Breaking News

How to Detect AI-Generated Content (and Can Google Do That)

Artificial Intelligence is taking the SEO industry by storm, and as it often happens in our community, there have been lots of discussions about how to use AI to manipulate ranking signals scale, especially content.

AI tools can write original content, and, from the reader’s point of view, good content. In fact, some AI-written content is scary good.

The truth is a well-written copy could save SEOs much trouble by helping them add meaningful text content to database-driven pages, like product categories, city-driven pages and more.

So can AI be an SEO solution?

Not quite.

John Mueller has been saying that Google doesn’t like AI content, and has often been quoted saying that AI created content is against the Google Guidelines.

In August and also on December 6th, Google released the first ever “Helpful Content” updates. According to Google, this update, “generates a signal used by our automated ranking systems to better ensure people see original, helpful content written by people, for people, in search results.”

I added the bold to “written by people, for people” because finding and penalizing AI content was one of the main focuses of those updates.

Here’s some coverage on how Google may be treating AI-written content:

But can AI-written content really be detected?

Yes, even if it’s very well-written.

In fact, I was genuinely surprised by that discovery because to a human eye, it often seems undetectable. It often sounds natural and flows well, and can even add a lot of value (to me, it is not really “fluff” content that exists for the sake of adding more keywords to a page).

And yet, it can be detected, and if the tools below can detect it, so can Google:

GPT-2 Output Detector Demo

GPT-2 Output Detector Demo is open, free and requires no registration. As a result, it is sometimes down, because a lot of people seem to be using it at a time.

However it is scary good. Make it your first choice when analyzing content for possible AI impact.

GPT-2 Output Detector Demo

GPT – Writer

GPT – Writer is another free and open AI detecting tool. I have found it just a bit less reliable than GPT-2 but it still managed to identify that my content blurb was not exactly written by a human being.

GPT - Writer

Content at Scale

AI Detector – Content at Scale is another tool that I have found quite reliable. It gives your content a human score, so aim at higher percent!

Content at Scale

AI Content Detector

AI Content Detector is a free detector but it has a limit of up to 200 words. When comparing it to the above tools, I didn’t find it reliable enough.

Originality.AI

Originality.AI is the only premium tool on my list. It charges 0.1 cent per credit and checks your content for both originality and AI.

Originality.AI

It does a good job detecting AI-generated content, so it is a good option if any of the free tools is down. It also offers API access and saves your scans.

My Own Take

Overall, when running my tests, I didn’t find it too challenging to bypass those filters. It takes just a few minutes of editing to make the text sound the way I’d say it. One quick edit per paragraph got me past all those tests. It did take me a couple of minutes to edit about 300-word text, so it is not exactly effort-free.

There’s no way to tell if Google using more advanced tools though.

Manual detection

There are a few approaches you can take to try to detect AI-written content manually:

  1. Look for stylistic quirks: Some (but, again, not all) AI tools have distinct writing styles that may be noticeable to a human reader. For example, they may use unusual sentence structures, or they may repeat certain words or phrases more frequently than a human writer would.
  2. Check the content for factual errors: AI systems may not always produce accurate or up-to-date information, so factual errors could be a sign that it was written by an AI.
  3. Analyze the language: Some AI systems create content that just don’t sound quite normal, so analyzing the language used in the content for any irregularities or inconsistencies could be a way to detect AI-written content.
  4. Check for a lack of personal voice or perspective: AI systems may not be able to express personal opinions or write in a personal style, so a lack of personal perspective in the content could be a sign that it was written by an AI.

Obviously, these approaches are not foolproof and may not always be reliable for detecting AI-written content. So when in doubt (i.e. when you were not that one actually writing content), try the tools above!)