The 'Criteria for Discarding' When Choosing One of Three AI-Generated Ideas

* "AI Practice Notes for Streamers" #13

When you have AI generate three project candidates, you often get stuck because "none of them are bad." If you simply choose the one that seems likely to get the most views, you risk creating a video that deviates from your channel's core focus.

Last time, I organized how to choose the next three projects from your past 20 videos. This time, I will define the "criteria for discarding" when narrowing those three down to one. Do not leave the decision to the AI; instead, have it identify the gaps in each candidate.

The first to be discarded are projects without supporting material

Even if an idea sounds interesting, if you lack the evidence or experience on hand, that project should be put on the back burner. If you use AI-generated text to fill in figures you cannot verify, documents you haven't read, or questions you don't yet have the answers to, the video will end up feeling shallow.

This does not mean you will "never handle it again." Keep it in your project notes and return to it once you have gathered the necessary materials, research, or comments.

Compare the three candidates using the same five criteria

For each candidate, write one line for each of the following five points:

• Whose question does this video answer, and what is that question?

• Do you have the documents, experience, or past videos to support the conclusion?

• Can you state the video's conclusion in a single sentence?

• Are you repeating the same points as in your recent videos?

• Does it lead to the next video you want viewers to watch after this one?

If you choose without a comparison table, you will be swayed by the idea with the strongest title. First, line them up with the same questions to make the missing information visible.

Do not let the AI "choose." Make it find the gaps

Limit your instructions to the AI to assisting with comparison, not predicting popularity.

Rather than having the AI rank candidates from first to third, having it point out "what is missing" leaves the final judgment to the streamer.

Finally, choose an idea where you can imagine the title and the first 30 seconds

When choosing one, try writing a draft title and the one sentence you will convey in the first 30 seconds. Can you show at the beginning of the video what you promised in the title? If that is ambiguous, discard that idea for now.

Points to keep in mind when increasing the number of title ideas are also summarized in Part 6: How to Increase Title Ideas with AI Without Diluting Video Content.

Plans you put on the back burner are not failures.

For the remaining two, leave a note saying 'what is missing' and 'when to review them.' Waiting for materials, waiting for comments, or waiting for the right interval from the most recent video. If the reason remains, the material you pass to the AI next time will also increase.

The purpose of increasing plans with AI is not to create all three every time. It is to reach a state where you can confidently release the one that you can explain the best.

If you are going to charge for it, sell the comparison table, not the 'chosen answer'.

If you are making a paid article, a sheet that allows readers to compare their own candidates is more valuable than a list of correct answers for someone else's channel. By providing a set of five entry fields, instructions to pass to the AI, and options for reasons to put a plan on the back burner, readers can use it while retaining their own judgment.

The entrance to the entire series is Part 1: How YouTube Creators Can Reduce 'Downtime' with AI.

▼ Read also

・For those who can't remember what they did this week. 'Weekly Review' created with ChatGPT Work

https://note.com/poliplus/n/ncc28addb2b67

・AI operational design for deploying one video across multiple media—Judgment and responsibility left to humans

https://note.com/poliplus/n/nfa84bc37af7a

▼ Summary of this theme

・AI x Content Creation Know-how Summary

https://note.com/poliplus/n/ncd6194c20727