top of page

How you can tell if content is AI generated


Featured image, split diagonally with one half a blue, wavey grid and the other half hands typing on a laptop


If you’ve been subject to a conversation with me around AI over the past few yars, you’ll know it has been an uphill struggle for me to get to grips with where I stand on using it within my career, both practically, and ethically.


Now that things have advanced somewhat, I can personally see the benefit of cautiously using gen-AI tools for brainstorming, getting a jump start on research, and outlining articles - but I draw a hard line at letting tools write for me.


Knowing how to spot AI content is super important for anyone sitting in a SEO or content role because unfortunately, when you can tell something is AI, it usually isn't very good.



The impact of AI on your website

Hosting AI content on your website can negatively impact search performance. That's a fact. A study by Originality.AI that was conducted during the latest Google algorithm update showed that 100% of the websites given a manual action penalty contained posts that were AI generated. Even more importantly IMO, it absolutely sucks for audiences.



 


Alright then, how can I tell if something is AI generated?


Easily. Here's my advice on identifying AI content based on experiences I’ve had over the years. 


📣 Sidenote: Today, I’ll be referencing Google’s own gen-AI tool Gemini for my examples because she’s who I’ve been chatting with recently.



1) Tools

Detecting the use of tools by using other tools feels a bit meta, but it is an efficient way of flagging AI generated content. 


If you want to keep this cost-effective, there are free tools out there, but I would recommend checking content on a few of them as I’ve found some to be more accurate than others.


 For tools that have an option to check via a URL or by copy and pasting content, I’d also suggest going for the copy and paste option. I’ve had a few instances where content that I know to be AI generated isn’t judged as such when going off URL based checks because the tool hasn’t been able to read the content within the page somehow.



🤖 Quillbot


  • Price: Free, premium options available

  • Limitations and features: Only check 1,200 words on the free version




🤖 Copyleaks


  • Price: Free for a few tests, then plans start at $13.99 p/m

  • Limitations and features: 1,200 credits per month on the basic plan, but this does include a plagiarism checker too




🤖 GPTZero


  • Price: Free, pro options available

  • Limitations and features: Only check 5,000 characters on the free version




2) Repeated phrases

Language models repeat words and odd turns of phrases that you wouldn’t normally use in your writing. These can be inappropriately placed, or just feel jarring to read in a sentence. 


There are endless weird repetitions, and I’d encourage you to go ahead and poke the bear a bit here to see if you can spot any by using the tool to write pointless articles (this can be quite funny too if you’re having a slow day in the office).


These are some I come across a lot that make me feel sad:


  • Fear not!

  • Happy hunting!

  • Elevate X to new heights!

  • Ultimately

  • Unleash

  • In addition

  • In contrast



3) Accuracy of information

It’s important to double check any information that comes out of an AI tool, especially if you are then planning on sharing this information with other people. 


Tools like ChatGPT and Gemini can now access up to date information, but they aren’t perfect by any stretch of the imagination. 


Sometimes it’s really easy to spot a mistake, like when I was using Gemini for some initial research on an article about types of laboratory robots, and it returned this…


A result from Gemini that talks about a type of laboratory robot, a 'washing robot'. However, the AI generated content has produced an image of a robot that has a washing machine in its abdomen
Source: Gemini

But there are times when the tools' mistakes aren’t as obvious as an image of an anthropomorphic washing machine, and it’s your responsibility to make sure you aren’t spreading misinformation.




4) Formatting

Generative AI tools tend to follow formatting patterns that are hard to unsee once you notice them. I’ll show you some examples of Gemini patterns below.


  • Lists with colons: When Gemini returns a list as an answer, it almost always does so in this format, with a bullet point or number, the main list item in bold, a colon, and then a paragraph of text.


Obviously, humans might also like this formatting and use it anyway, but seeing this in articles sets off my AI alarm.



Two examples of Gemini results for the prompts 'Can you give me a list of the best kinds of bird food?' and 'What were the most popular singles by David Bowie?'
Source: Gemini

No headers, only colons: instead of headings, Gemini often uses its favourite combo of bold text and colons. 


This is very similar to the point above, to be fair, but worth highlighting.


I’ve stumbled across articles that look exactly like the below where the publisher hasn’t bothered changing anything.



An AI generated Gemini response for the prompt 'Write an article about the best type of baked bean'
Source: Gemini



5) Structure

Articles generated through Gemini have very predictable, dull structures. I tried out a good few prompts asking the tool to write me articles on a variety of thrilling topics (beans, rugs, flowers - my favourites!), and found that each one had a similar layout to this…




Header: with a colon of course


A very generic introductory paragraph that waffles on for a bit about the topic.



Point one: Tedious text outlining the first point of the article, about 50 words or so.


Point two: More, more, more, here the tool might call Heinz beans tender but ‘one-dimensional’.


Point three: another not very original point with 50 or so words, but surprise:


  • Bullet points: with colons, of course, to spice things up a bit


  • Lists with colons: keep things nice and interesting


  • Of course: all these bullet points have around 20 - 30 words, so they are a bit more substantial than I’m making them out to be


Conclusion: often, this will actually say ‘conclusion’, but sometimes it is something dramatic like ‘The Final Verdict’. Around 50 words again.


Bonus Tip: if you’re lucky, you’ll get a bonus tip, too.



People using AI to write their articles will generally flesh their articles out, change the structure, and alter the wording. But you’d be surprised how many people don’t.


A test I’ve conducted before is to put the article title into the big gen-AI tools to see just how similar the outputs are to the article in question.



 

Found some AI content?

If you've followed the above and have successfully spotted AI content across the websites you manage, you should hustle to get a handle on it. 


There was a time that I was being asked to use inferior tools to help with my work, tools that I tested over and over again with the results being… well, crap. Aside from feeling it wrong to create content with AI from an ethical POV, these tools felt inefficient to me because as someone who spent every day planning, writing, and reading content, I could instantly spot AI wording. It took longer for me to beat a piece of generated copy into something valuable and human sounding than it did to write it from scratch.


Using AI to do the legwork isn’t always bad, but no matter how advanced tools claim to be, you can’t get language models to think for you. If you want to truly position yourself as an expert in your niche, then it’s only right that you take time to create content that provides genuine substance and adds value for users.


Comments


BREE SHEMILT

SEO, content writer, and creator of Shemeo.

BREE SHEMILT

SEO, content writer, and creator of Shemeo.

bottom of page