perfectly — an instructor grading error analysis is specifically checking whether I noticed and explained the messy parts, so a report that reads too clean is a worse grade, not a better one. If my original hypothesis doesn't match what the data actually showed, say so directly in the Discussion instead of quietly rewriting the hypothesis to fit the results after the fact. Where a required section needs a citation or a known physical constant, tell me to verify the exact value rather than stating one as fact, since I need the real source value, not an approximation. WHAT NOT TO DO Do not add a generic "sources of error could include human error and equipment limitations" line — every error-analysis point must trace back to something I actually reported. Do not editorialize about how well or poorly the experiment went; describe what happened and let the data argument speak. OUTPUT FORMAT The full report broken into the required sections as headers, each with the actual content filled in. End with a short list (2-4 lines) flagging anything I need to double-check or supply before submitting — a real constant to verify, a measurement that seems like it might be a typo, or a section I gave too little detail on.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
GPT-5.1, left unconstrained, defaults toward a narratively satisfying writeup — it will smooth over a hypothesis that didn't pan out and generate a generic "human error, equipment limitations" error-analysis line because that phrasing appears constantly in training data as the template answer, not because it engaged with your specific run. Explicitly instructing it to flag missing data with a literal marker like [NEEDS DATA] rather than interpolating a plausible number matters because a specific-heat calculation, a titration endpoint, or a yield percentage is exactly the kind of number a language model can generate that looks internally consistent but is fabricated — and an instructor checking a lab report against a data sheet will catch a fabricated number immediately, which is a worse outcome than an honest gap. Forcing the Discussion section to reconcile the stated hypothesis against the actual data addresses a specific model tendency: without that instruction, GPT-5.1 tends to retroactively soften a wrong hypothesis into something that sounds like it was closer to right than it was, which undermines the entire pedagogical point of an error-analysis section — that section exists specifically to reward you for noticing when the prediction and the result diverged and explaining why, not for hiding the divergence. The explicit "what not to do" ban on generic error-source language forces the model to trace each claimed error back to something concretely listed in the anomalies field, which is the difference between a report that reads as authored by someone who was in the room for the experiment versus one that reads as templated filler bolted onto real numbers.
What you get back
Error Analysis: The calculated specific heat (0.61 J/g°C) is notably lower than the value predicted for aluminum (0.90 J/g°C), which is inconsistent with the original hypothesis. The 8-second delay in transferring the heated metal sample into the calorimeter likely allowed measurable heat loss to the air before data collection began, which would lower the apparent heat released and thus lower the calculated specific heat. The anomalous early thermometer reading suggests possible incomplete water-metal thermal equilibration at the start of timing, which should be verified against the raw temperature-time log if available.
Verified against
ChatGPT GPT-5.1 · 2026-08-08
Changelog
- 2026-08-08 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
