I'm sure there's a small psychological benefit! And it's more memorable for the brand experience.

Not sure how you'd measure if it works, but it probably won't hurt either, so no harm!

This started as a scrappy personal fix and slowly turned into a way I think about information architecture:

If people can’t easily access what they already have, the system is working against them. 🚫

I wrote more about how this came together below.

#DesignThinking #UserFlow

Most recipe apps treat recipes like documents. 🔎

What worked better for me was treating *ingredients* as first-class data. If I have an avocado, I should be able to quickly find recipes that use it.

Once retrieval got easier, creativity followed with less effort.

#InformationArchitecture #Systems

Because of this conversational AI design token drift issue, I built a structured prompting workflow (Figma recommends this) that cuts usage way down.

Turned it into a custom Universal Prompt Designer GPT so anyone can run the same brief pattern.

If you want the link, reply here and I’ll drop it.

Where does AI break for you most often?

Reply with:
1️⃣ Device/Viewport constraints (e.g., mobile fold)
2️⃣ Business constraints
3️⃣ UX details (states / flows / edge cases)
4️⃣ Other (tell me)

#ProductStrategy #UX #AIDesign #ProductManagement

Pattern‑matching isn’t strategy.

And strategy is what drives revenue, especially during Cyber Week when every percentage point matters.

Full case study (with screenshots of the fails + the fix):

The real risk with AI design?

Layouts that look ready to ship but don’t solve the right problem, quietly burning time on flows that don’t convert.

AI rule for founders + PMs: Be skeptical of polished AI output. Never use it to bypass the foundational research that drives your metrics.

The result we shipped:

→ ~–2% conversion drag (well within the –7% threshold)
→ Meaningful AOV + LTV lift that offset the drag
→ A new internal pattern the org reused for future design projects

That’s what rigorous iteration looked like, not just a pretty upsell.

Here’s what the winning human design actually needed instead (the strategic gaps AI couldn't fill):

→ Marketing psychology (reframing the choice)
→ Behavioral insight (from real session data)
→ Cross‑functional negotiation (to decouple pricing)
→ Mobile‑first strategy (intentional viewport design)

All three looked “shippable.”

With enough extra prompting, a couple could have looked great.

But none of them understood the business problem.

They optimized pixels, not outcomes.

(Full side‑by‑side breakdown + screenshots is in the Medium case study linked later in this thread.)

What happened 👇

- Figma Make → buried the profitable tier **below the fold** on mobile (~85% of traffic)
- Lovable → walls of text and comparison cards that increased decision fatigue
- v0 → best UI by far, but still missed the critical mobile viewport behavior

So the question was simple:
Could AI crack what four human designers couldn’t?

I gave Figma Make, v0, and Lovable the same detailed brief: user context, business constraints, and a hard –7% conversion‑drag threshold.

Context: I led product design for a D2C beauty brand where a single screen determined whether we hit our AOV (Average Order Value) and LTV (Lifetime Value) targets.

Four previous humans had already taken a swing at it but failed. Conversion drag was still too high to justify the revenue lift.

❌ THE FAIL: AI Contrast Checkers Miss Systemic Flaws

AI contrast checkers only spot-check one element. They miss systemic issues like inconsistent link color or drifted tokens. Instead, bake compliance into your Design System.

#DesignSystem #ColorContrast #WCAGCompliance #SystemDesign #AIFail

✅ SUCCESS #2: AI Mapped Enterprise Keyboard Nav.

I used Google Gemini to map out the full keyboard pattern for a complex enterprise app. It gave us a strong, standards-based head start, saving days of foundational planning.

#EnterpriseDesign #KeyboardNavigation #ProductDesign #AIinUX #Usability

✅ SUCCESS #1: AI Stopped Me From Confusing Screen Readers

My mistake? Using big headings

for visual style, not to build the site's structural outline. The AI acted as a coach, ensuring the "table of contents" made logical sense to screen readers.

#A11y #WebDev #WAIARIA #Frontend #SemanticHTML