Summary
When we ask AI for more options, are we questioning our first idea or looking for a version we can commit to? I keep coming back to that distinction. We can question every detail of a solution without questioning the assumption behind it. And if every option starts from the same explanation of what users need, what have we actually given ourselves to choose between?
What are we counting as exploration?
“Let’s try a few more options.” And just like that, we carry on prompting, hoping the AI will come up with something better. We can reword the copy and generate different layouts but if every version rests on the same assumption of what users need, that never gets questioned.
Ten alternatives can still be ten versions of the same answer. We might have explored how to make an idea work without questioning why we chose it in the first place.
Rejecting every version wouldn’t necessarily change that. We could start again and still carry the same assumption into the next prompt.
Helpful for which part of the work?
A study published in September this year had 32 design students take on the same brief with and without AI.
Expert reviewers rated the work produced with AI higher for problem framing. But work produced without AI scored higher on innovation and proposed solution strategies. AI helped with parts of the task, not necessarily the parts we might assume. [1]
An earlier CHI experiment involving 60 participants found that those using an AI image generator stayed closer to an initial example and produced less varied ideas than those without an inspiration tool. The researchers examined design fixation: becoming attached to an example in ways that restrict what else we consider. [2]
Neither a student exercise nor a short visual task tells us how AI affects every designer. But these findings complicate the assumption that a process which feels helpful must also be expanding our thinking. [1]
More versions of what?
Take a hypothetical problem: people keep missing deadlines.
So we ask AI for reminder ideas. Earlier notifications, clearer wording, more frequent prompts. We now have several concepts to compare.
But we’ve already decided that people need reminding. Where did that explanation come from?
Someone might remember the deadline perfectly and still be waiting for information from a colleague. Someone else might have three tasks due at once and no authority to decide which comes first.
A reminder would not supply the missing information or resolve those competing demands. We would need to investigate the dependency or the priorities, not just improve the notification.
Those are different directions because they question why the deadline was missed. Changing when a reminder arrives doesn’t necessarily do that.
Of course, forgetting might be the problem. But it deserves investigation, not a free pass because it was the first explanation we reached for.
What assumptions are our options built on?
I’d try writing one sentence beneath each concept: “This would help because…”
If every sentence ends with “people forget their deadlines”, we’re comparing ways to act on one explanation. That may be intentional. But the important part is being able to recognise the difference.
We could ask AI for other reasons someone might miss a deadline. But a plausible answer still tells us nothing about whether it explains our users’ experience. To investigate, we could ask someone to walk us through the last deadline they missed: what they were waiting for, what else needed doing, and what prevented them from finishing.
Then we have an actual account to examine, rather than a plausible story we haven’t checked. We can compare it with other missed deadlines, looking for recurring obstacles rather than assuming one explanation fits everyone.
And if the evidence points back to reminders, we haven’t wasted the exploration. We’ve found a reason to pursue them beyond “that was our first idea”.
When you last asked for “more options”, did you need another version or another way of understanding the problem?
Sources
[1] Lin’s experiment on AI support in design ideation. PLOS One, September 2026. Read the PLOS One study.
[2] Wadinambiarachchi and colleagues. The Effects of Generative AI on Design Fixation and Divergent Thinking. CHI 2024. Read the research paper.
