Most users asked one question and left
iAsk AI delivers cited, conversational answers, but most people asked one question, got it, and left. The experience itself made it too easy to stop and too hard to continue.
Most people asked one question, got an answer and left.
Reduce drop-off after the first answer, make refinement options easier to discover, and make asking a follow-up feel effortless.
Session depth mattered: it tied directly to engagement and to monetization linked to query volume.
What the analytics showed
Analytics showed two problems: filters existed but went unused, and almost nobody asked a second question. A competitive review of Google and Perplexity pointed to the same pattern: people continue when next steps are easy to find. That suggested friction in discoverability, not missing functionality. The usability sessions picked that up next.
What eight usability sessions showed
I ran 8 moderated usability sessions to understand the drop-off, and three patterns kept surfacing.
Discoverability: people didn't notice a follow-up question was even possible.
Mental model: the input read as starting a new search, not continuing the current one.
Uncertainty: it wasn't clear whether the previous context carried over at all.
People didn't notice a follow-up was possible, and the input looked like the start of a new search.
Three hypotheses
Between the analytics, the competitive review, and the usability sessions, three hypotheses took shape:
Less effort to ask a follow-up → more follow-up queries per session.
Easier-to-notice refinement options → more people try filters instead of leaving.
Clear follow-up prompts on the results page → people keep exploring instead of exiting.
Search entry redesign
Filters existed, but the dropdown holding them was visually unremarkable, so nothing invited a click. Getting to what shipped took two rounds of iteration, including one direction that tested well enough to try, then didn't hold up.
First iteration: widened the search bar and redesigned the dropdown trigger: icon, clear label, a more prominent button style. Filters stayed in a dropdown, just a far more noticeable one.
Second iteration (tested and rejected): pushed further to single-click filters, no dropdown at all. People could see them immediately, but it cluttered the main input too much. Tested against the first iteration, which won. One detail survived: the graduation cap on the logo when Scholar is active.
Final direction: the first iteration's dropdown layout, plus the graduation cap kept from the rejected version. It confirms the filter is on and reinforces that the product is built for students.
Before and after
Drag the slider to compare the original search bar with the shipped one.
Before
Final direction
Follow-up prompting
Two things shipped together: a sticky input anchored to the bottom of the page (Hypothesis 1) for people who already had a next question, and a related-questions module attached to each answer (Hypothesis 3) for people who hadn't formed one yet. Neither covered both cases alone.
Ship the sticky input and the related questions together. Neither covered both cases alone.
Sticky input (Hypothesis 1): anchored to the bottom of the page, for people who already had a next question.
Related-questions module (Hypothesis 3): attached to each answer, for people who hadn't formed a next question yet.
Results
Multi-question sessions went from 5% to 43%, and clicks on filters rose 34%. Related questions were used more often than typed follow-ups.
Not every hypothesis had the same effect. Hypotheses 1 and 3, the sticky input and the related questions, did more to get people asking additional questions than the search entry redesign did. The search entry changes still got filters noticed and used more.
Hypothesis 3 did the most. Both follow-up changes were tested, and people used the People Also Ask section more often than they typed a follow-up question of their own. All of it shipped in the same three-week sprint.
How it was measured. I compared the month before the release with the month after. It was a before-and-after comparison, not an A/B test, so other changes in that period may have played a part.