Smart List AI

Smart List AI

Helping sellers create higher-quality listings faster with AI-assisted workflows

Helping sellers create higher-quality listings faster with AI-assisted workflows

Role

Lead Designer

Platform

iOS, Android, Web

Scope

4-5 Months

Background

Background

Poshmark had no AI listing feature. I built one from scratch.

Before Smart List AI, creating a listing on Poshmark meant filling in every field by hand — title, description, category, brand, size, condition, price — with zero guidance and zero intelligence from the app. Poshmark had never built an AI-assisted listing feature. No competitors inside Poshmark's niche had one either.

The data told a clear story: listing publish rate was only 56%, with the sharpest drop-off right after sellers entered the camera view. Inexperienced sellers routinely paused mid-flow to search elsewhere for how to write a title or description — and many never came back.

Two problems to solve

"It takes too long. I don't know where to start."

  • 10+ required and optional fields, entered manually every time

  • Sellers paused mid-flow to research titles and descriptions offline

  • Casual sellers abandoned before ever hitting publish

I don't know what makes a good listing."

  • No photo guidance — sellers guessed what angles to take

  • No best-practice coaching on titles or descriptions

  • Poor listing quality hurt sell-through invisibly

What we learned from sellers first

What we learned from sellers first

Before designing anything, I ran seller interviews and usability tests. There was no prior AI listing feature to reference — everything we learned came directly from sellers using the manual flow. Four insights shaped every design decision that followed.

Before designing anything, I ran seller interviews and usability tests. There was no prior AI listing feature to reference — everything we learned came directly from sellers using the manual flow. Four insights shaped every design decision that followed.

  1. Drop-off wasn't a single point — it was everywhere.

Sellers didn't stall at one field. They stalled mid-title, mid-description, mid-price. Every blank field was a decision they felt unequipped to make.


  1. Sellers left the app to find answers.

Inexperienced sellers paused mid-flow to search elsewhere for how to write a title or description — and many never came back. The problem wasn't just speed; the flow itself was pushing people out.

3. No one knew what "good" looked like.
With no photo guidance and no best practices for titles or descriptions, sellers were guessing. Poor listing quality didn't punish them immediately — it hurt sell-through invisibly.

4. Remove decisions, don't just automate them.

This was the insight that shaped everything after: the value of AI wasn't filling in fields faster, but eliminating the decisions sellers felt least equipped to make.

1. Drop-off wasn't a single point — it was everywhere.

Sellers didn't stall at one field. They stalled mid-title, mid-description, mid-price. Every blank field was a decision they felt unequipped to make.Drop-off wasn't a single point — it was everywhere.

Sellers didn't stall at one field. They stalled mid-title, mid-description, mid-price. Every blank field was a decision they felt unequipped to make.

2. Sellers left the app to find answers.

Inexperienced sellers paused mid-flow to search elsewhere for how to write a title or description — and many never came back. The problem wasn't just speed; the flow itself was pushing people out.

3. No one knew what "good" looked like.
With no photo guidance and no best practices for titles or descriptions, sellers were guessing. Poor listing quality didn't punish them immediately — it hurt sell-through invisibly.

4. Remove decisions, don't just automate them.

This was the insight that shaped everything after: the value of AI wasn't filling in fields faster, but eliminating the decisions sellers felt least equipped to make.

1. Drop-off wasn't a single point — it was everywhere.

Sellers didn't stall at one field. They stalled mid-title, mid-description, mid-price. Every blank field was a decision they felt unequipped to make.Drop-off wasn't a single point — it was everywhere.

Sellers didn't stall at one field. They stalled mid-title, mid-description, mid-price. Every blank field was a decision they felt unequipped to make.

2. Sellers left the app to find answers.

Inexperienced sellers paused mid-flow to search elsewhere for how to write a title or description — and many never came back. The problem wasn't just speed; the flow itself was pushing people out.

3. No one knew what "good" looked like.
With no photo guidance and no best practices for titles or descriptions, sellers were guessing. Poor listing quality didn't punish them immediately — it hurt sell-through invisibly.

4. Remove decisions, don't just automate them.

This was the insight that shaped everything after: the value of AI wasn't filling in fields faster, but eliminating the decisions sellers felt least equipped to make.

How we went from nothing to a shipped product

How we went from nothing to a shipped product

Research first — understanding the manual flow

Before writing a single spec or wireframe, I ran seller interviews to understand where the current listing flow broke down. The biggest finding: sellers didn't stop at one point — they stopped everywhere. Mid-title, mid-description, mid-price. Every blank field was a decision they felt unequipped to make. The feature needed to remove decisions, not just automate them.

Solving the latency problem by design

A major technical blocker emerged: generating a listing took 5–20 seconds after the seller tapped "Done." That was a non-starter. I redesigned the photo flow so ML processing started the moment a seller uploaded their first photo — not after Done. Adding more photo angle guidance (3→5 per category) bought ML more processing time while improving listing quality. By the time sellers tapped Done, the listing was nearly ready. A dead wait became an active, purposeful step.

Measuring what mattered — Click-to-Publish vs. Classic List

Solving the latency problem by design

We A/B tested Smart List AI against the existing manual listing flow. Every design decision was evaluated against one north star: did it improve Click-to-Publish rate without sacrificing listing quality? The results confirmed the core thesis — sellers publish more, faster, with AI-suggested content they trust.

A major technical blocker emerged: generating a listing took 5–20 seconds after the seller tapped "Done." That was a non-starter. I redesigned the photo flow so ML processing started the moment a seller uploaded their first photo — not after Done. Adding more photo angle guidance (3→5 per category) bought ML more processing time while improving listing quality. By the time sellers tapped Done, the listing was nearly ready. A dead wait became an active, purposeful step.

Measuring what mattered — Click-to-Publish vs. Classic List

We A/B tested Smart List AI against the existing manual listing flow. Every design decision was evaluated against one north star: did it improve Click-to-Publish rate without sacrificing listing quality? The results confirmed the core thesis — sellers publish more, faster, with AI-suggested content they trust.

Design Solutions

1-photo quick-publish experiment

We tested a radically simplified flow where sellers could publish with just one photo. Publishing rate increased — but sell-through dropped significantly. Listings published this way were less discoverable and had lower conversion. Speed without quality was not the goal.

Pricing Confidence

Poshmark had never surfaced pricing data inline during listing creation. Sellers were guessing, leaving the app to search elsewhere, and losing confidence mid-flow. Research was clear: sellers didn't need a price recommendation — they needed to see why a price made sense.

I designed a pricing module that pairs an AI-suggested price range with real market evidence: comparable sold listings, brand strength signals, recent sales velocity, and condition context. The goal was to replace "I don't know what to charge" with an informed decision a seller could stand behind — without ever leaving the app.

Final results overview
Final results overview
Final results overview
Final results overview
Final results overview

Reflection

Together, these results showed that Smart List V2 wasn't just helping sellers list more—it was helping them complete listings with less manual correction.

Together, these results showed that Smart List V2 wasn't just helping sellers list more—it was helping them complete listings with less manual correction.

Copyright 2026 by Kathrine Huang