The truth about AI-generated content

Surprisingly, this article isn’t biased.  Of course, we want humans to be the default writers on the internet. Alas, wanting something isn’t enough to make it happen, but Google has decided it’s on the same page as us and it’s put actions behind its words.

What is AI-generated content?

AI-generated content is something that’s been created by AI (artificial intelligence) or an LLM (large language model).  Usually what happens is a person gives an LLM (like ChatGPT, Gemini or Claude) a brief and the LLM brings back a finished article in seconds.

Why is AI content unhelpful?

LLMs can only work with what’s already out there in the world – they’re not creating these articles using independent thought, or their own experience.  This can mean articles are of poor quality for several reasons, including:

• They can be factually inaccurate
• They tend to regurgitate what already exists
• They tend to sound artificial
• For all of the above reasons, AI-generated content doesn’t tend to stand the test of time, especially in Google rankings. We’ve shared proof of this later on

How can you tell if content is AI generated?

1. From how it looks and sounds. Multiple em-dashes, commas before the word ‘and’ and dramatic ‘clickbait’ style beginnings of paragraphs can be signs that content is AI-generated
2. Gemini use a tool called SynthID which is like a watermark added once text, audio, images or video are created, to help it to identify what has and hasn’t been created by a human.  You can find out more about SynthID over at Google DeepMind or this LinkedIn post from Miguel Ramalhosa explains it nicely

Does AI-generated content help with SEO?

Absolutely not. And don’t just take our word for it.  Forget the hallucinations and the blatant misinformation and take a look at these studies to see for yourself:

From 526,000 impressions to zero

SE Ranking ran an experiment where they published 2,000 AI-generated articles which weren’t edited by humans at all.  They did this over a period of 16 months, using 20 brand new website domains.

To begin with everything looked great. 71% of those new articles were indexed and they gained some good rankings and over 122,000 impressions, which grew to more than 520,000 in month 2.

Once month 3 came around, their rankings fell off a cliff – immediately.  97% of pages didn’t rank in the top 100 search results and after more than a year there had been zero recovery. Fast forward to the end of the test and month 16 and most of the sites’ visibility had completely flatlined.

Thanks to Tony Pataky for sharing this example on LinkedIn.

Blog traffic declined by 40% in 3 months

Thanks to Olly Hodgson for sharing an example from one of his clients on LinkedIn:

A client of Olly’s found their blog traffic dropped by 40% in just 3 months. He found that they’d published 23 new articles between November and February, each one written by a freelancer who had used the same AI prompt template to generate them all.

Then came March and Google’s latest core algorithm update.

Tony’s client wasn’t the only one to lose traffic after using AI-generated content.  The update had an industry-wide, overnight impact. Then came Google’s spam update, that same week.  Combined, these updates had a clear goal – stop rewarding sites where content didn’t add value.

A LinkedIn post by Spencer Robert of MarketSpire provided some interesting statistics:

• Sites with original research and proprietary data saw roughly 22% visibility gains.
• AI-generated content published without editorial review got hammered. Double-digit drops.
• Thin affiliate pages and coupon-only sites hit hardest.
• AI content farms lost 60-80% of their traffic overnight.

In the same post, Spencer warns:

“This is the clearest signal Google has sent in years. Original thinking gets rewarded. Scaled content abuse gets buried.

“If your content strategy is still built around keyword volume and thin blog posts, this update is a warning shot. Change now or that traffic is gone for good.

“The businesses winning right now are the ones producing original, structured content that both Google and AI answer engines want to cite.”

How should AI be used when writing content?

When writing, think of AI and LLMs as your second opinion, your proof-reader and your idea validator.

Second opinion

Tell an LLM what you’re trying to achieve with a piece of content. Run your plan past it to see if there’s anything else it thinks users would find useful, then use what it suggests as a heading or bullet point – just make sure you write the copy beneath the heading yourself.

Proof-reader

If you know your content can come across as clunky, abrupt or anything in-between, it can be helpful to run it through something like ChatGPT first to see whether it’s hitting the right notes.  Similarly, it’ll let you know about grammar or spelling mistakes.  Just be mindful to not let it re-write it for you.

Idea validator

This is especially helpful for Digital PR ideas. You can find out whether they’ve been done before, whether it’s something that journalists might find attractive or whether it’s something that’s potentially tone-deaf because of macro elements you might be unaware of.

Research

LLMs can also be great for research, but be aware that they do hallucinate information and make things up entirely. If you are prepared to fact-check everything it gives you (using manual Google searches) then go ahead, but otherwise we’d steer clear of anything it can’t give you a clear reference or citation for.  Even then, we would check whatever it said is on the website page it’s referencing.

Tools and web apps

If you’re interested in creating your own tools as part of your website content, LLMs can be a big help.  Check out courses like this one from Udemy to help you get started.

Help with content that ranks well on Google and in LLM responses

If you’re looking for someone to help you with creating content that consistently shows up in search results and LLM answers, contact our wonderful team of humans for a chat!

We’re experts at creating content that ticks every box for Google’s EEAT (experience, expertise, authority and trust) signals and we’ve found this works brilliantly for showing up in LLM answers, too.  Find out more about how we can tick those boxes for you.