Instead of teaching people to prompt, we let CHALTTEOK ask first

"You know, something for Instagram. On⁠-⁠trend, but not too light. Copy for the new product." Say that to a colleague and they will usually come back with something good. Type the same words into an AI and the answer tends to feel flat.

CHALTTEOK is a prompt generator we made to narrow that gap. Here is what we saw as the problem, and why we decided not to teach people how to prompt.

The missing piece was context, not skill

When an AI doesn't give us what we want, we tend to blame the model or our own wording. But when we looked at what people actually typed, most requests were clear enough for another person to understand.

The difference was context. Who will read it, how long it should be, what form it should take. A person fills those gaps in, or asks. An AI doesn't ask. It fills them with the safest guess, and the result comes out bland.

No studying required

The usual answer is to teach prompting, and there are plenty of courses and templates for it. But that means finding time to learn, and then writing everything out by hand every time.

When LIGHT OF LIFE builds a product, we try to keep the technology out of the way. If a tool has to be studied before it is useful, we think the tool isn't finished yet. So we turned it around.

Instead of people adapting to the AI, CHALTTEOK asks for whatever is missing.

Asking, without turning it into a form

More questions give a more accurate result, but line up ten of them and most people close the window by the second. So we gave ourselves three rules for asking.

• Only ask what changes the result. If every answer leads to roughly the same output, the question goes.
• Show a suggested answer with every question, so people can pick instead of facing an empty field.
• Let people skip. Anything CHALTTEOK fills in on its own is marked in the prompt as an assumption.

We spent the most time on the third rule. Push for an answer and people leave; guess quietly and they can't tell why the result went wrong. Leaving the guesses visible means they only need to fix the ones that missed.

The finished prompt, in plain view

Everything gathered this way becomes a prompt a person can read. The offhand request above turns into this.

Role: social copywriter for a 20s-30s audience
Tone: on⁠-⁠trend, still credible
Goal: 3 options of Instagram copy for the new product
Limits: two lines each, hashtags included

We show the prompt along with the result for a reason. When you can see which line changed the outcome, you can adjust it yourself, and after a few tries you may start writing that way from the beginning. It wasn't built to teach, but there was no reason to stop people from picking it up along the way.

No copying, no rewriting

Copying a prompt into another window was another place where people stopped. So the prompt runs right where it was made. You can still send it to ChatGPT or Claude, but by default you never leave the screen.

The same went for results that missed. Asking people to explain in writing what was wrong just brings back the original problem. So instead, they choose a reason.

• Too long · Not enough substance
• Too generic · Wrong format
• Missed something important · Not sure it's accurate

Pick one and the prompt is refined in that direction. If you don't like it, you can undo it. Being able to undo is what makes people comfortable pressing the button, and pressing it is how the result gets better.

What we're still working on

The hardest part is still deciding what to ask. Too little and the output is bland; too much and people get tired. There is no clean rule for it, so we keep adjusting based on how CHALTTEOK is actually used.

CHALTTEOK is a beginning. Repeated explanations, complicated setup, screens you have to learn before you can use them: there is still a lot standing between people and what they want to do. We plan to clear it away one piece at a time.