Spot AI slop
AI Slop Detector: What It Can and Can't See on Your Site
What an AI slop detector can see in text, images, and pages, why its score can be wrong both ways, and how to read one before you fix anything.
An AI slop detector is a tool that scans text, an image, or a whole page for patterns that look machine-made or generic, then gives you a score. It spots tells fast. It cannot tell you whether your product is clear, true, and trusted, and its score can be wrong in both directions.
You built something with AI, and you want a quick, honest signal. A detector can give you one, as long as you know what it looked at. Here are the kinds, where each goes wrong, and how to read a score. For the definition of the word itself, see what AI slop means.
What does an AI slop detector look at?
There are 3 kinds, and the difference is the input.
- Text detectors read words and estimate how likely it is that a machine wrote them.
- Image detectors look at one picture and estimate whether a machine made it.
- Page scanners load a page and judge the look and the wording together: the headline, the layout, the icons, the photos, the footer.
Each one judges a single slice. A text detector never sees your layout. An image detector never sees your headline. A page scanner sees the page, but reading a page is not the same as using the product behind it. Tools also differ in how they reach a number, so read what a tool says its score means before you act on it.
What can a page scan see quickly?
Page scans are good at tells that sit on the surface and look the same every time. They are fast and consistent.
Think of anything you could spot by comparing two pages side by side: the same photo style repeated, one icon pattern used for every feature, corners rounded differently from card to card, a footer that was never updated. A scan can also check that the basics exist, such as a way to contact you and the usual legal pages. Whether those pages exist is something a scan can see. Whether they are right for your business is not, and none of this is legal advice.
What does a page scan miss?
3 things are hard for any pattern check, whatever it is built on.
- Who made it and why. A scan sees a page, not the person behind it. A founder who chose a plain layout on purpose and a founder who never touched the defaults can end up with pages that look alike.
- What your audience expects. A dense, plain page can be exactly right for a niche, and a polished one can be wrong for it. Patterns do not know your buyers.
- Change over time. A scan is one snapshot. It does not know that last week's version was better or that you fixed the headline an hour ago.
Why can a score be wrong in both directions?
2 terms help. A false positive is when a check flags something that is actually fine. A false negative is when a check passes something that is actually poor.
Text detectors show both clearly. A 2023 research paper (arXiv, first submitted 6 April 2023, last revised 10 July 2023, read 5 October 2026) tested several widely used GPT detectors on writing from native and non-native English speakers. In that paper, the detectors consistently labeled the non-native samples as AI-generated, while the native samples were identified accurately. The authors also report that simple prompting strategies could slip past the detectors, and they caution against using them in evaluative or educational settings.
Pages behave the same way. A scan can flag a clean template that a founder customized with care (a false positive). It can also pass a polished page with a muddled offer (a false negative), because polish leaves marks a tool can count, and clarity has to be understood.
W3C, the web standards body, makes a related point about accessibility tools in its evaluation tools overview (page updated 28 April 2020, read 5 October 2026). It says some checks cannot be automated, that tools can give inaccurate results, and that you should not rely on what tools say over the real experience of the people using a site. The page is about accessibility, but the habit travels. Treat a score as a clue.
What is an AI generator fingerprint check?
Our homepage lists an "AI generator fingerprint check" among the things the $27 report includes, and its FAQ says the audit "checks for the fingerprints AI builders leave behind." You can see how a report words that finding in the sample report. Here is what we do and do not claim for it.
- What it does. It is one check inside the audit of your live public pages, looking for the marks AI builders leave behind. It is a read of how the page looks.
- What it does not claim. It is not proof of origin. The audit reads what a customer sees at your link, "not your code or which tool you used," so it cannot tell you which tool made a site, or that AI made it at all.
- What it is not aimed at. The audit works on any public site, however it was made, including ones built with Lovable, Bolt, or Replit, and ones started from a template and edited with AI. We are not affiliated with any of them.
A template edited by hand and a template edited with AI can look alike, so treat a fingerprint note as a hint, not proof. A hint is still useful. It points at the part of the page where a visitor may feel sameness, and you can decide whether that is true.
What does our audit read that a text or image detector can't?
No person reads your app here. Is My App Slop uses AI to score what a customer sees against a fixed rubric, which is why a report is back in minutes. Instead of one AI-or-not number, it scores six areas of what a customer sees, and the report lists what it could not confirm under Not confirmed.
What does a pattern check flag on a language-tutor marketplace?
Hypothetical example: A founder builds a marketplace where learners book 1-on-1 lessons with tutors, then runs a pattern scan.
| What a pattern check flags | What a visitor notices |
|---|---|
| A generic photo of a smiling person with a laptop in the hero | Tutor cards show no teaching language or price until you open each profile |
| The same icon repeated on all 3 feature cards | The Book a lesson button sits below the fold on a phone |
| A headline that could fit any app: "Unlock your language potential" | Nothing says whether tutors are certified, or what a trial lesson costs |
| Card corners rounded differently from one card to the next | Time slots appear in the tutor's time zone with no label, so a learner books the wrong hour |
The left column is worth fixing, and a scan can find it. The right column is where a learner decides to book or leave. Fix only the left column and you may get a better score and a page that still loses people.
How should you read a score?
As a summary of what the tool could observe, and no more. Ask 3 questions of any score:
- What did it look at? Words, a picture, a page, or a path through the product.
- What could it not see? A tool that does not say has not told you its limits.
- What do I do next? A single number with no reasons gives you nothing to fix.
Look at the parts, not the total. Our report gives each of the six areas its own score, a What works note, a Slop tells note, and ranked fixes, and it gives an overall Verdict. Work from the ranked fixes.
A good score means fewer obvious problems. A weak one means there is work to do, not that the idea is bad. A score is a clue about the page, so treat the tells it finds as hints, not proof. Our guide on reading an audit report shows how to turn findings into a working list.
When is a detector enough?
Early on. With a draft and a few first users coming, a detector or a scan is a fair start. If you would rather look at your own site first, our 20-minute self-check walks you through it by hand.
Go further when more is at stake. If real money moves, people create accounts, or you store personal details, add a specialist who can review those parts directly. A score tells you how the front door looks. It cannot vouch for the locks.