AI visibility reports, from 0 → 1 to 6× ARR
I designed Discover, Peek's AI visibility report, from the first screen. It shows property owners and operators how often AI assistants like ChatGPT recommend their buildings to renters. I took it from a brainstorm to a live demo in four weeks, then grew it into a full dashboard for three different users.
Product
Peek Discover
Role
Sole designer
Team
1 PM, 8 engineers
When
Oct ’25 → now
Platform
B2B web app
6×
ARR in ten months
74 → 280+
properties on monthly reports
18
features shipped, as the only designer
After launch
I synthesized 30 customer and sales calls into three gaps, each for a different user.
→ 01
What worked
AI-generated market survey
What was missing
Trust: some of its prompts were wrong
→ 02
What worked
A visibility score, competitors and sources in one report
What was missing
A next step: customers asked "So what do I do?"
→ 03
What worked
One report per building
What was missing
The big picture: teams with dozens of buildings opened each one separately
Feature
Custom Topics
Let customers fix the prompts, without breaking the data or the budget.
For marketing managers, who know what their building competes on. AI wrote the prompts behind each score, and some were wrong: a Denver building with no pool was scored on pool searches.
I matched each input to the data behind it, with dropdowns and set price steps, so every custom prompt returns real data. Prompt control shipped as a Premium feature.
The pivot
My first design locked the prompts, since knowing what renters ask was our expertise. But AI wrote some wrong ones, and opening all 35 to editing would cost about 25 AI queries each a month. So hiding wrong prompts became the main fix, and with product and engineering we capped new ones at 3.
Result
1 in 5 customers tuned their prompts, hiding 314 wrong ones so far. It became one of three reasons to buy Premium.
Before
After
Edit or hide any prompt the AI wrote, and add up to 3 of your own.
Feature
Recommendations
Give each user the fix in the form they'll use: copy for marketers, a CSV for developers.
For whoever makes the fix: marketers who paste copy, and the developers they pass technical fixes to. Customers could see their score, but not what to change.
Peek's AI scans each building's website and flags what keeps AI assistants from citing it. I kept each card simple: the fix on the left, copy to paste on the right, and a CSV that marketers forward to their technical team, the way they already hand off work.
I designed listing-site fixes as well, but that data wasn't ready, so the MVP shipped with website fixes first.
The pivot
The report's "Areas of Concern" list named problems but never the fix. Instead of another list, we retired it, so each problem sits on one card, beside its fix.
Result
It went live the week Discover's ARR jumped 44%. By the end of July, ARR had doubled again.
Before
After
A recommendation card: what to change on the left, the copy that fixes it on the right.
Feature
Portfolio View
One page for two readers: the managers who fix buildings, and the leaders they report to.
For managers who oversee dozens of buildings, and the leaders they report to. They opened each building's report one by one, then summed it up by hand.
I interviewed three customers, wrote the PRD and cut the MVP to three parts: the portfolio score, the building table and the top fixes. Leaders get a one-page PDF, and teams get the detail as a CSV.
The pivot
A "Needs attention" card would have pushed the building list further down the page. Instead, we reused the table's filters, already in the build, so filtering by Critical brings the worst buildings to the top.
Result
It became the home page for about 80 people at 40 property operators, 7 weeks after the first customer interview.
Filter by Critical, and the buildings that need attention rise to the top.
In their words
“
Interactive marketing specialist overseeing 170+ properties, on the portfolio overview
Expansion underway beyond their first 10 of 170+ buildings.
What I learned
Design around AI's limits, not its best case. Users shouldn't have to manage errors or cost, so I build in a way to correct the output from version one.
Also shipped
Self-serve checkout · Report summary · Competitor tracking · Free sample report · AI sentiment reporting · Navigation updates · Citation share summary · and more
