AI slop, and what a website has to do instead
Slop is what you get when something is generated rather than decided. On a website it is fluent, competent, entirely forgettable, and indistinguishable from every competitor who used the same tool. The problem is not that AI wrote it. The problem is that nobody chose anything.
The definition
Not badly made. Unmade.
Slop is the word that stuck for AI output produced at volume with nobody deciding anything: text that reads well and says nothing, images that look like images of things rather than things. It is worth being precise about the cause, because the cause is not the technology.
- A model answers with the centre of everything it has read. Asked for a website for a business like yours, it returns the most average version of that.
- That average is fluent. It has correct grammar, confident claims and a sensible structure, which is exactly why it slips through.
- Nothing in it was chosen. No one decided this claim over that one, this order over another, this word because a customer used it on the phone.
- Slop is therefore a decision problem wearing a technology costume. The same model, pointed at real decisions, produces work that is not slop at all.
- Which is why we use AI on every build and still call this out. We are not arguing against the tool. We are arguing against publishing its first answer.
The tells
Eight things you can check on any page in two minutes.
Read your own homepage against this list. Most sites built without a strategy step fail five or more, and the failures are the same ones every time.
- 01
The headline would fit any competitor
The test that settles it: could a rival put their logo on your homepage and publish it unchanged? If yes, the page is not making a case, it is occupying a URL.
- 02
Claims with nothing behind them
"Innovative solutions", "trusted by industry leaders", "passionate about quality". Adjectives doing the work that a number, a name or a date should be doing.
- 03
Three features, in three columns, in a row
Almost always three, almost always icons, almost always the same three abstractions. It is the shape a model reaches for when nobody told it what to prioritise.
- 04
Nothing specific enough to be wrong
No prices, no timelines, no named clients, no numbers, no constraints, nothing a competitor could contest. Specifics are what make a page credible, and generated copy avoids them because being specific risks being wrong.
- 05
Words your customers never use
"Leverage", "solutions", "seamless", "bespoke". Nobody rings up asking for a seamless solution. They ask whether you can fix the thing by Friday.
- 06
Several equally weighted calls to action
Learn more, get started, contact us, book a demo, download the guide. A page that wants five things gets none of them, because nobody decided which one mattered.
- 07
Imagery that could belong to anyone
Generic stock or generated pictures of nobody in particular, in an office that does not exist. One real photograph of the actual team beats a gallery of it.
- 08
No evidence of a real customer
No testimonial with a full name and a company, no case with a before and an after, no logo anyone recognises. Proof is the single hardest thing to fake and the first thing missing.
The same business, twice
One of these was generated. One was decided.
Same trade, same city, same job to do. The page on the left is what you get when nobody chose anything — it is fluent, competent and interchangeable. The page on the right had a strategy session first.
- 1A headline any competitor could publish unchanged
- 2Adjectives where a number, a name or a date should be
- 3Two calls to action of equal weight, so neither is the one
- 4Three features, three columns, three abstractions
- 5Logos nobody can read and no customer anyone can look up
Neither business is real. The point is not that the left is ugly — it is not, and that is the trap. It is that a competitor could publish it unchanged, and nothing on it gives a stranger a reason to ring you.
The cost
It is worse than a plain site, not better.
The instinct is that a polished generic site beats a modest specific one. For a business that needs strangers to choose it, the opposite is usually true.
- Buyers have learned to recognise it. In 2026 a page that reads like it was generated signals that the business cut a corner on the one asset a stranger uses to judge them.
- It gives a buyer no reason to pick you. If your page and four competitors' pages say the same thing, the decision falls to price, and you did not want that fight.
- It cannot be improved, only replaced, because nobody can say what it was trying to achieve.
- Search rewards usefulness, not volume. Google's stated position is that it does not care whether a person or a model wrote a page, but its spam policies target content produced at scale to rank rather than to help. Generic pages do not fail because they are AI-written; they fail because they are not the best answer to anything.
- AI search makes it worse. An assistant answering a question needs something specific to quote. A page of fluent generalities gives it nothing, so it cites the competitor who published a real number.
The standard
What a high functioning website is held to.
The opposite of slop is not craft for its own sake. It is a page where every element is there because someone decided it should be, and the site is measured on whether that was right.
- 01
One job per page, written down before the build
A page exists to produce one action from one kind of visitor. If it cannot be stated in a sentence, the page is not ready to be designed, let alone generated.
- 02
A claim only this business can make
Specific enough that a competitor copying it would be lying. That sentence is the output of the strategy session and the input to everything else.
- 03
Evidence next to every claim
Names, numbers, dates, photographs of the actual work, a testimonial from someone who can be looked up. Proof is what converts, and it is the part no model can invent for you.
- 04
The invisible standards treated as requirements
Fast on a phone on mobile data, accessible to WCAG 2.2 AA, indexable, and a form proven to deliver. Not extras to add later: the things that decide whether the rest is ever seen.
- 05
Measured from the first visitor
Analytics and behaviour data installed before launch, so the weeks when the site is easiest to read and cheapest to fix are not spent guessing.
- 06
Owned, so it can keep changing
The code in your repository, the domain and hosting in your name, in a framework any developer can pick up. A website that cannot be changed is a brochure with a monthly bill.
FAQ
Questions people ask about this
What is AI slop?
AI slop is content generated at volume with nobody deciding anything: text that is fluent and says nothing, images of no one in particular, pages that could belong to any business in the industry. The term spread through 2024 and 2025 as the web filled with it. The defining feature is not that a model produced it but that no human made a choice about it, which is why the same models can produce work that is not slop when pointed at real decisions.
Does Google penalise AI-generated websites?
Not for being AI-generated. Google's stated position is that it rewards helpful, reliable content regardless of how it was produced, and its spam policies target content created at scale primarily to manipulate rankings rather than to help anyone. In practice generic AI pages do badly for a simpler reason: they are not the best answer to any particular question, so nothing about them earns a ranking or a citation.
How can I tell if a website is AI slop?
Apply one test. Could a direct competitor put their logo on the homepage and publish it unchanged? If yes, it is slop, whoever wrote it. Then check for the usual companions: claims with no numbers or names behind them, three features in three columns, several equally weighted calls to action, stock imagery of nobody, and no testimonial from a person you could look up.
Is AI slop actually bad for business?
Yes, in two specific ways beyond taste. It gives a buyer no reason to choose you over a competitor whose page says the same thing, which pushes the decision to price. And it gives AI assistants nothing quotable, so when someone asks an assistant to recommend a supplier it cites the business that published a real number, a real constraint or a real case.
What is the difference between AI slop and a website built with AI?
The decisions. We use AI on every website we build, and the output is not slop, because a strategy session decided what the business can claim that competitors cannot, a person read and rejected parts of what came back, and the result was measured against real checks for speed, accessibility, search and form delivery. Slop is what you get when the path runs from prompt to live with none of those steps in it.
How do I avoid AI slop on my own website?
Decide before you generate. Write down who the page is for, the single action you want, and one claim your competitors could not honestly make. Then insist on specifics: a number, a name, a date, a real photograph, a testimonial from someone findable. Finally, read the result and ask whether a competitor could publish it unchanged. If they could, you have not finished.
Start your brief
Start your brief
Tell us what you're launching and what it needs to do. We reply within one business day with a fixed price and the first available Monday.
- Websites from $15,000 ex GST, fixed. The exact figure comes with the reply.
- A reply within one business day, from Michael, not a form robot.
- A thirty-minute call to hear what the site has to do.
- A fixed-price scope, sent as a web page with a version number.
- Kick-off on the next available Monday, held for you. We take four studio builds a month.