Think AI Detectors Are Reliable? Read This First

bydocligot Published:

Are you still using AI detectors?

I ask because I just tested them myself. I ran my last post through several AI detectors. The results were mixed. One tool said it was likely written by a human. Another said it was probably AI. A third couldn’t decide at all.

  • Scribbr - 0%
  • Quillbot - 0%
  • Content Detector AI - 45%
  • Quetext - 68.79%
  • Copyleaks - 100%
  • Undetectable - 84%
  • Brandwell - ?
  • ZeroGPT - 19%
  • GPTZero - 100%

(See the links below)

Same writing. Same words. Different answers.

That alone should make us pause.

AI detectors don’t actually know who wrote anything. They don’t watch you think. They don’t see you plan, draft, delete sentences, or rewrite ideas. They don’t see the struggle of getting words right. All they see is the final text. From there, they make a guess.

That’s not opinion. That’s how the tools work.

Most AI detectors look for patterns. They measure how predictable the writing is. AI often chooses common, expected words, so very smooth writing can look “machine-like.” Detectors also look for how even the sentences are. Human writing tends to jump around more. AI writing can look steady and uniform.

But here’s the problem: humans don’t all write the same way.

Some people write clearly. Some write simply. Some write carefully. Some write in short sentences. Some are non-native speakers who keep things basic on purpose. All of these very human styles can trigger an AI detector.

That’s where false accusations come from.

A detector might say a real person used AI when they didn’t. In schools, that can mean failing an assignment. In workplaces, it can damage trust. In both cases, the cost of being wrong is high.

And the mistakes don’t go just one way.

AI writing can often pass as human with very little effort. A bit of editing. A few sentence changes. Maybe a rewrite in a different tone. Suddenly the detector says, “Looks human.” So the tool misses actual AI use while accusing real people.

That’s the worst kind of system: one that fails on both sides.

People often assume AI detectors are like plagiarism checkers. They’re not. Plagiarism tools compare text to known sources. AI detectors don’t have anything like that. There is no database of “who wrote this.” There is no authorship log. There is no proof to check.

All they can say is, “This looks similar to writing we’ve seen before.”

That’s not evidence. That’s probability.

And probability is not strong enough to punish someone.

To be fair, AI detectors aren’t useless. They can sometimes flag very obvious, unedited AI text, especially if it’s long. Used carefully, they might raise a question. But they should never answer it.

Responsible use means treating detectors as one small signal, not a verdict. Draft history, writing process, conversations, and context matter far more than a score on a dashboard.

After running my own human writing through multiple detectors and getting mixed results, the conclusion is hard to ignore.

If the same text can be labeled human, AI, and “uncertain” all at once, what exactly are we trusting?

So I’ll ask again:

Do you still trust AI detectors?

 

List of Detectors:



About Me:

Dominic “Doc” Ligot is one of the leading voices in AI in the Philippines. Doc has been extensively cited in local and global media outlets including The Economist, South China Morning Post, Washington Post, and Agence France Presse. His award-winning work has been recognized and published by prestigious organizations such as NASA, Data.org, Digital Public Goods Alliance, the Group on Earth Observations (GEO), the United Nations Development Programme (UNDP), the World Health Organization (WHO), and UNICEF.

If you need guidance or training in maximizing AI for your career or business, reach out to Doc via https://docligot.com.

Follow Doc Ligot on Facebook: https://facebook.com/docligotAI

 

 

Tags:

You might like

View all
Previous Post Next Post
Share to other apps
Copy Post Link