Stop asking questions. Start having conversations.
An ongoing investigation into writing with AI
In working through a stubborn plot problem recently, I noticed something unexpected.
The novel was alive in fragments. The characters had weight. The world felt textured. Several scenes already carried energy. But the engine of the story, the force that would make it move, refused to click into place.
For days I circled the same frustration: something is wrong, but I can’t see it.
I could have asked for plot ideas. I could have asked for twists. I could have treated the tool like an unusually articulate search engine.
Instead, I did something different.
I showed my working.
I laid out everything I already knew:
who the characters were,
what each of them wanted,
what had happened so far,
which tensions were unresolved,
which themes kept resurfacing,
what felt wrong but I couldn’t yet name.
It wasn’t elegant. It wasn’t a polished prompt. It was thinking in draft form.
Then the questions began. Not informational, but investigative.
What assumption are you making about the villain?
Who benefits if this goes wrong?
What are you protecting in this version of the story?
We circled. Refined. Reframed. Tested.
And then the villain clicked into place.
The idea didn’t feel given to me. It felt recognised.
The process didn’t replace my thinking. It accelerated something I would eventually have arrived at through solitary brainstorming, the slow massaging of half-formed possibilities until one finally cohered.
The dialogue compressed the wandering. It held the thread long enough for me to see the pattern sooner.
And that made me realise something.
What mattered was not that options were generated. What mattered was that my reasoning had been made visible.
In school maths, teachers insisted we “show our working.” At the time it felt unnecessary. If the answer was correct, why did the steps matter?
Now I suspect the steps are the thinking.
When I laid out my assumptions, contradictions, and half-formed ideas, I was not simply giving the tool more context. I was externalising the invisible parts of my reasoning.
Once visible, they could be examined.
That is what changed.
When I treat AI like Google, a place to ask isolated questions and receive answers, the output is efficient but shallow.
When I treat it as a conversational space, somewhere to explore uncertainty and test assumptions, the depth changes.
A question seeks an answer.
A conversation exposes reasoning.
The shift is subtle, but it alters the quality of collaboration.
So I am curious:
When did your AI collaboration move beyond answers and start changing the outcome?
What shifted?