- Bring real material
- Choose the question
- Explore different angles
- Check the claims
- Select and make
Bring an observation into the conversation
I start a brainstorm with something real: a question someone asked, an observation from the work, an example worth explaining, or a claim I want to examine. Then I give that material to the model. If I only ask for marketing ideas, there is very little for it to work with beyond the familiar ideas it has already seen.
Explain who the work is for and what they are trying to decide. Include the material you can actually show: a process, an image you own, a public document, or an example with permission. Mark anything you are not sure about. That gives the brainstorm edges.
Ask for different arguments, not different headlines
A warning, a walkthrough, a comparison, a disagreement, and an experiment can explore the same observation in different ways. Ask what each would help the reader understand. Then compare the actual arguments rather than choosing whichever title sounds most confident.
Ask the model to point out what evidence each idea would need. If an idea depends on a customer result you do not have, it is not ready. If it depends on a claim that cannot be checked, change the idea instead of asking for more persuasive wording.
- What is the point of the piece?
- What example makes that point visible?
- What could a reasonable reader disagree with?
Choose with the work in front of you
Keep the original question, proposed idea, evidence, and intended next step together. You can then see when the polished version stops answering the question that started it. A fluent script can make that drift difficult to notice.
Choose the piece you can make well with the material available. It may be a short explanation or a substantial investigation. Length and production effort are consequences of the idea, not proof that it is important.
Let a person own the final meaning
Before publishing, read it as the person who will encounter it without the context of your prompt. What will they think is being claimed? What is shown as real? What is an illustration? Is there anything you would be uncomfortable explaining if someone challenged it?
Keep the useful responses after publication. They can reveal another question, a missing example, or a genuine disagreement. Bring that back into the next brainstorm. The system becomes more useful when it works with what you learned rather than producing another unrelated list.
Work an idea through instead of asking for more ideas
Suppose you notice that a generated customer reply sounds helpful but commits the business to something nobody approved. That observation can lead to several pieces. A walkthrough could show where approval belongs. A comparison could contrast draft-only assistance with automatic sending. An opinion piece could argue that a human review button is meaningless when the reviewer cannot see the source material.
Ask the model to explain what each piece would need to show. The walkthrough needs a clear process. The comparison needs explicit conditions so it does not claim that one arrangement is always better. The opinion needs an argument that can stand without a fabricated customer disaster. Looking at those requirements helps you choose a piece you can actually make.
Now ask for the strongest objection to the chosen argument. Someone may reasonably say that reviewing every routine reply would defeat the purpose of automation. That objection can improve the piece: define which replies are routine, what the system may commit to, and what happens outside those conditions. You have moved from a slogan about human judgement to a decision a reader can use.
Write the plain explanation, then decide the format. A short video might demonstrate the approval boundary, while an article holds the fuller workflow and sources. You can use AI to help arrange both versions without asking it to invent another argument for every channel.
Keep a record of what you rejected
Save the reason an idea was rejected or delayed. Missing evidence, a weak argument, and insufficient production material are different problems. Another round of titles will not fix all of them. A useful idea may become possible when you have a real example to show.
Before choosing the final version, compare it with the observation that started the work. If the connection has disappeared, either restore it or explain the new argument clearly. The decision belongs to you, even when a model produced most of the wording on the screen.
Sources you can check
The links support the platform details and technical distinctions. Recommendations and examples are editorial judgement.