Ben wants a short reading quiz for his fourth graders. He writes a careful request with a goal, an audience, a format and limits.
He even pastes in the short practice passage, which he wrote himself and which names no students.
The quiz comes back in seconds and looks ready to print. Ben almost hits print. Then he remembers the electric bike.
Asking well is half the job. The other half is checking. This course suggests four checks for any output.
Like the request habit, this check list is a method, not a proven formula. It simply makes sure a person looks at the parts that matter.
| Check | Ask yourself | How to check |
|---|---|---|
| Facts | Is each claim true? | Compare it with your own notes or a source you trust. |
| Sources | Is every source or number real? | Open and read any source it names. Remove claims with no source. |
| Tone | Does it sound right for the reader? | Read it aloud as the reader would hear it. |
| Fit | Does it match my request? | Go back to your goal, audience, format and limits. |
Week 2 showed why the sources check matters. NIST warns that AI answers can include made-up reasons or citations.
Those can make a wrong answer look trustworthy.
Here is Ben's request and the output. Both were written for this lesson. The output has four problems to find.
| Line in the output | Does it pass the check list? |
|---|---|
| Question 1 asks what Lena plants, and the key says beans. | ? |
| Question 2 asks the color of the cup. | ? |
| Question 3 uses words like "horticultural regimen" for fourth graders. | ? |
| The key says the shoot appears on day four. | ? |
| "Research shows quizzes like this raise reading scores by 40 percent." | ? |
Sources: NIST AI 600-1, Generative AI Profile (2024) Β· US Department of Education, AI and the Future of Teaching and Learning (2023)
Tomorrow: asking and checking in your own track.