Start with your own site
A studio that sells craft over AI slop has to pass its own test first. Before the redesign, we ran the AI Slop Test on the old produlogi.com homepage, row by row, with the same rule we give clients: one yes in any row means the page fails.
The old homepage failed three of the five rows outright and passed one only because nothing was there to test.
Row one: visual
The test asks whether the page leans on default gradients, generic 3D, stock "AI" imagery or uniform card grids.
The old hero sat on an animated WebGL gradient loaded from a third-party script. Our SEO audit flagged it for performance: it loaded late, and it ran a loop that scanned the page while the main content was trying to render. It also failed on craft. A drifting gradient behind a headline is the most common hero on the web, and it said nothing about the work.
The portfolio section showed four stock-style images with alt text such as "AI product interface" and "Design system". None of them showed a real artefact.
Verdict: fail.
Row two: content
This row asks about placeholder data, invented metrics and anonymous praise.
The homepage had a testimonials block with a "15+" products figure and the phrase "full client satisfaction", with no named client behind either. A feature card showed a small chart labelled "2-4x Conversion" with no source and no project attached.
Neither claim came with evidence, and they sat on the page that asks a new visitor for trust.
Verdict: fail.
Row three: AI UX
This row checks whether AI features show more than the happy path.
The old homepage had no AI feature, so the row had nothing to fail. We record that as not tested rather than passed. The row applies the moment the site gets an AI feature, however small.
Row four: copy
This row looks for stock phrasing and claims without a verb and an object.
The H1 read "Design. Build. Ship." It named no category, no audience and no outcome. A visitor could not tell from the first screen that the studio works on AI products.
A second problem was worse than vague. The site called its training "HRDC-claimable" without naming the provider. The HRD Corp registered provider for the courses Marcus teaches is AiTraining2U, and the wording needs to say so every time.
Verdict: fail.
Row five: process
This row asks whether anyone outside the maker checked the page before it went live. We could not find a record either way for the old homepage, so we left the row open and made it a release rule for the new one.
What we changed
The redesign addresses each failed row.
- The third-party WebGL background goes. The new homepage is our own scene, built for this site and loaded only after the headline has painted, and each object in it stands for a part of our practice.
- The anonymous testimonials and the unsourced chart come down. Named quotes return only with written consent from the person quoted.
- The H1 now names the category and the outcome: design and ship AI products people trust.
- Every HRDC mention now names AiTraining2U as the registered provider. Our training page links to AiTraining2U for dates and claims.
- No page ships until it passes all five rows with zero fails, checked by someone who did not build it.
What the exercise taught us
Every failure on the list had been seen and approved at some point, one small decision at a time. The gradient looked fine in the moment. The "15+" figure felt harmless. The vague H1 read as confident.
Slop builds up out of defaults that nobody stops to question, which is why the test works as a checklist with a hard fail rule. A reviewer who has to write yes or no next to "anonymous praise" cannot talk themselves past it.
Two other Lab Notes cover the fixes in more depth. Write the design rules before the model guesses deals with the visual row. Design the states the AI demo skips deals with the AI UX row.
Run the test on your own homepage this week, then on the last AI feature your team shipped. If you want a second pair of eyes on either, we run the test as part of every consulting engagement. If you want to learn to run it on your own work, see coaching.