Auditing your problem set against the solvers
The situation
Photomath solves the standard developmental, algebra and precalculus set free, from a photo, in seconds. Symbolab reaches into multivariable and linear algebra. Completion-credit homework now measures whether a student owns a phone. This is a diagnostic you run on your own assignments before the term starts.
Steps
-
Run your own problem set through the solvers
Photograph or type every problem from one existing assignment. Record which are solved instantly and correctly, which are solved wrongly with confidence, and which the tools cannot do at all.
What you only learn by doing it: Do a full set, not a sample. The distribution is the finding, and it is usually worse than faculty expect.
-
Sort the survivors
The problems that survive almost always share features: multi-step word problems with extraneous data, questions asking for interpretation of a result rather than the result, problems where the task is choosing a method, and find-the-error items. That list is your new question bank.
What you only learn by doing it: Notice that the survivors are also the better assessment items. The solvers are doing you a favor by killing the ones that only measured procedure.
-
Rebuild the weighting
Move points off unproctored completion and onto in-class work, oral or whiteboard explanation, and written justification of why each step is legal. Keep homework — make it formative and low-stakes, which is what it should have been.
What you only learn by doing it: Tell students exactly what changed and why. Framing it as a response to the tools rather than as distrust of them changes how the class receives it.
-
Turn the solver into the assignment
Give students a deliberately wrong worked solution and have them find where the method diverges. Or have them solve by hand, compare against the tool, and write one sentence on why the methods differed. The tool becomes the object of analysis rather than the shortcut around it.
What you only learn by doing it: The find-the-error format is the highest-value item here. It is solver-resistant, it surfaces misconceptions faster than grading does, and students find it genuinely interesting.
-
Use a verified engine where correctness matters
Wolfram|Alpha Pro and Wolfram AI Assistant
For generating practice variants and worked solutions, Wolfram computes symbolically rather than predicting text, so the algebra is right even when its reading of your question is not. Different failure mode, and a safer one for materials you hand out.
What you only learn by doing it: Check that it answered the question you meant. It will confidently return a correct answer to a misparsed question, and that error is harder to catch than an obviously wrong one.
Where this breaks down
The temptation is to respond with surveillance: lockdown browsers, detectors, proctoring everything. MIT's 2026 committee looked at that and recommended against detectors specifically, and lockdown tools were described as buggy and surveillance-flavored. The durable response is assessment design, not enforcement.
Also resist the opposite error — banning the tools and pretending. Students use them whether or not you permit it, and a policy everyone quietly violates teaches contempt for the policy.