Desktop Reviews & topic visibility
Desktop set the production pattern: pipeline topics, sort unchanged, disclosure before interaction. Ops could monitor recurring themes without per-review manual tagging at launch.
AI Trending Topics
Surfacing recurring renter concerns from thousands of reviews — without replacing trust.
Designed a lightweight AI topic-labeling system that helped renters scan review patterns faster while preserving the credibility of original resident feedback.
Product Design · UX Research · Interaction Design
Overview
On high-volume property pages, renters struggled to spot recurring issues across hundreds of reviews. I introduced semantic topic labels inside the existing Reviews tab—patterns surfaced faster; resident feedback stayed the source of truth.
Problem
Renters scanned long threads for noise, maintenance, pests, and safety—while product and legal required AI help that would not weaken Verified Reviews. The problem was navigation and hierarchy at scale, not generated copy.
Renter experience
Newest-first feeds hid persistent themes. Without a query, people scrolled shallowly or left. Keyword search failed when wording varied for the same issue.
Product constraints
- Preserve original reviews and Verified hierarchy
- No AI summary blocks that replace resident voice
- Familiar Reviews layout and sort behavior
- Labels scalable from sparse to 100+ review pages
- Disclosure visible on first paint
Research & Discovery
Research paired analytics on Reviews depth and exit with moderated sessions on high-volume pages—mapping scan paths, mobile discovery, and where AI could assist without owning the narrative.
Methods
- Reviews tab scroll depth, topic taps, and exit rates
- Moderated tasks on 50+ and 100+ review properties
- PM, legal, and engineering constraint workshops
- Audit of summary-style AI patterns (rejected for trust)
What we learned
- Renters oriented via labels before committing to read
- Counts changed tap order more than label text alone
- Search refined an active topic—rarely started discovery
- Mobile needed the same hierarchy as desktop
Design questions
- Where do topics sit relative to sort and feed?
- What must disclosure say before first interaction?
- Global search vs. in-topic search—when each applies?
- Empty states when a property lacks label volume?
Key Insights
Insights came from observed scan and prioritization behavior—not feature requests.
- Renters scan for patterns first: they compare label counts before opening individual reviews.
- On mobile, discovery happens above the fold—topics must read as the primary hierarchy before the feed.
- Prioritization follows volume: higher counts reliably predicted which themes renters opened first in testing.
- Trust requires hierarchy: labels point downward to resident cards; nothing above the feed paraphrases them.
“I don't have time to read 100 reviews—I need to know if noise or maintenance keeps coming up.”Renter feedback, usability synthesis
Insight → Design Response
Each row ties observed behavior to a shippable decision and what we expected to move in Reviews.
User behavior
In sessions, renters paused on the label row and used counts to decide what to open—rarely reading cards in feed order first.
Design decision
Topic chips with counts fixed above the feed; active state shows current filter.
Expected outcomeShorter path to a themed subset before deep scroll—validated in moderated tasks on dense pages.
User behavior
When copy summarized residents, renters distrusted the tab and skipped to individual reviews anyway.
Design decision
No summary module; AI disclosure on first paint; Verified cards unchanged.
Expected outcomeMaintained compliance and avoided a second “voice” competing with resident content.
User behavior
Keyword search failed when renters did not know which term matched a recurring theme (e.g. “thin walls” vs. “noise”).
Design decision
Pipeline-driven semantic labels group varied phrasing under one chip.
Expected outcomeReduced failed searches and repeated scroll on the same underlying issue.
User behavior
Applying search before choosing a topic produced empty or confusing result sets.
Design decision
Topic filter first; search narrows within the active chip; reset restores full feed.
Expected outcomeClearer recovery when filters over-narrow—fewer dead ends in usability runs.
Strategy
Product tradeoffs—not a feature checklist. What we refused to build defined trust as much as what shipped.
No AI summary block
Rejected paraphrase renters could not audit line by line.
Semantic topic labels
Navigation via recurring themes; counts signal prioritization.
Topic-first discovery
Filter before read depth—in-topic search second.
Integrated into Reviews
One tab, one model—desktop and mobile parity for ops and analytics.
Solution
V1: filter states inside Reviews—labeled topics, active chips, in-topic search, reset—plus desktop visibility for pipeline-driven topics.
Before
- Newest-first feed; no cross-review themes
- Search only when renters knew the keyword
- Shallow scroll; early exit on dense pages
After
- Topic row establishes hierarchy above cards
- One tap to themed subset with visible count
- In-topic search + reset within Reviews
Topic labels as entry points
Active chip and count state made the current filter obvious. Renters validated labels by reading filtered cards—not by trusting label text alone.
In-topic search & recovery
Search applied only after a topic was selected—matching how renters refined, not started, discovery in research.
Mobile parity
Same hierarchy on mobile: topics → filter → cards. Filter state, sort, and disclosure shipped as one model—no forked mobile feature.
Impact
Post-launch we tracked Reviews tab behavior and moderated follow-ups—not generic “AI engagement.” Outcomes below reflect operational signals the team could observe.
Topic module interaction
Increased taps on trending topic chips vs. pre-launch baseline on high-volume property pages (analytics).
Less scroll-before-filter
In follow-up usability sessions, renters reached a themed review set in fewer interactions than unfiltered browse.
Trust signals intact
No change to Verified review placement; disclosure remained on first paint through launch QA.
Ops-ready labeling
Pipeline-driven topics reduced manual theme tracking in partner demos and support walkthroughs.
Reflection
The balance was AI-assisted discovery, clear content hierarchy, and business visibility into recurring themes—without letting the model speak for residents. Prioritization had to be honest (counts renters could verify), trust had to stay in the cards, and the business needed a system that scaled across properties. The strategic call was shipping navigation, not narration.