Usable GenAI Outputs Don’t Come from Short Conversations

The Trap of the Immediate Output

When talking with generative AI chat tools, a lot of people think these tools know best. This leads to an assumption that iteration doesn't really need to happen because they will produce usable content right away.

I look at this through two lenses. One, it's a communication problem—you can craft strong prompts to get closer to a product you can use with reduced iteration. If you want a refresher on how to set up your initial prompts, take a moment to review my previous article covering the PTCF framework (Persona, Task, Context, Format).

The second lens is that iteration and critical thinking can take time to work through. These tools create exceptional drafts, but the more you dig and consider what you want, iteration naturally occurs. It is so important that you understand what you are doing and what you need to complete the task to actually guide the conversation.

These tools don't know what you want, what you do, or what you need—they are simply responding to your prompt. Iteration and long conversations aren't a waste of time if it means the outcome is usable.

A Real-World Experience: Recipe R&D

Let’s talk about this through a real experience. This week, I had picked fresh, tart Evans cherries at my neighbor's house—a variety that grows incredibly well here in Alberta—and I wanted them to be the star of the show.

I envisioned a brunch cake that wasn't too sweet, using dark chocolate as a secondary, complementary flavor. Crucially, I needed a cake structure that was dense enough to hold up to fresh, wet, juicy cherries without turning into a soggy mess, while still feeling light enough for morning eating. I didn't have a recipe and honestly I didn't want to look for one.

So I started to chat with Claude and then Gemini to develop a recipe. Part 1 of this was Claude—I was struggling to get usable recipe suggestions based on my ask or initial prompt and follow-up prompts. I felt it wasn't understanding me for this task very well, so I switched over to Gemini to see if I could get better results.

Part 2 was refining my vision of the recipe with Gemini. In total, it took close to an hour of a conversation to develop one recipe. Let that sink in. You might be wondering how this took an hour.

My process looked like this:

Dissecting the Dialogue

My original prompt started with assigning clear parameters:

"You are an exceptional French baker, who has worked at top restaurants around the world. Your client has asked you to make a cherry and dark chocolate breakfast type of cake. Their expectations are that it’s a little bit dense but also light in texture. There’s some moisture that might add to the denseness. No icing may be a crumble top, but it’s really critical that the texture of the cake can accommodate fresh cherries, diced subtle vanilla is OK as secondary options and outline the texture as this is most important and it’s critical that it’s not too sweet. Streuselkuchen is a good option. Give 10 options"

Prompts during the conversation included:

"Brioche is too bready. We are looking at cake of sorts. Nothing said suggests bready"

"I think chocolate better throughout"

"Compare against the German style"

"I think chocolate better throughout"

"Evaluate ratio of fruit to base"

"Increase fruit to 285g"

"I didn’t tell you change format of ingredients"

"Is this enough topping or do we need to increase the amount amounts to be enough for each individual cake?"

In reviewing these prompts you can infer that Gemini was providing suggestions or recommendations, but they didn't always fit my needs. I was constantly updating, informing, making decisions to guide the conversation.

Following the Goal as Your Compass

I was also questioning what was provided by Gemini, like asking if there was enough topping. I am the expert in building the recipe because I know what I want the outcome to be.

Your ultimate end goal is the compass in the conversation. The AI acts as the engine providing momentum and raw ideas, but your goal keeps the needle pointing north. Without keeping that target outcome locked in, you will just drive in circles—or worse, not know when you've actually arrived. When you hold tight to that clear objective, an extended conversation isn't a struggle; it's just the natural work of building something genuinely usable.

The Cake

My vision for this cake came to life and as soon as I took a bite, I surprised myself!

It was cherry forward, with dark chocolate for balance as a secondary flavor. It had the perfect texture: not too dense, soft, but holding up to the cherries remarkably well. The cornflake topping provided a great contrast in every bite.

Check out the recipe and give it a go before cherry season is over!