A website looks AI-generated when its design is polished enough to seem finished but not specific enough to explain the product.

The strongest signs are not purple gradients, rounded cards, or a particular font. They are generic structure, interchangeable copy, decorative choices without a clear job, incomplete product states, and rules that change between screens.

None of these details can prove that AI made a website. Human designers use trends and templates too. An AI-assisted team can also produce careful, original work. What you can identify is an interface that lacks product-specific judgment.

Five simple tests make that difference easier to see.

1. The substitution test

Imagine replacing the company name, logo, and primary color with those of another business. How much of the page would still make sense?

If almost everything survives, the website is probably describing a category rather than a product.

A generic project-management page might promise to “streamline your workflow,” show three cards for collaboration, automation, and insights, then finish with a call to “work smarter.” The same page could sell a CRM, analytics tool, or AI assistant with only minor copy changes.

A specific page is harder to substitute. Its structure reflects what the product actually does. It may show the shape of a real project, explain an unusual workflow, compare a meaningful result, or address a concern unique to its audience.

Look for evidence that could not be moved to a different website unchanged:

  • real product inputs and outputs
  • language used by the intended customer
  • examples with believable constraints
  • proof tied to the promise being made
  • a page order that matches the buyer’s decision
  • calls to action that describe the next real step

Generic design is not only a visual problem. Interchangeable content often reveals it first.

2. The decoration-off test

Mentally remove the gradients, glows, illustrations, badges, shadows, and animation. Is the information still organized clearly?

Decoration can reinforce hierarchy, mood, and brand. It becomes a warning sign when it carries the whole design.

An AI-looking page often uses visual effects to create importance without deciding what is important. Several headings compete at the same size. Every feature gets an icon. Ordinary labels become pills. Sections are separated by background colors because their relationship is otherwise unclear.

With the decoration removed, the page becomes a sequence of similar blocks.

A stronger interface still communicates through:

  • the order of information
  • contrast between primary and supporting content
  • alignment and spacing
  • meaningful grouping
  • clear labels and actions
  • deliberate changes in density

This is why changing an “AI purple” gradient does not automatically make a website feel less generated. If the structure underneath remains generic, a new color only gives the same template a different coat of paint.

3. The second-screen test

The homepage may look convincing because it received most of the attention. Open a second route and compare how the same ideas are handled.

Check the navigation, headings, forms, buttons, cards, tables, icons, empty space, and page width. A coherent product does not need every screen to look identical, but repeated decisions should belong to one system.

AI-looking products often drift in small ways:

  • a primary action changes color or label
  • page titles use different scales
  • form fields gain a new height or radius
  • content widths shift without a reason
  • the same status uses different colors
  • spacing becomes denser or looser on each route
  • a component is rebuilt instead of reused

No single mismatch proves anything. The pattern matters. A page generated in isolation may solve its local prompt while ignoring the decisions made elsewhere.

The second-screen test also catches false consistency. Repeating the same card style everywhere can make a product look unified from a distance. If every workflow receives the same composition regardless of its content, the site is repeating a surface treatment rather than applying a system.

4. The uncomfortable-state test

Finished products must handle situations that do not fit the ideal screenshot.

Ask what the interface does when:

  • there is no data yet
  • a request is loading
  • a request fails
  • a form value is invalid
  • a label is unusually long
  • a table has hundreds of rows
  • a user lacks permission
  • a destructive action needs confirmation
  • the screen becomes narrow
  • content is translated

AI slop website design often stops at the happy path. The dashboard has perfect sample data. The chart always fits. Every account has an avatar. The form submits successfully. Mobile is treated as a smaller desktop.

These missing states matter because they expose how the website was conceived. A page designed as an image only needs one attractive arrangement. A product designed for use needs behavior, feedback, recovery, and boundaries.

The uncomfortable-state test is especially useful for polished interfaces. Surface quality can hide weak product thinking until real content or an error breaks the composition.

5. The product-language test

Read the page without looking at its design. Does the copy explain anything concrete?

AI-looking websites often rely on phrases that sound useful but avoid a testable claim:

  • “unlock your potential”
  • “transform your workflow”
  • “built for modern teams”
  • “insights that drive growth”
  • “everything you need in one place”
  • “work smarter, not harder”

Any one phrase can be appropriate. The problem appears when the whole page speaks this way.

Specific product language names the user, task, object, constraint, or result. It explains what changes after someone uses the product. It also admits important limits instead of presenting every feature as effortless and universal.

Compare these two descriptions:

Automate your workflow with intelligent tools built for modern teams.

Turn support emails into assigned tickets, suggest replies from your help center, and require an agent to approve each response before it is sent.

The second version gives the interface something to organize. It suggests objects, actions, roles, and states. The first version leaves the design to communicate a vague promise.

What do AI slop websites look like?

AI slop websites usually combine several weak signals instead of displaying one definitive giveaway.

They often look clean in a screenshot. The type is legible. The spacing is generous. The page includes the expected sections. Problems appear when you ask why each decision exists or try to use the interface beyond its ideal state.

Common symptoms include:

  • a page structure that could serve almost any company
  • marketing copy that makes broad claims without evidence
  • every idea placed inside the same kind of card
  • decoration that does more work than hierarchy
  • product previews filled with fake or meaningless data
  • actions that lead nowhere or do not explain what happens next
  • missing loading, empty, error, disabled, and success states
  • inconsistent components across routes
  • mobile layouts that only stack the desktop design
  • no sign that the output was reviewed against real content

AI assistance alone does not make a website slop. Slop describes the quality and review of the output, not the tool used to produce it.

A composite AI slop website example

Consider a fictional analytics product called Northstar.

Its homepage opens with “Turn data into decisions” over a blue-purple glow. Two buttons say “Get started” and “Learn more.” Below them are three cards named Insights, Automation, and Collaboration.

The product screenshot contains four metrics, an upward line chart, and a recent activity table. None of the numbers relate to a stated customer problem. The testimonials praise how “easy” and “powerful” the product is but do not name a result.

Inside the app, the dashboard repeats the same four cards. Filters do not affect the chart. The empty state is blank. A failed request shows no recovery action. The settings page uses different buttons and field spacing from onboarding.

No single detail makes Northstar an AI slop website. The diagnosis comes from the cluster:

  1. The promise is interchangeable.
  2. The visual effects have no product meaning.
  3. The preview imitates an analytics dashboard without explaining a task.
  4. The product only handles the ideal state.
  5. Repeated decisions do not form a stable system.

This example is fictional, but the method is practical. Evaluate whether the structure, content, behavior, and visual rules support one real product.

What does not prove that a website used AI?

Popular design choices are weak evidence on their own.

You cannot identify AI authorship from:

  • a purple or blue gradient
  • rounded cards
  • a dark landing page
  • a large centered headline
  • a common sans-serif font
  • stock illustrations
  • a standard dashboard layout
  • code built with a popular component library

Each choice can be deliberate and appropriate. The question is whether it works with the other decisions and serves the product.

This distinction prevents the diagnosis from becoming a style preference. Minimal, maximal, colorful, neutral, dense, and spacious interfaces can all be specific. They can also all be generic.

How to know if something is AI slop

Use three criteria:

  1. It is generic. The structure, copy, or visual language does not respond to the actual product.
  2. It is unresolved. Important states, interactions, content, or constraints are missing.
  3. It is unreviewed. Obvious inconsistencies and irrelevant details remain in the published result.

One weak choice is a design issue. A repeated pattern across all three criteria is a stronger sign of slop.

If you only have a screenshot, you can assess generic structure and surface choices. You cannot make a confident judgment about behavior, responsive design, or state coverage. Interact with several routes before reaching a conclusion.

For a complete review process, follow How to audit and fix AI-generated UI. If you want to understand why these patterns keep appearing, read Why AI-generated websites all look the same.

The goal is not to catch people using AI. It is to recognize work that still needs product judgment. A good website earns its design through specific content, coherent rules, complete behavior, and visible review.