Chapter 13. One passage, start to finish, and using AI without the slop
Using an AI without the slop
An AI can run these steps for you in seconds. That is the help. It is also the danger.
An AI makes smooth answers. A smooth answer is not a checked answer. I have said it before. I say it again, because it is the thing most worth remembering. If you read a smooth answer and feel at ease, check it twice.
Here is how to use an AI without adding to the noise.
What you give it
Give it three things.
The tests. The text of the tests, as I give them, in one long prompt. Appendix C says how to paste them in.
The run line. This is the first line. It tells the AI which tests to run. If you do not choose, the AI should say it is using the default. Appendix C gives the exact words.
The verse text itself. Copy it from a Bible you trust, and paste it in. Do this every time. If you only name a verse, the AI may recall it wrongly. A recalled verse is not a checked verse. When you paste it, the text counts as supplied.
I did not test which chatbots can take a prompt this long. Appendix C says so too.
What comes back
Expect a certain shape. It repeats your run line first, so you can see it ran what you asked. It gives the conclusion first, in a few sentences. Then it gives a part that shows its sources: what it was given, and what each test gives. Then it gives a part marked as judgment, kept apart. Then it lists what is still open.
Why keep them apart? So you can see what comes from the words and what is only the AI's opinion.
What comes back is a test result. It is not a verdict. It shows what the tests give on the words you pasted. It does not tell you what you must believe. Running the tests through a machine does not make them promise more. You can still be wrong with them, and right without them. The tests ask the AI not to flatter you. Watch for flattery anyway.
What it might look like
Here is a made-up example, so you know the shape. It is not from a real chat.
Selection received: the run line, copied back word for word.
Conclusion: the claim that Jesus said the disciple would not die is shown against by John 21:23. The claim that the disciple lived until Jesus came back is not shown.
Sourced: the verse text you pasted. Each test it used. What each test gave. Each limit.
Judgment: what the AI thinks the passage means. Marked as its own opinion.
Open items: what the words do not tell.
Read it from the top. Does each part say where it came from? If a part does not, ask.
Where an AI goes wrong
It goes wrong in the same ways people do, and faster.
It can add a step that sounds right. It can recall a verse wrongly. It can agree with you too soon. It can use a test outside its limit and not say so.
None of this makes the AI a bad tool. It makes it a tool. A saw cuts well and cuts the wrong board just as well. You hold the saw.
AI helped me make the tests in this book. It read the rules blind, raised objections, sorted findings, drafted wordings, edited my long argument and checked quotes. A person decided every label. I did. Appendix E tells it plainly. I say this so you know I am not asking you to do something I did not do. I checked, and I was checked.
Ask it twice
Here are three things you can ask after the first answer.
Ask it to quote the exact sentence it is answering. If it cannot, it may be answering a different one.
Ask it what would make its answer lose. The tests ask for this. A good answer can say.
Ask it to put the same two readings through the same tests, with the same words. If one reading gets a kinder answer than the other, find out why.
What you check
You check three things. Every time.
One. Every quote. Put each quote beside your own Bible. Do not check it against the AI. In my own work, quotes went wrong twice. Both times the editor, an AI, caught it. No reader caught the first, and my quote checker missed the second (chapter 12). If the quote is off, the steps built on it are weak.
Two. Every "shown". Go back to the words. Does the verse show that? Or was a step added? Here is an example. Suppose the AI writes that John 21:23 shows the disciple lived until Jesus came back. Look at the verse. It does not say that. It says Jesus did not say he would not die. So that "shown" is a step added. Put it back to "not shown".
Three. The limits and the labels. Did a test run outside its limit? Did the answer lean on a wording that is still being tried? Did it say a rule is "Passed" as if that were a reason? A label is never a reason.
One line to keep
A check that looks only for certain words can be walked around, and it cannot judge meaning. So do not trust an AI to check an AI by listing words. You read the meaning. You stay in the chair.

