Think AI content is safe to use for SEO in 2025? The results of this case study might change your mind.
I will show you precisely how easy it is to detect content generated by all of the most popular AI content tools.
Now you might point to Google’s statement that:
“appropriate use of AI or automation is not against our guidelines“
But are you ready to bet your business on that?
Google already penalized sites that were abusing AI content in March 2024.
Before that, Google were actively rewarding sites that published AI content with more traffic. But then, Google killed them all which have never recovered:
Plus the Google Search Leak, DOJ Anti-trust case, and our own SEO Myths case study shows a history of inconsistent statement vs action.
Always take what Google says publicly with a grain of salt.
My point is simple:
Using AI generated content is an uncontrolled risk because you never know how Google are going to change the algorithm. So, why publish AI content that can be easily detected as AI generated?
It’s amazing how many people are taking that uncontrolled risk with absolutely zero clue of how easy AI content is to detect.
And problem with SEO’s is we always follow the same 3 step process:
Over the last 20 years, that has happened with keyword stuffing, duplicate content, spun content, anchor text, schema stuffing, scholarship links and more.
And now I’m seeing just about every writer, SEO and agency abusing AI generated content at scale. Google already gave us the warning shot in March 2024 but since then, things have got much worse.
For example:
I’ve had 3 consultations recently where US based agencies are using ChatGPT to publish medical advice along with specific drug and dosage recommendations on their clients blogs with zero oversight.
In every case, the agency had not informed their medical based clients that they were using AI to generate the content. Disgusting, immoral and unethical.
So with all of this in mind, I wanted to show you just how easy it is to detect AI-generated content in 2025 using publicly available tools.
The test process was simple:
- Generate content using 14 different popular AI tools
- Run the content through 11 AI content detectors
And just to spice it up a bit, we compared the results to our original AI content detection test from January 2023 which gave similar warnings.
Here’s what happened…
What Will I Learn?
We wanted to make this test as comprehensive and fair as possible.
Here’s exactly what we did:
Step 1:
- We generated 5 articles with 14 different AI tools.
- We used the same topic for all articles to ensure fair comparison.
- That’s 60 unique pieces of content in total using the most advanced AI models, including GPT-4o, Claude 3.5 Sonnet, and Gemini Flash 2.0.
Step 2:
- We ran every single article through 11 popular AI content detection tools
- This included multiple rounds of testing.
The results are based on an average detection score, so it’s not taking into account just one output in case the AI generator had a brain fart.
We also included 5 pieces of my own human-written content to ensure we had a reliable baseline to compare the AI content against.
The goal was clear – See if AI content detectors can really detect AI content.
Here are the results…
In the graphic below, you can see exactly what happened when we tested each popular AI content writer and generator against the best AI detection methods:
The number for each tool represents the AI detection score (out of 100).
The higher the number, the more confidently the detector flagged it as AI-generated content:
| Originality | Sapling | CopyLeaks | GPT-Zero | Quillbot | Winston AI | BrandWell | TurnItIn | ZeroGPT | Phrasly AI | Writer | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Human | 27 | 4 | 0 | 3 | 0 | 3 | 0 | 18 | 8 | 5 | 1 |
| Article Forge | 100 | 99 | 41 | 63 | 0 | 66 | 90 | 90 | 85 | 83 | 10 |
| ChatGPT 4o | 100 | 100 | 99 | 99 | 96 | 94 | 84 | 84 | 81 | 81 | 20 |
| Claude Sonnet | 100 | 100 | 100 | 97 | 98 | 96 | 72 | 53 | 47 | 46 | 20 |
| DeepSeek | 100 | 100 | 100 | 86 | 80 | 50 | 58 | 45 | 33 | 33 | 19 |
| Frase | 100 | 99 | 100 | 89 | 78 | 76 | 72 | 72 | 46 | 51 | 12 |
| Jasper | 100 | 90 | 100 | 91 | 84 | 85 | 84 | 84 | 60 | 59 | 10 |
| KafKai | 100 | 100 | 100 | 98 | 92 | 54 | 90 | 90 | 84 | 83 | 25 |
| Koala.sh | 100 | 91 | 100 | 89 | 88 | 65 | 78 | 32 | 23 | 23 | 9 |
| Phrasly AI | 100 | 100 | 100 | 96 | 100 | 95 | 32 | 77 | 84 | 82 | 17 |
| Rytr | 100 | 100 | 99 | 100 | 96 | 91 | 84 | 90 | 97 | 96 | 16 |
| ScaleNut | 100 | 98 | 96 | 70 | 54 | 75 | 78 | 78 | 60 | 61 | 10 |
| SurferSEO AI | 100 | 100 | 100 | 99 | 85 | 90 | 66 | 90 | 58 | 57 | 9 |
| WriteSonic | 100 | 85 | 100 | 90 | 87 | 41 | 25 | 25 | 40 | 0 | 3 |
| Google Gemini | 100 | 100 | 100 | 85 | 97 | 99 | 90 | 65 | 93 | 93 | 27 |
The results are mind-blowing…
Straight off the bat you can see that all AI content was detected with 100% accuracy by Originality.ai.
Sapling and CopyLeaks also caught all the AI content with a high degree of confidence.
Interestingly though:
My five human-written pieces of content were the only ones that passed all of the AI-content detection tools.
They barely topped 20% on the major AI detectors which shows these tools can accurately distinguish between AI and human writing.
Bottom Line:
These AI detectors are only getting better and have kept up with new generative AI models. So if I can detect AI content with publicly available tools, you can bet that Google can as well.
The question is:
Do you want to gamble on the fact that Google won’t correct SEO’s who are abusing AI content again?
For me, the risk far outweighs the reward.
My team and I will continue to produce quality human-written content that provides real value.
I’m not saying that AI content can’t rank today, but I am saying it is an uncontrolled risk that is currently being abused at a huge scale. The second correction is coming, just like it has many times before over the last 20 years.
But for the sake of context, let’s look at how these 2025 results compared with our original AI-detection test in January 2023. Remember, Google nuked a ton of AI sites in March 2024.
Looking at our 2023 results compared to 2025, it’s clear the game of “cat and mouse” has dramatically shifted.
In 2023:
But then, in 2025:
- No AI tool consistently passed detection across multiple tools
- Even the most advanced AI models (GPT-4o, Gemini Flash, Claude Sonnet) are instantly caught
- Detection accuracy has skyrocketed to near-perfect levels
What got me is how one-sided this game now is.
AI generative models have advanced significantly over the last couple of years.
But, AI-content detection has evolved faster.
Remember:
If it’s written by a machine, it can be detected by a machine
And that’s never going to change.
Comparing the results in these two case studies that were just 2 years apart, reinforces my point that using AI content for SEO is a major uncontrolled and unnecessary risk.
These include some of the most recommended AI content tools from even big names in the SEO industry.
Let’s take a more in-depth look at each of them.
ChatGPT is still the most used generative AI tool on the planet, with a 39% market share in the US and a 37% market share in the UK.
Here’s how it performed in our test:
- Model Version: GPT 4o
- Avg. Detection Rate: 85.27%
- Originality’s Detection Rate: 100%
The premise is simple – Ask ChatGPT to write something (or ask a question) and get back a well-worded piece of content in a number of seconds.
Magic, right?
It even includes advanced capabilities such as:
ChatGPT has a free and paid version, which is why so many users have adopted it, but it failed all of our detection testing.
WriteSonic is an AI content marketing platform that helps you plan, write and optimise your SEO content.
They have a suite of tools to help you produce AI content that sounds like you. This includes training your own AI model so it writes just like you.
How did it perform in our test?
These are the results:
- Model Version: WriteSonic 6 Instant Article
- Avg. Detection Rate: 50.82%
- Originality’s Detection Rate: 100%
WriteSonic actually performed the best over every tool we tested in terms of AI content detection.
But that’s still over half the content being detected as AI, putting it into the “fail” category. Originality.ai flagged every piece as AI.
WriteSonic is very focused on content marketing in general.
That means the platform is designed to help you generate:
They also have “PhotoSonic”, which generates AI images.
Keep in mind that AI images are also detectable.
Koala.sh launched in April 2023 with a massive lifetime deal.
Koala was second in terms of AI detection:
- Model Version: Koala & Claude Sonnet 3.5
- Avg. Detection Rate: 59.91%
- Originality’s Detection Rate: 100%
The tool essentially helps you create custom one-click, SEO-optimised AI articles in seconds. All you have to do is:
The big difference is that you can create a custom outline and add custom prompts to each section.
This should (in theory) help you get a very unique content output. But it still wasn’t enough to get anywhere near passing our AI content detection test.
DeepSeek burst onto the generative AI scenes after creating a better AI model than GPT-4o for up to 95% cheaper.
They are a major competitor to ChatGPT.
Their results in our test were:
- Model Version: R1
- Avg. Detection Rate: 63.91%
- Originality’s Detection Rate: 100%
Despite having a lower max output token length that was less than other models, they performed better in the test.
It’s nowhere near close enough to pass AI-content detection, but it’s an interesting aspect of the case study either way.
DeepSeek’s specialisations include:
Their open-source platform gives developers access to create their own AI models and tools.
Article Forge is also primarily designed for long-form content.
It creates 1,500-word articles in less than 60 seconds, including:
You can even publish the content directly to WordPress from the platform.
Article Forge promises quality and unique content. The content quality is subjective, but one thing is for sure…
It’s easily detectable-
- Model Version: Article Forge
- Avg. Detection Rate: 66.18%
- Originality’s Detection Rate: 100%
This is despite them claiming on their site that their articles will “Pass AI Content Detection”.
Regardless of their claim, the test results show their content isn’t unique and is easily detected as AI-written content.
Scalenut is like an end-to-end AI-powered content writing and marketing platform.
What I mean by that is you can use AI to do the:
They have a traffic analyser built into the platform to measure the results. They even have a humaniser to make your content seem like a human wrote it.
But Scalenut content still didn’t pass the AI detection test:
- Model Version: Scalenut
- Avg. Detection Rate: 71.00%
- Originality’s Detection Rate: 100%
Scalenut offers many other SEO AI writing features, including content briefs, keyword planning, SERP analysis and more.
Frase is a powerful tool that started out by helping writers create website content and optimise their content perfectly to rank higher in Google.
It has evolved significantly over the last few years, developing its own AI writer to produce content faster.
They are also part of the Copyrytr group, which has a collection of AI tools including:
This has helped them develop their own AI content generation models faster.
How did the content perform in our test?
Worse than last time:
- Model Version: Frase
- Avg. Detection Rate: 72.18%
- Originality’s Detection Rate: 100%
Frase is a great platform for helping improve and optimise your content. But I would stay far away from the AI writer.
Check out my complete Frase review to learn more about it.
Claude Sonnet 3.5 is a revolutionary AI model.
We used it to create our free AI Google Quality Rater tool because of Claude’s powerful analysis features.
It’s also said to be the best writing model, powering most of the popular article writers available at the moment.
That’s why I was excited to test it out!
But, the AI content was easily detectable:
- Model Version: Sonnet 3.5
- Avg. Detection Rate: 75.27%
- Originality’s Detection Rate: 100%
Claude Sonnet 3.5 is very useful for a lot of different things. It has excellent:
Just don’t use it to generate content.
Jasper has been the market leader in AI content creation for a long time.
They were one of the first tools to be released and have some of the most advanced AI writing features of any tool I tested.
In our last AI content detection test, they were also one of the few tools that could beat AI detectors.
But not anymore:
- Model Version: Jasper
- Avg. Detection Rate: 76.91%
- Originality’s Detection Rate: 100%
Jasper uses its own AI models alongside Open AI (ChatGPT) and Anthropic (Claude).
They’ve become more of a complete marketing platform, allowing you to create brand voice guidelines and build custom workflows for almost any type of writing.
The quality of their writing output is superior to most of the other tools I tested by a long-way.
The platform allows extras like team management, custom knowledge base and art generation.
SurferSEO also started out as an SEO research and content optimisation platform.
They were one of the first tools to develop a one-click AI article writer.
You add your keyword, select your settings, review the AI-generated outline and click generate.
Easy, right?
SurferSEO AI scored 9th overall in our test:
- Model Version: SurferSEO GPT 4o
- Avg. Detection Rate: 77.64%
- Originality’s Detection Rate: 100%
SurferSEO is still an incredible tool that I rate highly and use almost every day!
It has a ton of valuable features like:
… and more to ensure your content is optimised properly.
Just avoid the AI writing features.
Learn how I leverage SurferSEO to rank 1 with my complete SurferSEO review.
Kafkai is an AI content generator that also specialises in competitor analysis.
The unique thing they offer is a proven SEO workflow.
You start by using the Kafkai platform to find easy keywords your competitors are ranking for. Then, you write content to compete for those keywords.
Make sense, right?
You target easy keywords and generate more traffic.
But like every other platform, their AI content is detectable:
- Model Version: Kafkai
- Avg. Detection Rate: 83.27%
- Originality’s Detection Rate: 100%
I like the process they bring to the market. Although the AI content reads well, it was easily detectable by almost every content detector we tested.
Phrasly AI started out as just an AI humaniser.
But they’ve developed a couple of other products, including AI writing.
The biggest differentiator is that Phrasly is simple to use.
My thought is that with a built-in AI humaniser and AI detector, Phrasly would produce content that passed the AI content detection test from the get-go.
That wasn’t the case-
- Model Version: Phrasly A
- Avg. Detection Rate: 83.91%
- Originality’s Detection Rate: 100%
They were in the bottom three of all the tools we tested. The AI content didn’t even pass their own detector.
In saying that, they do currently have the best AI rephrasing tools, and the platform’s simplicity makes it easy to use.
But the AI content is still easily detectable.
There is no doubt that Google Gemini is powerful.
It’s considered to be one of the four big AI models at the moment (along with ChatGPT, Claude, and DeepSeek).
But not when it comes to producing undetectable AI content:
- Model Version: 2.0 Flash
- Avg. Detection Rate: 86.36%
- Max Token Length: 8,192
Like Calude and ChatGPT, there are a ton of other great use cases for it.
Google Gemini 2.0 Flash is good at anything related to reasoning and problem-solving. It also has advanced coding capabilities and handles large, complex prompts well.
Rytr is primarily an AI writing assistant that is best at coming up with different angles for copywriters, marketers and entrepreneurs.
Rytr is one of the most affordable AI writing tools with a cost-effective unlimited content writing plan.
This has made it extremely popular over the last couple of years.
But it scored the worst of every tool we tested:
- Model Version: Rytr
- Avg. Detection Rate: 87.91%
- Originality’s Detection Rate: 100%
Rytr also didn’t do well in our first AI content detection test in 2023. And it has continued down that path.
Where Rytr really stands out is marketing copywriting. They offer very attractive pricing and a simple platform to generate copy quickly.
You can generate in 30+ languages, including tone adjustments and writing style.
But I would stay far away from any SEO content generated by Ryrt.
Here’s the list of AI Content Detection Tools we used:
I’ve used and tested a lot of these tools over the years, so there is quite a bit for us to unpack here.
But based on the results above:
🏆 Best AI Detection Tools
What about the best AI detection tools?
These are the three winners from our test results:
- Originality.ai: Clearly the best overall, catching all the AI-generated content with ease
- Sapling: Came in a very close second with high AI detection rates
- CopyLeaks: Delivered excellent AI detection rates and did the best job at identifying my own content human-written
Personally, I will continue to use Origniantly.ai in the future, with CopyLeaks as a backup and second opinion.
You can learn more about each AI content detection tool used in the test below:
Originality.ai is the OG AI content detection tool.
They’ve got the most consistent results and have only continued to improve over the years.
They were the only tool on the list that essentially detected every piece of content as AI, with an average detection rate of 99.94% from all tests.
That’s incredible!
Originality.ai is also one the only tools that detects humanised and paraphrased content. That’s why they made our best AI SEO tools list.
The interesting part is that Originality.ai did detect my human-written content at an average of 27% AI from all tests. So, there is still some room for improvement.
But, Originality is good at what it does and is continuously being developed to stay up to date with modern AI writers.
It’s the one I trust moving forward!
Want to learn more about this tool? Read my full Originality.ai review.
Sapling’s user interface is basic.
But don’t let that fool you…
The team behind Sapling have built an accurate AI content detector designed to detect even the latest AI models.
Sapling scored an average detection rate of 97.21% from all articles and even handled my human-written content much better than Originality.
Sapling is simple, easy to use and most importantly – Accurate.
Copyleaks has more of an education focus.
They started out detecting AI content that students submit for university and school assignments.
Because of the way they trained their AI detection model, you get far fewer false positives than other tools. And our test results showed the same…
Copyleaks was one of the only AI detectors that scored my human content as 0% AI. And to top that off, they had an average AI detection rate of 95.30%.
This cements Copyleaks as one of the most accurate tools for both Human and AI content. Impressive!
GPTZero is an advanced AI detector that was inspired to stop ChatGPT.
It uses advanced variables to track AI models and detects their digital fingerprints.
GPTZero had an average AI detection rate of 89.40% which was far better than the results back in 2023.
They are currently developing their technology for educators to stop students from taking advantage of AI for college and school assessments.
Quillbot AI detection technology has developed significantly over the last couple of years.
They have become one of the most well-known AI detectors.
In my last test, I received a bunch of emails and messages asking me to test their detector. That’s why I added them to the mix this time.
Quillbot had an average AI detection rate of 80.97%, placing it just inside the top 5.
They’re popular because of their suite of tools, which include paraphrasers, grammar checkers, plagiarism checkers, translators, and more.
Winston AI is more than just a content detector…
You can content to the platform and get a bunch of insights including your readability.
Winston AI had an average AI detection rate of 76.91% in our test.
Their user interface is easy to use, and the tool is robust. They’re also working on developing their models for education use.
BrandWell detects AI by looking for common patterns used in AI-generated content and then forecasts probable word choices based on individual models.
It colour codes every sentence on the page, highlighting the likelihood of the content being AI-generated.
They scored an average AI detection rate of 71.64% and definitely worked better against some tools more than others.
BrandWell’s goal is to create an AI detection algorithm that can even detect edited AI content accurately.
I agree, that’s the future!
TurnItIn, as the name suggests, has a strong academic focus.
They scored a solid average AI detection rate of 69.64% from all articles.
It’s clear that TurnItIn works better in the academic world. They even have a dashboard that allows for better analysis of AI-written academic papers.
ZeroGPT is all about speed.
Their main point of difference is fast AI content checking and bulk uploads.
As for the detection rate?
It was a lower-than-expected 61.10% average AI detection rate.
ZeroGPT is affordable and has a straightforward, easy-to-use platform layout. There is also a free version, but the site is littered with display ads.
As I mentioned before, Phrasly is better at humanising AI content than actually detecting (or writing) it.
And the results show. Phrasly scored a low 60.67% average AI detection rate for their score.
It placed it comfortably in the bottom 3 of all the AI content detectors we tested.
Last and very much least, was Writer.
I know that not all AI content detectors are built the same. But Writer really proved this point with a terrible average AI detection rate of just 14.77%.
That is incredibly low and essentially means that Writer passed most of the AI-generated content.
Writer has created a bunch of other excellent apps and features. But the AI content detector isn’t one of them.
Just spend 15 minutes reading through a Reddit forum, and you will see a multitude of people claiming that humanisers are enough to beat AI content detectors.
But the reality is different…
Humanisers aren’t guaranteed to work primarily because AI’s still producing the content. It might slightly lower the detection rate, but you’ll likely still get caught out.
Plus, tools like Origniality.ai are constantly updating their detection models to catch humanised content.
If that’s not enough for you, there are a few more reasons to avoid them…
The goal of a humaniser is to make the AI content “undetectable”.
How do they work?
The goal is to essentially add nuances that make it sound more “human”.
The problem is that this adds extra words into the sentence, disrupting the flow and affecting the readability.
You have to ask yourself…
Is humanising AI content really providing value to your readers?
And that’s not all.
Many humanisers resort to tricks like messing with the formatting, adding double spaces or replacing letters with unique characters that kind of look the same to the naked eye.
I mean, look at this result I had using a humaniser:
This is essentially what we used to call auto spinning in 2012, and spinning things automatically instead of manually wasn’t a recommend practice back then either.
The simple fact is that:
If it’s written by a machine, it can be detected by a machine!
The truth is that humanisers are still playing the game of cat and mouse. You might fool basic detectors, but you’ll likely get caught in the end.
They aren’t a foolproof solution for AI content and are still high-risk to use for your website SEO.
Can Google detect content written by AI content creators?
Short answer – Yes because if we can do it with publically available tools, so can Google who have invested more in AI than any other tech company.
And they will continue to invest a massive $75 billion more in 2025.
Not to mention the fact Google published a paper in 2020 showing how they can use text generation models to detect low quality content.
A simillar test by Jon Gillham, comprehensively showed that AI content detectors could easily detect AI-written content as did Surfer.
It’s also likely that email spam filters (the biggest is opearated by Google) and social media networks will start to integrate AI content checkers in the future.
This could mean that your emails go directly to spam or posts on popular social media sites get limited reach.
So are AI based content generators valuable at all?
Yes, if you use them properly.
AI generators are great for:
…for human writers.
From my testing, it’s clear to me that the best AI SEO tools can come up with some amazing ideas and angles for content creation.
But when it comes to actual content writing for a website – stay away!
If Google decides to penalise sites using AI content in the future, it could completely destroy an entire website.
Don’t get caught up in the hype.
Protect your site and avoid using AI writers for website content or anything to do with SEO. We won’t be using it to create social updates or emails either!
88 Responses
Join My Facebook Group
Subscribe to my YouTube
Connect on Linkedin
Follow My Twitter Page
Well thats interesting, but today Google does its own thing, in one side of their mouth and out the other. Traditional SEO right now can not be depended upon either. Google answers many queries with AI, and then they rank Reddit and Quora for almost everything. After reading your article (From An Email You Sent) I just did a search for “do lakes freeze in the winter” and guess what the results were? First Google answered it, then in the second spot was Reddit, then in the 5th spot was Quora, both forums. No scientific answers of how lakes freeze and what temperatures, just people with different opinions chatting in a forum. Oh and Reddit ranked twice, once talking about one specific lake. So there is much more at play, and I have seen great sites that have amazing human wrote content go from 500,000 visitors a month to 5,000 or even less. Search is going to be more spread out in the future and very quickly, as more AI’s turn into being able to answer all questions, AI’s are only going to get smarter, and Google should be the furthest thing from ones mind. I have not done a search myself on Google in months, until today, just to prove a point that Google does what ever it wants. Youtube and pinterest can drive just as much if not more traffic than Google did. They are both search engines as well. Think out side the box, lots of platforms can get traffic from.
No no no! I have tested many of my handwritten articles from years ago, and they come out high on the AI-written score. But I am pretty particular about good grammar and punctuation. If that makes a difference. Also, I haven’t tested this myself, but others have tested parts of the Bible, and other “legit” documents, and they too got high scores. Most of the content writers I know consider these apps to be nothing but a money grab based on fear-mongering.
Yes Will. You are right. You can insert thesis or books from 1999. And they can have 0-100 % AI-Score.
It remains a stochastic parrot, working in all directions.
Oh wow, I couldn’t have seen this at a better time. I’m just in the research phase before launching a blog, podcast, and YouTube (yes, all at once because I’m a glutton for punishment). I’ve been bingeing Search Logistics and some others in the process. American literacy is abysmal, so I’d been considering using Brisk Teaching AI to adjust reading level of spun content. I am definitely reconsidering that. I guess it’s back to researching what constitutes specific reading levels.
Matt,
I have to disagree with your thoughts on using prompts to humanise your writing, because I have been using a prompt that is a combination of multiple prompts that I’ve picked up over the last 8-9 months, and the result very rarely (I write blog posts and Medium articles every day) go above 5% AI detected on any detector software. With some, like ZeroGPT, I usually get 100% human writing.
I used ChatGPT to merge all the prompts into one, without repetition or fluff, and I will gladly send it to you if you wish.
Terrific data and thorough as usual Matt. One thing to point out is that the article you used to test the tools is SEO related, and I bet the AI tools now have SEO as a topic their models look out for more and might be better at detecting, then say, “preparing christmas diners for relatives”
I still think the message to avoid ai-generated content is still very sound, and moving forward Google will get better and better at detecting it – just like they have for every other seo “trick” over the years.
good stuff as usual. still miffed when you were sick and missed Thailand right before Covid 🙂 I have built an AI detection system called SentiScan and am just coming out of beta now. It’s not just for AI detection. It scans writing quality and gives advice on improving it. This isn’t a sales pitch because it’s not being sold yet. But if people want to test the scanner, ping me online. Won’t cost a dime.
I am leaving my job because I work in YMYL and I am being asked to use AI to create content. I agree with you Matthew, it’s disgusting to jeopardise our clients in this way, and they have no idea what we are asking them to sign off is unoriginal content that has the ability to tank their online businesses. I think it is unethical, but I won’t be around to be help responsible when that inevitable algorithm update hits
Can you please share the link where Google says that “appropriate use of AI or automation is not against our guidelines“.
I am just reading this article to enhance my knowledge regarding AI because I have heard a few people say that AI content can be harmful to you. So, I am just getting knowledge that behind the scenes, what say Google?
Here you go: https://developers.google.com/search/blog/2023/02/google-search-and-ai-content
I appreciate your work Mathew, I always have. I too have drunk the hype flavored cool-aid my damn self and I can see how people looking for that elusive ez-button would/could get sloppy using AI content for SEO. BUT… When done correctly, it can come up with some AWESOME (legit) content. As an SEO for 15 years+ that also owns a local retail shop; I figured out how to get chat-GPT to spit out 100% human written, plagiarized free content for each product we buy and sell! It’s not a silver bullet (damn close though)! Meaning, just create and then post. It still takes that human touch (and time) to format everything, tweak and rearrange some of the content etc. What is the “correct way” to do it? It’s all in the prompts. Here’s my list of 11 prompts I use for each product I want to write about from start to finish.- Enhanced Post Prompts (beginning to end) -1) I am a (Business Type) and want to attract search engine traffic from people looking to (Buy | Sell) (Main Keyword).Do you understand?2) can you create a list of LSI keywords associated with “(Main Keyword)”Do you understand?3) I want to write a long blog post that is SEO optimized using these LSI keywords throughout the content. Do you understand?4) this blog post is to be written in second person, from the perspective of (Business Name) making offers to (Buy | Sell) (Keyword) using the LSI keywords throughout the content in a natural way.Do you understand?5) My business name is ( ________ ), we have been a (business Type) in (City, State) for (amount of Time). We have five-star ratings with hundreds of satisfied customers.Do you understand?6) Create a blog outline with the above focus keyword and LSI keywords in mind.Do you understand?7) For the content, you will include the above information and LSI keywords written throughout the content with SEO optimized H2 headings for each topic point you came up with.Do you understand?8) Going forward, all of the above is to be used for your reference in writing a comprehensive, SEO optimized blog post.Do you understand?9) This content is to be written in second person, and in a conversational style of writing as it were coming from (Business Name). It will have a persuasive tone and appeal to do business with us.Do you understand?10) With the above instructions in mind, expand on the outline you provided. Do not include your instruction or reason for doing so.Do you understand?11) ContinueIT WORKS!!!Doing it this way, I have found the output to be damn near flawless each and every time! It adds that personal touch that (for me) has gotten buy AI detection with scores of 97% to 100% human every time! And… 100% unique, every time!Enjoy 😉
That is a very detailed process however does it pass ChatGPT’s own detection tool?
Hey MattThis is spot on… I have been using Jasper boss mode since it started. I now see more and more de-indexed pages from google for my client’s sites…. and I really think this new wave of chat interest will do 2 things… 1. We will see a proliferation of so-called “writing experts” pop up and tout their expertise in writing cheap content for small low-budget businesses, thinking they get a deal..only to find it penalizes their site in googles eyes and2. The massive increase in content on the web is going to be so overwhelming, that written content as we know it will not be a strong signal for search as it is now… these lame marketers will destroy that… and, as Google has already told us, they are now looking for E-E-A-T-S… specifically EXPERIENCE in the content that you are writing about… we will see a rise and sudden fall of these so-called experts pretty quickly
Indexing in general has got a lot harder but we’ll be publishing a case study about that soon!But yes I agree with your points, ultimately it’s a cat and mouse game and the cat is catching up to the mouse.
These AI detection tools can be tricked very easily, the question is what sort of AI detection does Google has?
I would assume they have some of the best resources on the planet for that given their investments
There is no such thing as duplicated content between different domains. Duplicate content exist only as internal thing around your domain. Period. And thats also a google statement.
I assure you when several high DR sites clone your entire site via proxy meaning they publish updates/changes just as quickly as you do – your rankings will suffer. We have had a couple of clients have this attack used on them, one in crypto and one in gaming.
Removing one or two punctuation or adding punctuation at a wrong place, most of the times makes 100% fake content to 100% real content. So getting around these publicly available AI detector is easy, not sure about Google though.
Yes the public tools are built with very limited budgets (often zero) and often by a one man band or a very small team. Google don’t have such bottlenecks.
Hi Matthew, I love this article! It was incredibly informative and comprehensive.You did a fantastic job of breaking down the different types of content AI can create and detail how Google can detect AI-generated content.It is important to note that while AI-generated content can be detected and penalized by Google, there are still some legitimate uses.For instance, businesses may use AI-created copy for internal documents or reports where accuracy is not paramount.Or, when creating website copy, businesses may appreciate the ability to quickly generate a large amount of content using an AI tool.Overall I believe that if used responsibly and with caution, AI-generated content can add value to specific processes. However, regarding SEO campaigns, AI-generated content should be avoided.Google has clarified that they disapprove of AI-generated content, which could lead to penalty action.Businesses must understand the implications of using AI-generated content and ensure they do not violate any guidelines. I appreciate the thoughtful insights you provided in this article on how businesses can use AI responsibly.This stimulates an exciting conversation about how we use technology in our marketing efforts, and I think we must discuss these topics openly.Thank you for such a valuable piece!
Yes there is lots of legitimate uses especially for producing anything that is not front facing content or part of email marketing.
AI gives you the bricks, you build the house.
Exactly!
Hi Matt,So, today I started my own case study to test onsite BASIC optimisation of a website (a few of them) only using the ChatGPT to do so. Here is my list:Keywords research – I’ve fed the AI the most information I can to spit out the best longtail keywords for a particular niche, then I am cross referencing that with data from other keyword tools to see the actual numbers of searches. I can, I think anyway, explain in detail to the AI, how to account for the search numbers locally, then asking it for the rankings…still trying to figure this one out.I will then input some text based on the niche, basically just a 300 to 600 word Homepage for a site I am working on, and ask it for the best headings (h1, h2) for the output and finally, Ill ask for the meta data. I am really looking forward to seeing how this all plays out.My last project, for the AI; I want to see if it can really build me a website from scratch. I need to fully understand how to input this all into the AI in order to teach it….if that’s even the right word.Thanks so much for this case study, I’ve been obsessed with it since it’s publication.
That’s a very interesting project are you doing it on a new domain or aged?
Thanks for the in-depth analysis. My main fear around AI content comes down to this… if I’m outsourcing content and I don’t have funds for investing in AI-detection tools, what do you suggest to make sure content writers are producing unique, accurate content that isn’t AI-generated?
Unfortunately managing writers and reviewing their work is a continous process that never ends. You should be using QueText at a minimum to check every piece of submitted work.
You explained AI content and what problems to face in the future but what about the spun content? Is it safe or not?
Human spun content when done properly is unlikely to be detectable at any point, however creating spun content to that calibre is labour intensive.
I am caught up in the AI ‘movement’–Ha!However, I was brought back down to earn by your surpurb review of all the AI players.Thank you.I have yet to commit to ChatGPT–or any tool, but your comparisons have given me direction!Incidentally, in one of the reviews you mentioned ‘sweet’ instead of suite–was that intentened?Thanks again for a very cogent discussion…you’ve opened my eyes to possibilities–and have given me a track to run on–instead of the wide open spaces!-Steve Lanning
Sweet, thanks for the headsup I fixed it!
Paraphrase bro, like everyone knows that. How else you make it unique…
Can you explain more please?
This is super frustrating. I am literally in Originality manually typing text and scanning it and it comes up AI. IT IS NOT AI — IT’S ME. As a marketing professional, I want to assure my clients that the copy I provide is human-generated since it is. However, every single “ai copy checker” I’ve used has identified my written copy as AI, even when I switch to a more casual voice. The writer.com and Originality are actually the worst, however, the ai detector at copyleaks.com seems to be much more sensitive to human content.Overall, this is terrible. This is already affecting my income as I have to waste time rewriting human content that’s being detected as ai. I really hope this is resolved soon. 🙁
Would you mind replying with a 300 word sample of your writing? Have you been in touch with either of the detection tools to report the false positive?
It was written by my own hand?
Hi Mathew, Thank you for the interesting AI test. You can run these Ai App content through another Ai App which will give them a more human touch. I don’t want to tell you, as it would be inhumane.Kind regardsBryn
You mean take the AI output and run it through Quillbot?
This is awesome Matthew! I’ve shared our own results with Originality.ai with some other SEOs and inevitably, many of them choose not to see it as a warning of the perils of AI content, but a challenge to game the system. One popular method to avoid AI detectors spotting the content as AI-generated is to use AI rewriting/spinning tools like Quillbot. I haven’t done the tests yet, but I feel like this will still be readily detected by QueText, and more importantly Google, since the content and context are still the same.
I would like to see the results of the test if you did! I showed you mine… 😛
Hey Matthew, great article. Have you done any single variable tests to rank an AI Generated article? Despite being flagged as AI by the detectors, does the article or post get ranked against competition?
Take a look at https://www.kevin-indig.com/how-well-does-ai-content-perform-in-seo/
AI definitely helpful, but we can’t use it in raw basis. I am still using with few ways around it. However it’s still involved 20 % of human work.
I think its good for planning, researching and creating content specifications for humans to execute.
It would be great to know if the content was used verbatim from these tools.If you were to edit and add your own original elements, how would the results look?
The content was copy and pasted verbatim with no editing or tweaks. I have now added links to the source content for you to download.
Great post. I’m glad someone addressed it in detail. What do you think about sales pages and landing pages? Especially when you’re using paid ads to drive traffic to them.
If organic traffic isn’t an issue – I don’t see why you couldn’t use it for paid campaigns. Perhaps Google could update their quality score in Adwords to reflect AI content on the landing page https://searchengineland.com/google-study-generative-models-redictors-page-quality-392460 but it would be easy to fix if they did
Thanks for sharing and you provide best information for SEO. result is not acceptebel but and i am very sad for result, but it’s help me to save my website. i never use AI-generated content.
Yes I wasn’t happy with the result either! I thought I could finally retire and have AI do all of these case studies for me =D
Excellent, Thanks for helping, I was not using AI Content, but other team and SEO’s buddies are using it and keep suggesting me to use it. Still, I was not using, the tool to create content from AI tools, so it will open other SEO’s Eyes if the want to take long-term benefits from SEO and Google.
And I’m sure they are probably seeing that content rank at the moment!
Great post! Thank you for the case study. I was little concerned about AI content. But I’m glad that I have taken great decision to not to put AI content on my website. I agree with you that Google can easily detect that. Chatgpt is useful for generating content ideas and outlines as you said. Thank you for the solution.
No problem glad to help!
Are you now saying AI writing tools can’t be used as an assistant.
What I’m saying is what is published above in the article. I don’t see any reference to saying “AI writing tools can’t be used as an assistant.”
Thanks for a great review! Do you think using AI-generated meta tags for improving CTR is ok with search engines (assuming this is short-form content and it’s harder to know if it was AI)?
That depends on how aggresive Google choose to be in the future https://searchengineland.com/google-study-generative-models-redictors-page-quality-392460
Seeing spun content here with all green checks. And ChatGPT with all red crosses made me cringe so hard. My eyes are tearing up right now And to even lower the level of ignorance and competency of your knowledge in any of this, that makes me cringe even more, is seeing Jasper and all the other crap in here praised more than ChatGPT which is GPT3.5’ish, while Jasper is using GPT3 api to provide you with those articles. You litterly said my Suzuki is better than your Ferrari hahahah, unsubscribe button has never been clicked so fast. Matt, you lost it brother, accept your defeat.
All I did was take the output of tools and use it as an input for other tools and screenshotted the results.I’m sorry that made you emotional but the data is the data.If you don’t understand why human spun content doesn’t get detected as AI content, it’s because human spun content is in fact, written by a human. And yes tools like Jasper are BASED on GPT3, however they have trained the model with their own action specific data sets which greatly impacts the output.You didn’t unsubscribe…?
Excellent article, Matthew and team. As an SEO, I have dabbled over the years with some of these tools. Most of them, I didn’t use for any length of time as I wasn’t happy with the results. Jasper is one that did stick for longer but even before reading this article, I felt it wasn’t the right route. I am referring to the AI route, in general.As an author and media producer, using non-human content doesn’t sit easily with me. One thing that we did use consistently was voice to text, originally with Dragon Naturally Speaking. I very much liked writing like this and then going back in and refining the work. It is quite a spontaneous way to write and easier than being on a the keyboard all the time.As for spun content, I hate this with a passion so yes, I was surprised initially to read the results. But I guess it does make sense. As a human is doing the spinning.Bring back good old fashioned writing for a purpose with creativity and proper research, that’s what I say!! 😉 Google does recognise it, I have had many experiences of this.
I think we agree with our outlook! You just can’t beat engaging enterating content backed with research and data.
Very interesting research Matt, and timely too! Without AI detection I had visions of the web turning into a giant spam bot.
Arguably, it’s already a giant spam bot =D
I believe the fundamental problem is that people are lazy. Its a lot of work putting a decent, well-written and informative web article together. But yes, I think there is a lot of hype at the moment and in the future there will be a lot of “publishers” that regret cutting corners.
Yes we saw exactly the same thing in the golden era of auto generating spun content about 10 years ago except spun content was only applicable in the SEO arena whereas ChatGPT is applicable in every industry.
You didn’t clarify what sort of articles you wrote, and how you instructed the AI to write those articles. It would have been useful to actually view some of the articles. If you just give the AI one word, or perhaps one sentence then the final work is more likely to be ‘risky’ than if you give paragprah level instructions. My 1500 word AI article gained 98% Human from Writer… and the only plagarism QueText found was the (my) URL I copied/pasted the text from (into their tool).I would strongly imagine that Google will leave margins for error, in penalising AI work (because if something is only 2% AI written – do you really penalise the 98% human written)?
I have updated the article with links to all the source content for your reference and a link to https://searchengineland.com/google-study-generative-models-redictors-page-quality-392460
This came at exactly the right moment. In the UK, the Good Morning TV programme was talking about AI in education and said that the UK may have to go back to just exams as students are getting A+ grades by using it for GCSE and A Level course work. My son had to write a short essay this morning for college so I suggested he used AI to see what happened. He did and when he read it out was told it was easily the best in the class. He then told them how he had done it. I am just starting a new blog and have thought that I would need to use Ai just to keep up with everyone else but this article has changed my view.So as a thought, could I use AI to give me the bones (plus a little flesh) of a post and then Spin it as I, like you, did a number of years ago? Would this make it unique enough to get past AI and plagiarism detectors? Would this be a way forward to at least speed up the process?I would be really interested in what you thought about this.
Well since publishing OpenAI have released their own detection tool https://platform.openai.com/ai-text-classifier which we have now updated the case study with.It’s likely any students that were using Open AI to produce their work secretly are under pressure right now.I think using AI to create outlines and find common questions is the best use case. Any actual writing for whatever reason should be done by a human hand.
Very comprehensive, and thanks for that post. I have tested some of the data I generated with chatGpt and it’s 99.9% fake. I had misgivings about using it and I’m glad I read this post.Will use it for ideas but not for full text. Thanks again.
No problem Dave!
Thanks Matthew, this is something that definitely needed to be said given the amount of hype and all the new tools hitting our inboxes of late (MarkeingBlocks 2.0 anyone?) I was curious about this, “Content at Scale is similar to Writer because the tool was designed to produce AI-written long-form content fast.” and why you didn’t test their content generated against the detection tools? Might be interesting to see how unique their own content would be.Also curious if a combination of AI content + a top shelf spinner would work, a first draft and editor as it were?Thanks again.
I will add them to the testing!Yes perhaps a process that embraced an AI written outline, that was then written by a human writer would work. Spinning of course is optional depending on how and where you want to use that content.
Ai content is against Google guidelines but what about human-edited Ai content? Google is against automatically generated content which I guess copying and pasting Ai generated content directly on your website without editing or fact check.What about this tweet by Danny Sullivan – https://twitter.com/dannysullivan/status/1589681355868504064?s=20&t=ZrOVzgU5B_ciSBalduNSag
There are some nice uses of AI generated content out there, for example instead of reading 100 comments/review from users – AI can write a quick summary from that.Its unlikely the wider SEO community will use it that way though.Also please take a look at https://searchengineland.com/google-study-generative-models-redictors-page-quality-392460
Very interesting read. For now I am convinced by your arguments and tests. Let’s see in the near future if AI detection tools will still work. OpenAI of course thinks they will not be able to very soon. Will be exciting 😉
You should run some of your own content through the tools to see what they find
Thank God you posted this. Hopefully people will actually listen to you! It’s just that, all hype!
I think its a pivotal change in history that will challenge many industries, but its not ripe for use in SEO.
Thank you.Excellent article. So many people are always looking for the easy way. Well I always say there is no easy way. Go through the process and create good quality content that grabs the end-users attention and give them some benefit and answers their question(s)That’s the secret,Jeff Goldsteinlmtw.us
That is exactly right Jeff!
Great study. Makes me feel relieved in a way but only time will tell. 🙂
SEO is always a cat and mouse game but we are repeatng histroy except before it was duplicate content, then auto spun content, then auto scraped and spun content and now it’s AI content.
Google have subsequently issued a clarification that they are not against AI content, just poor content written purely for search engineshttps://mobile.twitter.com/dannysullivan/status/1589681355868504064?t=QRmZkIj8SwOl8TkGYCAnOA&s=19
I’m sure the SEO community won’t use AI to generate poor content written purely for search engines, nor will writers try to pass off AI content as their own.
I agree with Matt, some human content when written specifically for search begin rankings can become and seen as AI even if its written by human.So in my view, don’t over optimize your content in eyes of SERP because it will look like its by AI, that’s what Google is trying to do here. Also, if AI can write human focused work then its good.