Prompt to
Product

A designer's field test — building the same news app across 4 AI builders to find out who actually delivers on the brief

ROLE

UX Researcher

BUILDERS

Lovable, v0, Bolt, Emergent

YEAR

2026

01 - OVERVIEW

A designer's field test

This project started as a question with no clean answer — with so many AI app builders launching at once, which one is actually worth using? Rather than reading about them, I tested them by building the same app in four different AI builders using the same starting prompt and evaluated each on output quality, design decision, iteration responsiveness, and free tier usability.

"What started as a tool comparison became something more useful: a lesson in why the design process still matters even when AI is doing the building."

02 - DISCOVERY

Jumping in blind

The moment I wrote the first iteration prompt for Lovable, something clicked. I was able to fit more features into that single prompt than I had used across multiple days of building in Replit — because Replit had already shown me exactly what to ask for.

Building blind in Replit, with just an idea and no direction, led to the discovery that a PRD was necessary. Without one, there was no way to build with intention or evaluate with consistency. That document became the foundation and direction for the rest of the project.

What Replit revealed

Coming up with ideas for features on the spot, adjusting and iterating on small critiques and wasting tokens, chatting instead of being specific, having no path or design journey to work towards, letting AI decide and drive the project

What was learned

The discovery of PRD and writing one to set the guidelines to where each AI builder project should be working towards and will be evaluated on the journey and quality of the iterations to get to the final product. Keeping the evaluation and projects consistent and uniform and allow judgement to be based solely on AI builder.

initial_prompt.txt
Abridged Verison of Product Requirements
Project: AI App Builder Evaluation — Mobile News App Contributors: Charlyn Truong Status: In Progress What & Why Testing four AI app builders — Lovable, v0, Bolt, Emergent — by building the same news app with the same prompt across all four. Goal is to observe, not rank. Replit used as discovery and control standard. The Problem Too many AI app builders, not enough clarity on which actually performs. This project creates a firsthand, structured comparison so designers and early-stage builders can make an informed choice. Objectives - Observe how each builder interprets the same prompt - Track how well each handles iteration and small corrections - Document what's built out of the box vs what needs prompting - Compare all outputs against Replit as a neutral baseline What Success Looks Like Must have - Interest selection screen functional on first output - Feed renders with article cards and category tags - Cards are tappable and attempt to open article - All iteration prompts reflected in final output - Builder generates an app name and basic design system - Builder can publish or share a live preview - Builder allows rollback to previous version Nice to have - Working article links - Consistent color system across screens - Builder suggests design alternatives when prompted The App — 8 Core Features 01 Interest Selection - Onboarding screen to pick topics, personalizes feed 02 Article Cards - Headline + 2–3 sentence summary, tappable 03 Summary Modal - In-app overlay with key points before full article 04 Engagement Actions - Upvote, save, share, comment per article 05 Navigation - Bottom nav — Home, Trending, Profile 06 Read Indicator - Visual marker on already-viewed articles 07 Commenting - View and post comments inside the modal 08 User Profile - Saved articles and preferences, no login required Not Building Real news API · Auth / login · Backend · Production deployment · Cross-builder feature parity Evaluation Dimensions Output Quality · Design Quality · Ease of Use · Iteration Handling · Free Tier Limits Builders Lovable · v0 · Bolt · Emergent · Replit (control only)
Mobile-firstNews appCore FeaturesSame prompt, 4 builders

03 - THE BUILDERS

Meet the cohort

Lovable logo
Lovable
★★★★☆
Creative, consistent, and likes to add a bit of sunshine in everything they do, helps them stand out.
View case study
v0 by Vercel logo
v0 by Vercel
★★★☆☆
A little undiagnosed OCD in organization but gets a bit lost in the chaos.
View case study
Bolt logo
Bolt
★★☆☆☆
Says they got it but gets overwhelmed easily, needs some hand holding to get the job done, correctly.
View case study
Emergent logo
Emergent
★★☆☆☆
Wants to be a one and done, good at asking questions, but runs out of energy quick unless they really like you.
View case study

04 - ITERATION WALKTHROUGHS

See what each builder made

Pages
Home
Comparing Lovable, v0, Bolt & Emergent
Lovable — Home
Lovable
v0 — Home
v0
Bolt — Home
Bolt
Emergent — Home
Emergent

05 - SCORECARD

What they were up against

Dimension
Rating
Assessment
01
Output Quality
Aesthetic UI, organization, information architeture, creatvitity
02
Ease of Use
Able to navigate page with minimal to zero, beginner friendly, limited confusion
03
Iteration Handling
Prompt to changes immediately, takes everything into account, iteration makes sense and made obvious
04
Free Tier Limits
Tokens / prompts in a day, able to obtain all features and publish final product
✓ PassMet the standard
◑ PartialMet with caveats
✗ FailDid not meet standard
— N/AInsufficient output
HeuristicLovablev0BoltEmergent
01
Visual consistency
Does the design system hold up across screens — colors, type, spacing, and components?
✓ Pass
Strong design language from the first output. Colors, type, and components stayed consistent across every iteration.
◑ Partial
Clean and minimal but leaned web-first. Consistency held on desktop view — less so when viewed as mobile.
◑ Partial
Visual output stayed consistent through all iterations, there was not any surprises or major changes that affected the visual aspect of the app. It stayed consistently minimal.
✓ Pass
The design is consistent throughout the app. It kept visual consistency for the first prototype and is likely to continue with iterations.
02
Navigation & wayfinding
Can a user always tell where they are, how to go back, and where to go next?
✓ Pass
Bottom nav, clear back arrows, and modal exits all implemented correctly. User always had a clear next step.
◑ Partial
Navigation required multiple iterations to correct. Modal behavior and article exits needed specific prompting.
✓ Pass
Pathways are easy to identify and the app is easy to navigate. Made sure to add exits or return buttons and none had a roadblock.
◑ Partial
The navigation and arrows are clear and the journey is easy to follow. However, the login part was a roadblock and required to create an account to proceed.
03
PRD adherence
How closely did the final output match the specified features, flows, and interactions?
✓ Pass
Implemented iterations exactly as requested. Kept features and UI consistent — no regression between prompts.
◑ Partial
Good on additive changes but needed precise language for details, especially the article modal and scroll behavior.
◑ Partial
Overall, Bolt did have all the specified features, though it does struggle a bit with the flow on the comments modal.
— N/A
Due to the limited tokens, no iterations were made towards the goal brief. Emergent does not have enough to be rated in this area. However, this is a decent starting point and had more ready than the other V1s.
04
Mobile experience
Does it feel like a native mobile app? Tap targets, layout, scrolling, and transitions.
✓ Pass
Felt native from the first output. Tap targets were appropriately sized and transitions felt intentional.
◑ Partial
Strongest feature was the ability to switch between web and mobile view. Occasionally felt adapted rather than native.
◑ Partial
Mobile frame boundaries were a persistent issue — content occasionally bled outside the frame. Required explicit correction.
— N/A
Not enough was done for this prototype to be evaluated in this area. The current experience is very short and limited.

06 - RATING & THOUGHTFUL ADDITIONS

How they stacked up - overview edition

ReplitLovablev0BoltEmergent
Overall rating
My rating
★★★★☆
3.75 / 5
★★★★★
4.5 / 5
★★★★☆
3.5 / 5
★★★☆☆
2.5 / 5
★★★☆☆
2.5 / 5
Thoughtful additions
Web to mobile view
Working links
Expo Go preview
Live shareable mockupnot published
Publish to web / apppaid only
Notes
Extra notesBecame the standard of what to expectHas the most personalityBest org of previous versions, but can get confusingReports "Missing Project Files"Like the tabs for the project

07 - TAKEAWAYS

Lessons to take onto the next project

This project taught me more about the importance of the design process than it did about AI tools.

01

Going into a builder without a design direction produced something — but reacting to what came out wasn't the same as designing with intention.

02

All AI builders are excellent at executing a clear brief and much weaker at filling in what wasn't specified.

03

The builders that produced the strongest outputs weren't necessarily the most technically capable — they made the best design decisions in the absence of explicit instructions.

04

That's a design judgment call, not an engineering one — and it's where the gap between a good builder and a great one actually lives.

Return to Projects
trainON
visualize muscle