AI content CAN rank. I’ve run a lot of tests, on small scale sites like this, to large scale sites getting millions of unique visits per month.
What I found is that on its own, many times AI content can rank, however it typically does need some human intervention in order to rank well. It also needs manual human copy editing in order to sound right.
In my testing, for best results, a piece of content that is anywhere from 75%-80% AI generated will still rank well.
This podcast was recorded before the “helpful content” update from Google, but my opinions have not changed. AI content can create good content, though much of it isn’t going to be groundbreaking or new, it can pull ideas and concepts together into a single article that is useful for a user.
On this podcast article, I wanted to talk about AI content and my tests on its success as a viable option for creating content that ranks on search engines.
I’ll start with some background on the latest publicly available AI system, an example output that I pick apart, some additional example outputs, and then my opinion on how to utilize an AI system to write content.
Background on the latest publicly available AI system:
First, a little background on where most of the AI content is coming from:
GPT-3 is the model that Open AI allows access to. It is the latest publicly available AI tool that you can use for content. Go ahead to their site @ openai.com – I’ve used several 3rd party tools that interface with GPT-3, as well as working directly within their API.
So, what is GPT-3?
GPT-3 is a text completion endpoint that allows users to input text prompts and receive generated text completions. The API is powered by a set of models with different capabilities and price points. In a nutshell, GPT-3 takes an input text and gives you output text. This can be in many forms depending on your input instructions.
- You can provide the tool several paragraphs and ask it to summarize.
- You can provide sample sentences and ask it to continue the content.
- You can provide a single sentence and ask it to write a paragraph or more.
- You can provide it a sentence with a sentiment label, and ask it to label additional sentences.
- You can provide the tool a topic and ask it to create a list of ideas or requirements.
- Use it to generate REGEX code even
- You can have it write a poem or creative story based on a single prompt.
It’s all based on what you want it to output, you put it in an instruction set.
There are 4 main engines for GPT-3, which have different characteristics and costs associated.
Starting from the most capable, but also the most expensive:
- Davinci – Davinci is the most capable engine and can perform any task the other models can perform. This includes tasks such as summarization for a specific audience and creative content generation. However, Davinci requires more compute resources, so it costs more per API call and is not as fast as the other engines.
- Curie – The Curie model is best used for tasks that require complex classification, text sentiment, summarization, and answering questions. The Curie model is powerful and fast, making it a good choice for tasks that require these capabilities. It’s often used for chatbots and generated Q&A.
- Babbage – The Babbage model is best used for tasks such as moderate classification and semantic search classification.
- Ada – The Ada model is best used for tasks that don’t require too much nuance, such as parsing text, address correction, and simple classification tasks.
Here’s a very simple example use case: I’ve downloaded reviews from yelp and put them in the test set directory. I run the AI as a sentiment analyzer where it will classify each post based on language used. The sentiment analyzer would know which words in the review indicate positive or negative sentiment. From there, we can then use the tool to create responses based on sentiment. For positive, we can feel confident that it can thank and praise. For negative sentiment we may want to have a customer service rep actually step in and look into the issue further.
You can build this into your own toolset with API integration to nearly automate positive feedback, while delegating negative feedback responses to a human. You can even enlist the AI to help write the feedback to the negative feedback, however these cases tend to be better handled by a human as you may need to research or follow up with the reviewer.
Most of you probably want to know how can you use GPT-3 to help write blogs?
- Help come up with ideas for blogs
- Research for blogs
- Draft and revise blogs
- Write catchy headlines for your blog posts.
- Write engaging introductions to your blog posts.
- Write thought-provoking conclusions to your blog posts.
- Help simplify or summarize ideas into easier to read sentences.
You can use GPT-3 for many uses, such as topic ideation, which is what I tend to use it for the most. If you have a certain subject or topic that you want to cover, but don’t know how many angles or ideas there are to talk about, GPT-3 can assist in this area creating lists from main topics based on it’s knowledge or data that you provide.
Let’s try a quick example:
What are 5 things I can buy my dog for Valentine’s day?
1. A new toy
2. A new bed
3. A new blanket
4. A new collar
5. A new leash
Again, this is a quick easy example of a great starting point. I asked them to expand on the above:
1. A new toy – Dogs love getting new toys, especially if they’re interactive. There are a ton of great options out there, so find one that your pup will love.
2. A new bed – A comfortable bed is a must-have for any dog. If your pup’s current bed is looking a little worse for wear, consider buying them a new one for Valentine’s Day.
3. A new blanket – Not only do dogs love to curl up on soft blankets, they can also help keep them warm in the winter months. Give your furry friend a snuggly blanket to keep them comfy all winter long.
4. A new collar – Even if your dog’s collar is in good condition, it never hurts to have a spare around. Plus, collars can be a great way to show off your pup’s personality!
5. A new leash – Like collars, leashes come in many different styles and colors, making it easy to find one that suits your pup’s personality perfectly. Plus, a new leash is always a great way to celebrate any special occasion!
You’ll quickly see just how capable the AI is, and I hope your gears are turning. It’s taking all the ideation phase that takes so much time and generating it in seconds. The list above is exact, unedited.
AI Content Examples:
Other AI tools help model the interface and allow you to run commands. An example being requesting a list of ideas from a general topic. I’ll go ahead and do an actual marketing example. This particular example was run using the Jasper.ai 3rd party tool:
What are the top 5 SEO ranking factors?
The top 5 “most important” ranking factors for SEO are:
#5. The # of backlinks pointing to your site.
#4. The amount of time spent on your page by visitors
#3. Social media shares & likes, especially from Google+
#2. Site speed (how fast the pages download)
#1. The # of people who visit your site daily
Let’s dive into what we see here. Note – these 5 are unedited and actual outputs.
Let’s start from the top
#5 – “The # of backlinks pointing to your site.”
I think we can all agree backlinks are a major component to ranking a site. I’ll give that an A. It doesn’t go into the types of backlinks, but this would be acceptable for anyone to give as an answer.
#4. The amount of time spent on your page by visitors
So, time on site. Even though Google doesn’t use any Google Analytics data, I think we can all agree that if people spend no time on your site, we’re doing something wrong. Though time on site isn’t exactly a ranking factor, I think it’s a byproduct of doing all the other things right. I’ll give it a pass since this is a pretty common thing to say. This is often confused with dwell time, which is time spent on the site after coming back to the result page and trying a different result. This is something that Google has somewhat confirmed, though it may be more tied to rankbrain as Google tries to understand if the result actually aligns with the particular sentiment of that searcher.
#3. Social media shares & likes, especially from Google+
Here is where we start seeing the issues with using a tool like GPT-3. GPT-3 was trained with data from CommonCrawl, WebText, Wikipedia, and a corpus of books.
The base training was completed at the end of 2019. That means it doesn’t know about current events from January 2020 and beyond.
It also thinks that Google+ still exists. It’s an area where you have to pay careful attention when talking about recent events or things that have shifted including trends. The AI tools might have what we would think would be outdated information.
#2. Site speed (how fast the pages download)
This is really a good one, but similar to the previous one, it won’t have any understanding of core web vitals as a ranking factor as this wasn’t really a thing in 2019. When I asked the AI what are core web vitals, it said “There is no one definitive answer to this question. However, some of the most important web vitals that should be considered include website speed, website security, and website uptime.” It gets the general idea, but it doesn’t know the defined core web vitals according to Google as they didn’t exist in that form in 2019.
#1. The # of people who visit your site daily
This isn’t really a ranking factor, but more a byproduct of good rank. You can see how it uses data from the web, and so it can think that things that are myths or common misconceptions might be fact, and can’t always tell the difference between them.
Overall it came up with some good ideas and some that showed its age. It also wasn’t especially accurate which is something that you’ll find with the AI, it will just make up facts and figures. You need to fact check everything that is writes.
Now this data was generated using the Jasper AI tool.
I ran this same query directly on the OpenAI playground API getting the raw results with no modeling from 3rd parties. Here is how it responded:
My prompt: “What are the top 5 ranking factors for seo?”
The top 5 ranking factors for SEO are:
1. Content
2. Links
3. RankBrain
4. Technical SEO
5. User experience
I would say if I asked that to someone in an interview and that’s what they came back with, I’d be pretty happy.
That was 100% direct from the API with no copy editing.
You can see that even how you interface with the API and adjust your way of asking questions, you can get a much different type of response.
So, the big question is, can Google detect AI content?
In general for content, in a study recently done, judges barely achieved above-chance accuracy (52%) at correctly classifying GPT-3 texts as human or AI written. This basically means that people had as good a chance of just guessing or flipping a coin than correctly guessing.
There has been some work that researchers have done to identify AI written content, and most specifically in scientific journals. They’re mostly looking for oddly phrased sentences or content that seems out of place. This can definitely happen, and why you can’t just drop a topic and let the AI build an entire article.
Now, on to what you probably care most about – does the content perform?
My answer is yes, very much so. But also realize that I’m not using 100% content from the AI, it’s definitely been edited and at the least, it has been nudged into a direction that I’m looking for. I act much more like and editor and the AI is my copywriter.
In my tests, I utilized AI content to write a few short story type blog posts. I published three of these short stories, and saw visibility within the index within a week, and impression data has continued to hold steady. One of the stories is currently ranked position 4, and has been in that spot for several months. The search volume is about 30 searches per month, but the page has stayed consistent.
I also used AI content to augment other pages with much higher volume. The AI content had about 75% original content, with about 25% of it being copy edited. Those pages currently rank in the top 4 spots across about 25 pages, garnering about 10k unique visits per month with an average increase in traffic in the double digits.
My tests of completely AI generated content did show the ability to rank for long tail keywords, while my 75% generated content showed the ability to rank #1 for competitive terms.
Does Google care if you use AI generated content?
Here’s a link to their advanced SEO page on automatically generated content, which is about the closest documentation I’ve seen on AI content. https://developers.google.com/search/docs/advanced/guidelines/auto-gen-content
Let’s look at each of their cases that can cause manual actions on your content:
- Text that makes no sense to the reader but which may contain search keywords.
- This is often from spin sites or where the content doesn’t flow naturally at all, but contains a lot of the target keywords. When you come across nonsensical content, this is typically this form.
- Text translated by an automated tool without human review or curation before publishing.
- Large blocks of automatic language translations just don’t quite work. “All your base are belong to us” is now a meme, but this was a 1989 translation from Japanese to English which for any native speaker it would be very jarring to read or say. It should have been something along the lines of “The enemy has taken all of your bases.”
- Text generated through automated processes, such as Markov chains.
- This is a bit closer to what we have going on with GPT-3, but it does differ. I’m not going to get into the technicalities, but I’ll try to explain as simply as possible what a Markov chain is, and what types of outputs we see. First, a Markov chain is trained with a corpus, which can be as little as a sentence, or as much as wikipedia. It then uses modeling to determine the most common next word after a previous word, and has variables allowing for more randomness. You can also set the how many words should use the prediction versus if they were already modeled. It’s very similar to how GPT-3 works, with the main difference being the neural network used in GPT-3 better understands the content and context instead of using only predictive models.
While I was at it, I asked GPT-3 if they were a Markov chain – and here’s the prompt and response from the AI tool:
Is GPT-3 or GPT-2 a Markov chain?
GPT-3 is not a Markov chain. GPT-2 is a Markov chain.
Why?
GPT-3 is not a Markov chain because the states are not independent of each other. GPT-2 is a Markov chain because the states are independent of each other.
- Text generated using automated synonymizing or obfuscation techniques.
- This is your standard spinning, where your script takes certain words and uses a dictionary or thesaurus to lookup similar meanings to change the sentence enough to be unique. Often used to duplicate another high ranking piece of content but make it seem unique.
- Text generated from scraping Atom/RSS feeds or search results.
- This is a straight copy paste, but is effective to pull content from sites like WordPress which have full blog posts in their feeds. This is exact copy content and should be pretty obvious that Google would not see this as different or better content. Consider however, that a proper combination of pulling or scraping content and getting it indexed first is a kind of indexing arbitrage, where your stolen content may look newer to Google only because you submitted it to them first.
- Stitching or combining content from different web pages without adding sufficient value.
- This is probably the most common form, where content is taken from multiple sources. You see this often on sites like a medical or dental site, where content from webmd, wikipedia, and other local sites are copied and pasted to act as if they are their own with little actual content changes. Plagiarizing from multiple sources is not looked at favorably according to Google.
https://youtu.be/kPVNl7btFHQ?t=3186 – Google’s Office hours where it’s discussed. Summarize what is said, if the content is well written and provides value, does it really matter how it was written? There’s a point at which you can’t distinguish between well written content sources, so the ability to differentiate and whether they would is still a little ways out.
So in summary, I feel that AI generated content has the ability to create quality content, but you need to fact check, and help direct the output. The content itself also needs copy editing, so don’t expect to use it 100% straight from the tools if you want the best results.
If you aren’t thinking about using some AI content generation in your workflow, then go ahead and listen to the following pieces of content that were generated,
I’ll be giving the prompt I used into GPT-3, and its exact, unedited output. Let me know what you think – is the API based content good enough to be used?
If you are listening, you may want to go to opinionatedseo.com/s2/e21 and read them for yourself. I’ll have a few more examples on there that I won’t read aloud here.
Okay, here we go:
Can an AI write really good content?
It is difficult to say for certain, as there is no one definitive answer to this question. Some believe that artificial intelligence can be used to generate high-quality content, while others are not so sure. Ultimately, it depends on the specific AI and the type of content that is being created.
Is AI generated content safe for use in ranking on Google?
There is no definitive answer to this question as it largely depends on the specific nature of the AI-generated content. However, in general, content that is generated by AI is likely to be safe for use in ranking on Google if it is high quality and provides value to users.
Can Google detect AI written content?
Google is not able to currently detect AI content because the technology is not yet sophisticated enough. However, as AI technology continues to develop, it is likely that Google will be able to detect AI content more accurately.
What is the best seo tool out there?
There is no single “best” SEO tool, as the most effective tools depends on the individual’s goals and needs. Some popular SEO tools include Moz, Ahrefs, and SEMrush.
What is the most important thing a website needs to rank well in Google?
The most important thing a website needs to rank well in Google is high-quality content.
I then wanted more content, so I updated the prompt:
Write a paragraph about the most important thing a website needs to rank well in Google?
The most important thing a website needs to rank well in Google is high-quality content. This means the website needs to have well-written, informative, and interesting articles that people will want to read. Additionally, the website needs to be updated regularly with new content to keep people coming back.
How can I convince my boss to increase our SEO budget?
The first step is to make a strong business case for SEO. You can do this by demonstrating how SEO can help the company achieve its goals. You can also show how much traffic and revenue the company can expect to gain from SEO.
If your boss is still not convinced, you can offer to do some of the work yourself. This will help prove that SEO is worth the investment.
Can you tell me 5 things I need to create great SEO content?
1. A catchy headline
2. Engaging introduction
3. Well-researched content
4. Concise and well-organized content
5. Useful and relevant images
Can you expand on each of the above items?
1. A catchy headline – The headline is one of the most important aspects of your content, as it is the first thing that readers will see. Make sure that your headline is catchy and enticing, and accurately represents the content of your article.
2. Engaging introduction – The introduction is also important, as it sets the tone for the rest of the article. Make sure that your introduction is engaging and draws the reader in.
3. Well-researched content – Good SEO content is well-researched and provides accurate information. Make sure that your content is well-researched and fact-checked.
4. Concise and well-organized content – SEO content should be concise and well-organized. Make sure that your points are easy to follow and that your article is easy to read.
5. Useful and relevant images – Images are an important part of SEO content, as they help to break up the text and make the article more visually appealing. Make sure that your images are relevant to the topic of your article and that they add value to your content.
What are the best resources to learn about writing good SEO content?
There are a number of great resources available for learning about SEO content. Some of the best resources include:
-SEOmoz: This is a great resource for learning all about SEO. They have an extensive guide to writing good SEO content, as well as a wealth of other information on SEO.
-Search Engine Journal: This website is dedicated to helping people learn about SEO. They have a number of articles on writing good SEO content, as well as other helpful information.
-Copyblogger: Copyblogger is a website dedicated to helping people create effective content. They have a number of articles on writing good SEO content, as well as other helpful information.
Can you write a few sentences on what a marketer should be good at?
A marketer should be good at creating effective campaigns, building relationships with customers, and analyzing data. They should also be able to think creatively and be able to understand and communicate with customers.
Can you give me a list of the top 10 attributes of a good marketer?
1. Good marketers are creative and innovative.
2. They are good at developing relationships with customers.
3. They are good at analyzing data.
4. They are good at thinking critically.
5. They are good at problem solving.
6. They are good at communicating with customers.
7. They have a good understanding of the market.
8. They are good at marketing research.
9. They are good at branding.
10. They are good at digital marketing.
Can you expand on each of these?
1. Creative and innovative – Good marketers are creative and innovative, and they are always looking for new and better ways to reach their customers.
2. Relationship builders – Good marketers are good at developing relationships with customers. They understand the importance of customer loyalty and they work hard to build lasting relationships.
3. Analytical – Good marketers are good at analyzing data. They understand the importance of data-driven decisions and they use data to make informed decisions about their campaigns.
4. Critical thinkers – Good marketers are good at thinking critically. They are able to see the big picture and they are not afraid to question the status quo.
5. Problem solvers – Good marketers are good at problem solving. They understand that every campaign has its own set of challenges and they are able to find creative solutions to any problem.
6. Good communicators – Good marketers are good at communicating with customers. They understand the importance of clear and concise communication and they are able to communicate with customers in a way that is meaningful and relevant.
7. Knowledgeable – Good marketers are knowledgeable about the market. They understand the trends and they are aware of the changes that are happening in the industry.
8. Research oriented – Good marketers are research oriented. They understand the importance of research and they are willing to put in the time and effort to conduct accurate and reliable research.
9. Brand builders – Good marketers are good at branding. They understand the importance of a strong brand and they know how to build and maintain a strong brand.
10. Digital marketers – Good marketers are good at digital marketing. They understand the importance of a strong digital presence and they know how to use digital channels to reach customers.
Now let’s say I wanted to continue to expand on each of these, I did a follow up prompt:
How can I become more Creative and innovative as a marketer?
One way to become more creative and innovative as a marketer is to read more. Reading helps to expand your knowledge and it helps you to come up with new and innovative ideas. You can also attend workshops and seminars to learn more about creativity and innovation. Additionally, you can try to be more open-minded and you can challenge yourself to come up with new and different ideas.
And then tried a different way: How does one become more creative?
Some ways to become more creative include: being open-minded, challenging yourself, being curious, and practicing mindfulness. Additionally, you can read books and articles about creativity, attend workshops and seminars, and try to be more innovative in your thinking.
How can I build better relationships with my customers? Name 5 ways:
1. Be responsive – Be responsive to customer inquiries and be willing to help them with whatever they need.
2. Be helpful – Be helpful and provide valuable information to your customers.
3. Be social – Be social and engage with your customers on social media.
4. Listen to them – Listen to your customers and understand their needs.
5. Treat them well – Treat your customers well and make them feel valued.
Name three reasons why AI content is a good way to build your blogs.
1. It can help you generate more content ideas.
2. It can help you automate your content creation process.
3. It can help you improve your blog’s overall quality and consistency.
Please expand on each of these points
1. AI content can help you generate more content ideas.
There are a number of AI-powered content generators out there that can help you come up with new content ideas. For example, the AI-powered content platform XYZ can help you come up with topic ideas, headlines, and even full articles.
2. AI content can help you automate your content creation process.
Many AI-powered content creation tools allow you to automate the creation process, which can save you a lot of time and effort. For example, the AI-powered content creation tool XYZ can help you automatically create summaries of articles, which you can then use as the basis for your own blog posts.
3. AI content can help you improve your blog’s overall quality and consistency.
AI-powered content creation tools can help you improve the quality and consistency of your blog content. For example, the AI-powered content creation tool XYZ can help you automatically review and improve your blog posts for grammar, spelling, and style.
Note – actual company names replaced with XYZ as they weren’t accurate.
This is an important detail, maybe at the time of the training, those companies existed and provided those services, but I couldn’t find them or they had changed names to the point where they were no longer able to be found.
And I thought this would be a great way to end this Podcast:
Is SEO dead?
No, SEO is not dead.
Can you tell me why?
There are many reasons why SEO is not dead. Firstly, search engine algorithms are constantly changing, so SEO techniques that may have worked in the past may not work today. However, if you are able to keep up with the latest changes and adapt your SEO strategy accordingly, you can still achieve good results. Secondly, SEO is becoming increasingly more competitive, so it takes more effort and dedication to achieve good rankings. However, with the right tools and knowledge, it is still possible to rank your website high in the search engines. Lastly, search engines are still one of the most popular sources of online traffic, so investing in SEO is still a good way to reach your target audience.
I wrote a blog post and podcast about using AI generated content for SEO. It’s full of examples and some reasons why it can be helpful in a content marketer’s workflow. Can you write a few sentences to summarize the article and end the post?
Including AI generated content as part of your SEO strategy can help you to achieve your desired results. By automating the content creation process, you can save time and resources while still delivering high-quality content that is relevant to your audience. Additionally, AI-generated content can help to improve your website’s ranking on search engines.
And can you say goodbye to our listeners?
Thanks for listening!
About 75% of the content on today’s podcast was generated with AI. The opinion portions were my own, but much of the definitions, summaries and even the opening and closing portions were generated from AI.
I hope this gives you something to think about, and feel free to check out some of the resources that I’ll link on the website.