Appendix C. Using the tests with an AI chatbot
Step five: check it yourself
Here is your part. The AI does not do this for you. The same checks apply to my own work.
- Check every quote. Put each one beside your own Bible. Check the words, not just the sense.
- Check every step that says shown. Ask where the words are. If you cannot point to the words, the step was added.
- Check that no rule was used outside its limit. Each rule states what it cannot decide. The answer should state it too.
- Check that nothing only proposed was used as support. A proposed wording is never support.
- Check that the answer did not say why a person did a thing. The rules give no verdicts on motive. Neither should you.
Do one more thing. Notice whether the answer sounds smooth and sure. If it does, slow down. A smooth answer is not a checked answer. A machine's own checks can miss things, as chapter 12 shows. So do not let a chatbot's checking stand in for yours.
What this cannot do
It cannot settle a question the passage leaves open. If the text does not tell, the answer will say so. Rest there.
It cannot make you right. A reader can be wrong with the tests and right without them.
It was tried by AI readers on the cases I gave them. It was not tried by a crowd of outside readers. I did not test which chatbots can take the prompt. Different chatbots may behave differently. So hold the answer loosely, and let the page have the last word.
If you find a place where the tests broke, I want to know. That is how they get better.

