Automated design audits for AI-built sites
Your AI built it.
Now make it look designed.
Paste your URL, get your shipshape score in about a minute. The full audit adds a FIX.md your coding agent can apply.
Free, no signup. No public URL yet? Join the waitlist instead.
Where slop hides
Slop is rarely one big mistake. It is a dozen small defaults you scroll past, choices no one made, shipped anyway. Below are six examples, caught on deliberately vibe-coded demo pages and marked exactly the way the audit marks them. They stand for many more: the catalog behind the scan checks for 71 of these tells across six categories.
Color & gradients
The same gradient on every surface
Hero, buttons and stat numbers all wear one stock indigo-to-violet ramp. A color nobody picked reads as template, not product.
Typography
A headline painted, not ranked
The gradient clip is decoration standing where hierarchy should work. Scale and weight rank a page; paint does not.
Authenticity
Numbers with nothing behind them
Happy families, uptime nines and perfect ratings for a product nobody has opened. One unverifiable number poisons every true sentence around it.
Components & effects
Glass panels and violet glow
Backdrop blur plus purple shadows is the default costume of generated dashboards. Depth comes from contrast, not fog.
Components & effects
Emoji standing in for icons
A rocket, a sparkle, a padlock: placeholders shipped as UI. They render differently on every platform and mean nothing specific.
Copy voice
The phrasing every AI tool ships
The guilt-relief opener, the 'not just X' pivot, the adjective triple. Copy that could sell anything describes nothing.
And none of it is a redesign.
The fix removes the defaults and stops. This kids-app demo keeps its butter palette, its coral, its headline and its layout; what goes is the stock ramp, the badge row and the invented numbers. The look changes exactly as much as the defaults dominated it, and no further.
We score the defaults, not your taste.
AI tools make hundreds of design decisions you never saw: the stock gradient, the glass cards, the badge spam, the spacing that drifts from section to section. shipshape finds the decisions that were made for you, so every one that stays is yours. Remove machine defaults. Keep your style.
How it works
Scan.
We inspect the rendered interface: your page in a real browser, desktop and mobile, exactly as visitors see it.
Score.
You get a shipshape score from 0 to 100 across six categories, and you see where machine defaults are showing through. Deterministic checks: same URL, same score.
Shape.
The full audit ships a FIX.md written against your page. Drop it into Claude Code, Cursor, Lovable or Bolt. Your own tool applies it, in your codebase, in your style.
Verify.
Run the included verification scan and see what improved. Shipped is where you started; shaped is the state you can prove.
The method
Design rules, not AI vibes. The catalog behind the score holds 71 explicit checks written and calibrated by a designer; 63 of them are deterministic rules a machine can verify. No model looks at a screenshot and gives its opinion.
Color & gradients
stock ramps, palette sprawl, contrast floors
Typography
flat hierarchy, stock stacks, broken outlines
Layout & spacing
drifting gaps, equal-card grids, overflow at 390px
Components & effects
glass panels, glow shadows, badge spam
Copy voice
AI phrasing, filler adjectives, undecided microcopy
Authenticity
sourceless stats, placeholder people, dead links
Every finding tells you how sure we are:
- Detected
Technically measured: nine border-radius values, a 3.1:1 contrast ratio, seven spacing values in one component. No judgment call involved.
- Pattern match
A known machine default: the generic AI SaaS hero, the badge wall, the repeating copy pattern. Named as a pattern, argued with evidence from your page.
- Needs evidence
A claim we can see but not verify: the 10,000 happy customers, the uptime nines. We never call a stat fabricated; we say it needs a source.
FIX.md
## Shape 03: Hero hierarchy
Priority: High
Issue:
The headline, badge and supporting copy compete
for primary attention.
Change:
- Remove the announcement badge
- Increase H1 size from 56px to 68px
- Reduce supporting copy width to 560px
- Keep one primary CTA
Preserve:
- The headline wording and the brand color
Verify:
- One element leads the viewport at 1280px and 390pxEvery directive says what to change, what to preserve while changing it, and how to verify it worked. Drop the file into Claude Code, Cursor or your coding agent and let it run.
The NEVER rules matter most: they stop your tool from putting back what it just removed.
shipshape on shipshape
We run this page through its own audit before every release. Here is the current result, deductions included. A perfect score would mean we hid something.
AI-default presence: Low
- LayoutLow
One vertical rhythm everywhere
Every section uses the same vertical padding. Our own catalog flags uniform spacing as an unexamined default. Here it is a chosen one: the page reads as one continuous report. We take the deduction.
- Copy VoiceLow
The core promise repeats
Defaults versus style appears in close variants in three sections. Deliberate emphasis, but repetition is repetition, and the check does not care about our intentions.
- ComponentsLow
Mixed focus styles
Form controls carry custom focus rings; the FAQ rows use the browser default. Consistent would be better. This one we will fix.
Pricing
Free tells you where the problems are. The full audit tells your agent exactly how to fix them.
Free
$0
shipshape score
- Overall score
- Category scores
- Every finding weighted, three in detail
- Shareable score card
Full Audit
$29 one-time
The findings, plus the file that fixes them
- FIX.md, ready for your coding agent
- Every finding, prioritized
- Before and after guidance
- One verification scan
No subscription. Launching soon, and the waitlist gets it first.
Who's behind this
I'm a designer with more than 16 years of experience across digital and physical industrial design. I've spent my career designing products, interfaces and physical objects, and more recently building and shipping my own software products with AI-assisted development tools.
shipshape grew out of a problem I kept running into in that process: AI can get a product surprisingly far, surprisingly fast. But many of the design decisions it makes along the way are defaults rather than deliberate choices. I turned the design principles I've used throughout my career, together with the best open design-engineering knowledge, into one explicit, calibrated rule system. Curating what belongs in it and proving it against real sites is the work; the sources are named on the method page.
The judgment is designer-authored. The audit is automated.
- 16+ years
- of design experience
- Digital + physical
- product and industrial design
- Now shipping software
- building my own AI-assisted products
- Designer-authored
- rules, not vibes
What shipshape isn't
- Not a redesign service.
- We show you what to improve. Your tool implements it.
- Not a usability study.
- shipshape evaluates observable interface decisions, not real-user behavior.
- Not an accessibility certification.
- Accessibility-related design checks are included, but this does not replace a full accessibility audit.
- Not a style generator.
- shipshape doesn't choose your aesthetic. It helps make your existing direction intentional.
FAQ
What exactly do you mean by “slop”?
The defaults nobody chose: the stock purple gradient, the glass cards, the invented stat row, the filler adjectives every AI tool ships when you don't push back. Slop is not a verdict on your taste. It is the absence of decisions, and it makes real products look like templates.
How long does a scan take, and what does the score mean?
About a minute. The shipshape score starts at 100 and each finding deducts points by severity and visibility. Alongside it you get an AI-default presence level: how much of the page is running on decisions a machine made for you.
Do you store my code?
No. There is nothing to hand over: you give us a URL, we render what's publicly reachable, like any visitor's browser does. No account, no access, no code.
Is this a human review?
The scan is automated so it stays fast, cheap and repeatable. What it checks is not: every rule in the catalog was written and calibrated by a designer, and the paid audit only keeps findings that survive strict evidence rules. No vibes, either way.
What if my page can't be scanned?
You only pay after a successful scan. If a paid run fails anyway, you get an automatic refund.
Which tools does FIX.md work with?
Any agent that reads markdown: Cursor, Claude Code, Codex, Lovable, Bolt, v0.
Will this make my site look minimal?
No. Fixes remove machine defaults, not style. If you chose a look on purpose, it stays. Every fix is anchored to your existing palette, type and voice; only the concrete slop tells always go: fabricated stats, gradient spam, badge noise. A playful site stays playful.
Is a score of 100 the goal?
No. 100 means the page follows every rule in the catalog, not that it's a perfect site. The score counts concrete, checkable defaults, so a page with zero of them scores 100. A memorable site sometimes breaks a rule on purpose, takes the point hit, and is better for it. Well shaped starts at 90; our own page sits below 100 and we publish why.
See what your AI decided for you.
The full audit is coming
All findings, evidence crops, and the FIX.md your coding agent applies, $29 once. The waitlist gets it first.