Across multiple research rounds, usability wasn't the problem. Trust was. Users could complete tasks fine. They just weren't sure they should trust the result. The product needed to feel credible before it could feel easy.

One insight was counterintuitive. AI outputs that arrived instantly felt less trustworthy — speed was undermining confidence. So we designed for what we called good friction. Structured questions before the draft. Smart defaults that show the AI knows your context. Visible reasoning behind every clause. By the second round of testing, the early trust concerns had resolved and the conversation moved to AI-output confidence — evidence the first layer was fixed.

The contract editor needed to feel like editing a real document, not reviewing AI output. We designed a smart fields system with clear states, and a review form that mirrors the document structure. Every AI-generated field is auditable: you can see what the AI filled, what it flagged, and what it left for you.

Early visual testing showed the palette felt too casual for a legal product. We redesigned the colour system — moved from green to a more measured purple, refined the accents, adjusted both themes. Negative feedback dropped by over 70%. Trust and modernity scores rose. The concerns disappeared entirely.