Artificial Research

Overview

Powerful model, fragile trust

A startup had a genuinely good research model and a signup curve that flattened the moment trial users saw their first answer. People didn't doubt the technology — they doubted whether they were asking correctly and whether the answer deserved confidence.

I redesigned onboarding, the asking flow and the results screen around one principle: show your work. Sources, confidence and next steps became part of every answer.

Challenge

A black box nobody asked twice

Interview after interview surfaced the same pattern: users typed one careful question, got a confident paragraph, and left — unsure what had happened. Without provenance or guidance, the product felt like a party trick.

• Single-shot answers with no sources, no alternatives, no follow-ups.
• Empty-input paralysis — new users stared at a blank box with no examples.

Approach

Answers that show their work

Results now arrive as structured briefs: the direct answer first, then the sources it rests on, a confidence note, and three suggested follow-ups. Onboarding demonstrates all of this with the user's own first question, so the lesson sticks.

Two rounds of testing took the flow from "clever but confusing" to a trial-to-paid jump the founders could see in their weekly numbers.