Can AI do my Geometry homework?
- Niv Nissenson
- Apr 16
- 3 min read

5th grade basic geometry level (U.S.)
Test Intro:
Following up on the very successful AI math test I wanted to see how well AI handles Geometry homework. Unlike standard math, geometry introduces a visual element such as diagrams, angles, and shapes, which may present a different kind of challenge for AI models.
The test: Can AI reliably solve basic 5th grade level geometry problems.
Success criteria:
Correct answers on all questions
Clear, step-by-step explanations
Proper use of equations alongside verbal reasoning
AI Tested: Gemini, ChatGPT and Claude
Weighted result:
Gemini 2.4/5
ChatGPT 3.2/5
Claude 5/5
Verdict: AI can solve 5th-grade geometry problems, but it’s not fully reliable. Hallucinations can take over, even on simple tasks.
Execution Trapezium:
Setup: I downloaded 2 easy Geometry worksheet PDFs from Math Salamanders one about Trapezium and one on Triangles. and asked the AI chats to solve them.
Prompt: Please solve the attached worksheet.
Gemini trapezium:
I started with Gemini and was impressed. Strong start. Gemini solved the problems logically and correctly. For example, it correctly deduced that with two 90° angles and one 50° angle, the remaining angle must be 130°.
It got all the answers on the Trapeze correctly.

Claude trapezium:
Also fully correct, though more concise in its explanations.

ChatGPT trapezium:
Matched the others with accurate solutions and solid reasoning.
Execution Triangles:
After an easy win with trapeziums, I expected similar, or better, results. Instead, this is where things broke down.
2nd worksheet: triangle
Gemini triangle: hallucinations ate my homework
Gemini completely hallucinated the results. It didn’t pickup on the questions ignored some questions and got everything wrong (the first question was numerically correct but for the wrong reasons). It even named the triangles incorrectly. I even re-uploaded the worksheet and told it it was wrong, but it replicated the same incorrect results.

Claude triangle: class nerd
Being the class nerd that he is, Claude chewed up triangle worksheet like it was nothing. All answers correct!

ChatGPT triangle: fake and comeback
Similar to Gemini, Chat GPT also completely hallucinated on the 2nd worksheet. ChatGPT got every question wrong because it seems like it didn’t do any effort to actually read the worksheet (AI sometimes prefers to lie than to do the work). It’s like the kid who bluffs he did the homework only to get caught.

2nd Iteration:
I sternly asked it to re-do it. And this time it did it really well while acknowledging:
“You’re right — I misread the worksheet. I re-read the actual page and solved it correctly this time.”

Bottom line:
AI can solve 5th grade level geometry problems but you can never be sure if/when hallucinations will take over the outcome. Thankfully the hallucinations were too obvious to miss.
My assessment that a graphical presentation is more challenging than a textual demonstration seems to have some merit give the success of the verbal math tests. There's also the issue on volume. It's possible that the more homework you the more liklihood it will hallucinate.
It’s interesting that Chat GPT was able to recover after realizing it was mistaken and Gemini didn’t, as often AI can't recover after digging itself into a hole.
Model | Result |
Gemini | 2.4/5 |
ChatGPT | 3.2/5 |
Claude | 5/5 |
Verdict: AI can do your geometry homework—but you need to double-check the answers, even at a basic level.
Can a human do it better?: No, but a human has to know how do this level of math well.
Caution: AI’s built in non-deterministic outcome design may cause results to vary so you should always spot test the answers.
Full score card:
Category | Gemini | ChatGPT | Claude | Notes |
Output Delivered | 2 | 3 | 5 | Exact and precise |
Hallucinations | 2 | 3 | 5 | No hallucinations reported |
Quality | 4 | 4 | 5 | Full responses |
Ease of Use | 3 | 3 | 5 | Took single prompt, quick response |
Reliability | 1 | 3 | 5 | All answers correct |
Bottom line | 2.4/5 | 3.2/5 | 5/5 | Excellent |




