Which assignments stopped working, and why
AI for educators · Lesson 1 / 21
The fifteen-second test
Take the wording of one of your assignments, paste the whole thing into any chatbot, and look at what comes back. If you would grade that fifteen-second answer a B or better, the assignment no longer measures what you designed it to measure. It measures internet access. It is a diagnostic you can run across your whole assignment bank in one evening.
What broke
- The literature review or survey paper. "Describe the causes of X in five pages" is produced better than the average student manages, and faster than anyone manages.
- Summarising the textbook in your own words. A summary used to prove someone had read the chapter. Now it proves someone can copy.
- The standard end-of-chapter problem. If the problem exists online in a thousand variations, the worked solution arrives with step-by-step commentary.
- The general-topic essay. "The role of technology in education" is the ideal generation prompt: no data, no personal experience, no constraints.
What survived
What holds up are assignments that depend on things the model does not have: data from your own session, one specific person's mistake, a handout you gave out on paper, an argument that happened in the room on Tuesday. Anything requiring a choice that must then be defended holds up. Process holds up: the draft, the second version, the explanation out loud.
ASSIGNMENT BANK DIAGNOSTIC 1. Paste the wording into a chatbot. 2. Grade the answer on your own scale. 3. Tag it: "a bot can pass this" / "needs my material" / "needs process". 4. Rewrite the first category, do not delete it. 5. Count the share: if more than half fall in category one, start with the ones worth the most points.
Insight. A broken assignment is usually not a pointless one — it is simply written around an output instead of a process. The same material starts working again if you ask for a comparison of two solution paths and a justified choice rather than a finished text.
Common mistake. Responding with a ban on typed submissions. Going back to handwritten essays does not restore the meaning of the task; it just adds friction for everyone and punishes the students who write slowly.
Pro tip. Keep a file with your diagnostic results and the date on it. Run the same list again a year later: models change, and some assignments will migrate from the process category down into the first one. That is your own obsolescence curve, and it beats any forecast.
Cheat sheet
- One paste into a chatbot tells you more about an assignment than any hunch.
- Reproducing information broke; choice, process and local data survived.
- Rewrite the high-stakes assignments first, not the easy ones.
- Repeat the diagnostic annually — the obsolete list keeps growing.
1. What is the fastest way to check whether an assignment is obsolete?
2. Which assignments proved most resistant?
3. What should you do with an assignment a chatbot passes at B level?