Answers feel complete
Fluent responses can make us less likely to question them.
A polished answer.
Accepted before verified.

A browser extension that builds critical thinking habits into AI workflows without interrupting the task at hand.
Turn AI answers into a starting point for deeper thinking.
Deciding whether an answer deserves our trust is still difficult.
Generative AI makes information and content easier to access. But as answers become faster, more fluent and more complete, it can become harder to notice what needs checking, which perspectives are missing, and whether we really understand the answer.
Fluent responses can make us less likely to question them.
A polished answer.
Accepted before verified.
Searching again, comparing sources and asking follow-up questions all take extra work.
One answer.
Several detours to verify it.
More generated content does not necessarily lead to better judgement.


We wanted to understand more than whether people liked AI.
We wanted to know how they used it, trusted it and questioned it.
Focus groups helped us understand how people use generative AI for everyday learning and information seeking, and when they trust, doubt or double-check its responses.
What was missing was a moment to pause and reconsider the answer.
From what we could build
to what the product should do
Early in the project, we explored many possible features. Adding more did not make the product's value clearer. We revisited our research alongside practical constraints, narrowed seven candidate features down to four core capabilities, and defined the MVP for Critique AI.
Identify product opportunities through user research and analysis of 13 competing products.
Evaluate seven candidate features by impact and feasibility.
Focus on four core features that give the product a clear purpose.
Turning critical thinking into an everyday interaction.
Suggest new directions for inquiry, encouraging users to keep exploring beyond the first answer.
Show the sources behind detected claims so users can trace information back and check it for themselves.
Present opposing views with sources, helping users consider different positions and weigh their arguments.
Provide an ongoing workspace to save and organise research evidence and reflections across topics.
Good AI products start with more than what AI can do next.
The most valuable outcome for me was learning how to make choices among many possibilities. Through research, prioritisation and testing, we gave an open-ended AI brief a clearer product direction, with four core features to support it.