Editors Note (03.02.23): Our original ChatGPT SEO guide created on 15th December 2022 explored how successful SEO content could be created using AI. The latest update provides a fresh 2023 view of this and discusses the impact of using ChatGPT to provide a content refresh. We have also covered the latest AI content detection tools available such as GPTZero and the new AI Classifier tool from Open AI. Check our sections 9 and 10 for more info on these updates.
Introducing AI for SEO: Why we’ve created this guide
Table of Contents
In this post we will explore the following topics:
- 1 Introducing AI for SEO: Why we’ve created this guide
- 2 A bit of background information about us
- 3 Overview of OpenAI GPT-3 and its potential for SEO and content creation
- 4 Why did we want to test yet another “AI tool”?
- 5 Preparing to use AI for SEO – what steps did we take?
- 6 Implementation – a comprehensive guide covering how we used OpenAI GPT-3
- 6.1 Step 1- identifying the keywords
- 6.2 Step 2 – let the AI brainstorm potential titles for the content
- 6.3 Step 3 – use the AI to up with an outline/structure for the content including headings
- 6.4 Step 4 – create the all-important intro
- 6.5 Step 5 – create a table of contents
- 6.6 Step 6 – create and revise the content for each heading
- 6.7 Step 7 – use ChatGPT SEO prompts to suggest the URL, meta data and HTML
- 6.8 Step 8 – add formatting including bold, bullet listed, hyperlinks and revise keyword usage
- 6.9 Step 9 – create a featured image using AI (DALL-E)
- 6.10 Step 10 – let ChatGPT create an article in full
- 7 The findings – our test results revealed
- 7.1 Does the content suffer from duplication? – No
- 7.2 Does the content rank well in Google? – yes
- 7.3 Do the articles generate real traffic? – yes
- 7.4 Will people read and engage with the content? – yes
- 7.5 Could AI replace content creation and SEO altogether? – maybe
- 7.6 Is AI a useful assistant or a hindrance to SEO? – assistant
- 7.7 If we deemed it was useful, would it save time, improve quality, etc.? – yes
- 7.8 Which article wins?
- 8 Does ChatGPT only do text?
- 9 Can SEO be enhanced with a ChatGPT content refresh? (Jan 2023 update)
- 10 Can Google detect ChatGPT and does it matter? (Feb 2023 update)
- 11 Final Thoughts – a review of the benefits of using AI for SEO
The use of AI within marketing has been around for some time now but what about using AI for SEO? With AI-driven tools to generate leads, optimise campaigns and personalise customer experiences, AI has been invaluable in helping marketers to identify marketing trends and make more informed decisions.
However, this guide isn’t going to focus on any of that – in the advent of OpenAI’s release of ChatGPT (https://chat.openai.com/chat) on November 30 2022, we decided to put ChatGPT to the test when using its AI for SEO. To make this experiment fair, the content created was 100% generated by ChatGPT with no human edits whatsoever and we documented the whole process. As a result, we’ve even created our own plugin here. Read on to find out more.
A bit of background information about us
We like to think we’re a bit cutting-edge. Opace were one of the first digital agencies in the Midlands to openly specialise in open source and we were blogging about this back in 2011 along with the applications of social media. In more recent years, we’ve been one of the first agencies on the scene to discuss technologies like VR, the Metaverse and their potential impact on digital marketing. Now here comes AI – we’ve never been more excited as an agency than we are right now.
We’ve now spent around 2 weeks experimenting with ChatGPT SEO, content creation and coding. It’s a game-changer!
Overview of OpenAI GPT-3 and its potential for SEO and content creation
We won’t go into much detail here as the topic has been covered in depth below, take a look – it’s a cracking read:
Actually, the above statement is a bit of a lie. We asked ChatGPT to produce the above article, with various prompts from us along the way – and that is the whole purpose of this guide. We wanted to test the tool to see whether it was any good for SEO and content creation.
Having played around with truly appalling article automation and spinning tools around 10 years back, and more recently ‘so-called’ AI content creation tools, we were ready to give up on the idea of using AI for SEO altogether. Then after experimenting with tools like DALL-E, AI Writer and Elai in our 25th November post covering AI marketing tools, we started to get more hopeful and spent a few days investigating other AI image generation tools like Stable Diffusion.
Then only five days later ChatGPT comes along and changes everything. In the subsequent sections, I’ll explain how we tested the tool and why it’s an overwhelming success.
For more information on the differences between GPT-3 and ChatGPT, please read here.
Why did we want to test yet another “AI tool”?
Quite frankly, it took a few days before we built up the motivation to give it a go.
We started to hear crazy stories about this new AI tool writing essays, horror stories, fictitious dialogues between famous people, and even erotic novels in perfect English.
We then thought ‘hang on, what about SEO’?
Based on what we were hearing, we started discussing questions such as:
- Could AI replace content creation and SEO altogether?
- If not, would it be a useful assistant or a hindrance to SEO?
- If we deemed it was useful, would it save time, improve quality, etc.?
As any digital specialist knows, writing great quality content is at the heart of SEO and digital marketing in general. However, for content to be great, it needs to tick certain boxes.
For example, for SEO content to be ‘great’ it must be:
- Useful: providing useful information to the reader in order to answer a specific need or question
- Relevant: serving information that is relevant to the topic at hand
- Engaging: engaging the reader, with interesting text with visuals, videos, and other forms of media to keep the reader interested
- Unique: delivering value to the reader that can’t be found elsewhere
- Clear: communicating information that is clear, to the point, and easy to follow
- Well-written: conveying the text in a well-written manner, free of grammar and spelling errors
- Actionable: prompting the reader with actionable steps in order to accomplish their goal or solve a problem
- Credible: giving the reader a sense of credibility with accurate facts and figures to back up the content
- Optimised: allowing the reader to find the content in search engines based on relevant keywords and phrases
- Up-to-date: discussing the latest information with up-to-date facts and figures
This is a challenge for even the most experienced human writer, so we didn’t think AI would have a chance.
Before testing ChatGPT, our theory was that it would produce reasonable content but would fail on most of the points above. In particular, we expected that the content would be largely duplicated based on experience with other AI content generators and we were certain that the content wouldn’t be favoured by Google.
Read on to see what happened.
Preparing to use AI for SEO – what steps did we take?
Firstly, the tests needed to be fair and provide data that we could compare somehow.
Establish the experiments
We established two experiments:
- Experiment 1 – Firstly, to test how ChatGPT would perform given human decision-making to inform the AI over the course of around 2-3 hours – the idea here was to utilise our expertise throughout the content creation process, continually attempting to optimise and improve the content
- Experiment 2 – Secondly, to let the AI decide the content entirely where we would input a title and let ChatGPT produce the content) – here we wanted to see what the tool could produce in around 60 seconds or less with virtually zero input from us
Define the rules and principles
We set some very strict rules for both articles:
- The article title must be AI-generated
- ALL text/content must be 100% AI generated with no human alteration
- The structure and formatting must be AI-generated
- The URL, meta data, image ALT tags and even images themselves must be AI-generated
- The only human input must be constrained to the initial idea and keywords provided
Before saying any more, we have been impressed, well more than impressed – we were absolutely shocked by how good the content is! Fort test 1, we threw everything at the tool and it never let us down.
ChatGPT understands intent and context, it has a deep knowledge of facts/information and the written language is exceptional.
Many have even been calling ChatGPT the “Google-killer” due to its accurate and insightful responses.
Determine the hypotheses for evaluating the experiments
Whilst not a strictly fair comparison, the two experiments were intended to prove two hypotheses:
- AI content that has been carefully structured by a human will be of better quality
- AI content that has been optimised with the knowledge and expertise of an SEO specialist will rank better
So let’s now get stuck into the nitty gritty.
Implementation – a comprehensive guide covering how we used OpenAI GPT-3
We thought the most interesting test would be to get ChatGPT to write about itself and cover the topic of AI, text creation and SEO in detail. We could see that the topic had been covered by authorities on the subject as well as OpenAI themselves, so quite honestly, we weren’t expecting a great outcome at the beginning.
These are the steps we followed:
Step 1- identifying the keywords
Google Keyword Planner ideas
In good old-fashioned SEO style, we used Google Keyword Planner to check the search volumes a few obvious keywords:
Stuck for more keyword ideas, test the capabilities of ChatGPT SEO by asking for ideas
We asked ChatGPT to provide 10 keywords based on the ones Google Keyword Planner gave us:
Evaluate search volumes for new keywords
As every SEO person knows, the keyword planning stage is incredibly important. We decided to use the below keywords as a basis for both articles:
- “gpt 3 ai”
- “gpt 3 text generator”
- “openai gpt3”
- “openai chat”
- “ai seo tools”
For a more detailed guide on using GPT-3 and ChatGPT for keyword research and meta data see here.
Step 2 – let the AI brainstorm potential titles for the content
The next step was to ask ChatGPT to brainstorm catchy article tiles using the keywords above.
These were really good but we thought it would be interesting to test the tool more and ask it to revise the titles to use the keywords exactly,
These titles seemed better but we wanted to push the capabilities of ChatGPT SEO by asking for more focus on a particular keyword.
There were some subtle differences as a result of each question but they largely seemed unchanged. We’ve been using ChatGPT a great deal more since this experiment and have learnt that there is only so far you can push the tool when it comes to keywords. This seems to be due to its ability to infer meaning and correlate keywords to be one and the same thing. That in itself has been an eye-opening experience, seeing first-hand how clever this AI is (and most likely, therefore, how clever Google is).
Step 3 – use the AI to up with an outline/structure for the content including headings
Choosing a tile was tough, but we decided on ‘How OpenAI GPT-3 is Enhancing AI Chat and Text Generation for SEO’ for experiment 1 and ‘The Evolution of AI SEO Tools with OpenAI GPT-3’ for experiment 2.
We asked the AI to create headings for a 1000-word article using the keywords, each in its own heading. The first list seemed a bit limited, so we asked ChatGPT to revise the list to include 10 headings with more detail about what the first and second sections should be.
Step 4 – create the all-important intro
Knowing how important the introduction is, we spent some time asking ChatGPT to produce this around the given keywords. We hadn’t really picked out a primary (or focus) keyword, so at this stage, we asked the AI to revise to place more focus on “AI SEO Tools”.
Step 5 – create a table of contents
Even though we have a plugin to insert a table of contents automatically, we asked the AI to produce a TOC so we could see what it came up with.
Step 6 – create and revise the content for each heading
Next, was what we had been waiting for – we asked ChatGPT to create between 200 and 400 words for each TOC heading.
The tool only seemed to manage around 400-500 words at a time, so we broke the work into two parts to create the entire article.
At this stage, we felt that the original headings were better and wanted to drop the TOC-style headings
Still unhappy with how the article looked, we asked the AI to re-write the intro to make it more catchy and include data.
We felt the conclusion was missing something, so we asked ChatGPT to produce a snappy call to action for the conclusion.
Step 7 – use ChatGPT SEO prompts to suggest the URL, meta data and HTML
We really weren’t sure what to expect with these tests, but first, we wanted to test get ChatGPT to recommend a Google-friendly URL.
Given the tool only seemed able to work with around 400-500 words at a time, we thought we would try another interesting experiment. At this stage, we uploaded everything to our WordPress blog and published the article in basic text format.
We then asked the AI to read the URL and provide the HTML code.
In hindsight, we think that ChatGPT probably just remembered the previous text and provided the HTML output based on this. The tool clearly states that it cannot get information from a specific URL and we’ve since put it to the test with little luck.
To see what else was possible, we used carefully worded ChatGPT SEO prompts to create the meta data. We’re only touching the surface of what’s possible here. To get a greater understanding, see our AI meta data guide here.
This is what it came up with:
At this stage, we made a very strange decision and decided to change the primary keyword for something more generic ‘OpenAI GPT-3’. This was probably a mistake but we also wanted the AI generated article to score well with the Rank Math SEO plugin for WordPress. After all, nobody likes seeing red or amber traffic lights when it comes to SEO.
Then we asked the AI to add hyperlinks from five different websites. This was actually quite disappointing, but in hindsight, we should have guessed this wouldn’t be possible.
Initially, we were excited when seeing what appeared to be relevant hyperlinks being added.
IMPORTANT – we learnt afterwards that the AI was actually coming up with fictitious URLs that seemed relevant to the domains added. Strangely, some of the URLs actually worked, which was most likely a fluke, but more on this later.
The next logical step was to add some formatting, so we asked the AI to highlight keywords in bold.
Then a fairly radical overhaul, we asked ChatGPT to revise the entire article for the new primary keyword ‘OpenAI GPT-3″.
Then back to Rank Math to get some validation using by seeing those all-important green traffic lights. Yes, we know – this was a mistake and we shouldn’t have changed the article at this stage just for some green shiny traffic lights. That said, the experiment was still intact – all of the content was still AI-driven and our only input was regarding keywords.
A few more tweaks were needed to the headings based on the keyword change.
The SEO suggestions made by Rank Math are often useful but we don’t always agree with them. We thought we would put ChatGPT to the test again and ask it to make its own SEO suggestions to improve the keyword density.
We thought it would be nice to add some reference sites at the end of the article for further reading, so we asked the AI for suggestions.
Remember that important note we added previously about hyperlinks being fictitious, well here’s the proof – these are all made up by ChatGPT. It become immediately obvious as soon as we saw our own brand “Opace” mentioned on marketingtech.co.uk. Apparently, we’ve launched our own AI text generator, but we knew we hadn’t.
We asked for some additional sites – for below at least, some of these actually worked, which was most likely a fluke.
A bit more formatting was needed as we’re big fans of bullet lists.
If that wasn’t enough, we were keen to add a new supplementary section on SEO. We already had ‘What is OpenAI’ so we feel ‘What is SEO’ was also needed.
Nearing the end of our experiment, we asked Chat GPT to revise the conclusion based on a specific keyword density. This worked but it didn’t quite achieve 3%.
Finally, we asked ChatGPT to suggest a list of tags that we could input into WordPress.
Step 9 – create a featured image using AI (DALL-E)
Given everything else was AI-generated, we weren’t going to resort to finding stock images. This is what we got.
For more information on DALL-E and AI image generation, please see: https://www.opace.co.uk/blog/ai-marketing-tools-for-content-creators
What we thought would be interesting was to get ChatGPT to actually come up with the text to put into DALL-E for a second image, so we asked the AI to suggest a good descriptive sentence to get DALL-E to produce an image about the ethics of using AI in an abstract format.
This was what we got – it’s all personal preference but these seemed far more interesting than the original images.
Step 10 – let ChatGPT create an article in full
This one will be kept short. We decided on the title ‘The Evolution of AI SEO Tools with OpenAI GPT-3’ as this felt like a much better match for the keyword ‘AI SEO Tools’.
We let the AI generate the article structure and headings and then continue to produce the entire article. Once this was produced, we requested meta data and created a featured image using DALL-E. The whole process took a minute or so to complete and the resulting article is below:
The findings – our test results revealed
Remember the initial theories? We’ve summarised these below:
- the content would be largely duplicated based on experience with other AI content generators
- the content wouldn’t be favoured by Google
Does the content suffer from duplication? – No
The first article passed Copyscape with flying colours.
We felt we could live with this. In the spirit of staying true to the experiment and the next test, we didn’t want to make any changes.
Does the content rank well in Google? – yes
This one was probably the biggest surprise to us. Both articles were indexed overnight and ranking in Google the very next day.
Initial tests actually showed both articles ranking for some of the same keywords – and on page 1.
The first one we noticed was “OpenAI SEO”, it seems like a good search term given that this is what we are trying to test. Seeing our AI-generated SEO ranking right next to a detailed article from Search Engine Journal was nice. Sadly, the keyword isn’t showing as having any search volume according to Google Keyword Planner. As most of us know in the SEO industry, there’s no point in being number 1 if nobody is looking.
Now that the articles have been live for a few days, clearer results can be reported. The keywords “GPT-3 for SEO”, “GPT-3 SEO” and similar rank around position 5 on Google. That’s an impressive result but still a relatively small search volume.
Sadly, we’re not ranking on Page 1 for the primary keyword “OpenAI GPT-3” but we’re not far off, ranking at position 14 currently.
For the complete list of ranking keywords, see below:
We feel this is astonishing. Within only 4 days, we’re showing promising rankings for good keywords. We’re only 5 positions away from Page 1 for “OpenAI GPT-3” with 370 UK monthly searches. Crikey, we’re ranking position 24 for “GPT-3” with over 10K monthly UK searches. We never expected to be ranking at all for a keyword like this with 100% generated AI content.
Do the articles generate real traffic? – yes
Since publishing the two experiments only a handful of days ago, they are generating more traffic to our website than any other content within the same period of time.
And what’s more – we’ve not shared the content at all (no social media), we’ve not linked to the content, and we’ve not even changed the US spellings to UK British to avoid modifying the content in any way.
And even more than that – look at the average time people have spent reading the first article, it’s a whopping 10 minutes or more.
Will people read and engage with the content? – yes
The above image alone shows that the content is sticky enough for people to spend 10 minutes or more on the page.
The second AI SEO Tools article doesn’t perform as well but it’s still number 2 on our list according to Google Search Console.
Could AI replace content creation and SEO altogether? – maybe
In our opinion, potentially, yes – but only in terms of the writing.
Probably the most important (and time-consuming) part of creating the first article was coming up with the right keywords, title and structure for the article.
Google Keyword Planner played a role in this and we feel search volume will always need to play a role in deciding keywords. AI generated ideas alone are not enough.
Human decisions were made throughout article 1 to tailor the content and enhance/improve what was being said.
Article 2, which had next to zero human input is ok. It ranks in Google, but obviously not as well.
Our main concern with article 2 and potentially both articles, is with regards to the actual data and facts included. Article 2 made some questionable decisions when claiming things like “One of the leading examples of this type of AI SEO tool is Articoolo”. There are far better tools to cite but we didn’t want to interfere with the process.
This alone shows that human proofreading and data checking will be needed for the foreseeable future.
Is AI a useful assistant or a hindrance to SEO? – assistant
We think this is clear by now – at best it’s close to replacing content writers and at worse it’s an incredible assistant.
If we deemed it was useful, would it save time, improve quality, etc.? – yes
This one is interesting. The first article took so much time to produce that one of our copywriters could have produced their own article in a similar amount of time. However, this was largely due to experimentation with ChatGPT and trying to find the right questions. It’s clear that AI tools like this can make huge time savings in terms of researching the topic.
With some refinement, this process can be reduced to a matter of minutes. In fact, since publishing these two articles, we have come up with a set of pre-defined questions that can be used to create human-directed AI content very quickly. We’ll come back to this in a future update, so watch this space.
To answer the quality element of this question properly would require an article of its own. That said, what we feel was most impressive was the AI’s ability to keep the article on track and focused in terms of answering the question.
A part of the pre-defined questions mentioned above was to get the AI to evaluate its own content. These specific queries are below along with the answers for the first article:
It passes for each question other than UK British spellings, which we expected.
Which article wins?
Article 1 is the clear winner!
Does ChatGPT only do text?
While the purpose of this guide is to cover AI SEO and written content, we can confirm that ChatGPT covers so much more than just written content.
For examples of other applications, WordPress users can check out a wide range of ChatGPT plugins to experiment with creating different types of content.
Can SEO be enhanced with a ChatGPT content refresh? (Jan 2023 update)
A couple of months after our two experiment articles were published, we decided to give our AI SEO tools article (experiment 2) a refresh, but why?
Firstly, we were never happy with the content provided in this article and some of the tools referenced. Without ruining the experiment and re-writing or editing the article ourselves, we wondered what would happen if we used ChatGPT to enhance the article by using different prompts and queries.
Secondly, we know from experience that updating content and keeping it fresh/up-to-date can provide huge ranking improvements.
Interested to find out what happened? It worked.
We asked ChatGPT to provide a brief history of AI SEO tools. Based on its response and inclusion of IBM Watson Analytics, we revised our original prompts to update the article to include different companies that have been significant in the field of AI SEO. We told about a wide selection of players (see below) and they all got a mention:
- Moz Pro
- Google Cloud Platform (GCP) including Google Natural Language, Google Cloud AutoML and Google Cloud AI,
- Amazon Web Services (AWS) and AI–powered tools, such as Amazon Comprehend and Amazon Polly
- Microsoft Azure with its range of AI–powered tools like Azure Cognitive Services
We even vetted the above to make sure these weren’t made up and each one came back as having some role to play in AI and SEO.
The end result was an enriched article but also the ranking improvements we had hoped for. Again, using 100% AI-generated text, we saw a ranking improvement from position 24 (where it settled by 19th December 2022) to position 14 on 10th January 2023 for the keyword “AI SEO Tools”:
Can Google detect ChatGPT and does it matter? (Feb 2023 update)
At this stage, you may be wondering why it matters whether Google can detect ChatGPT. We’ve shown that this AI content can rank well and drive traffic. We’ve also shown that refreshing content using ChatGPT can have a positive impact on SEO and rankings.
With all that said, we’ve been carrying out SEO for a long time and are well aware that Google will penalise plagiarised or low-quality content. Whilst our test articles rank, we honestly don’t know whether Google can detect them as being AI-generated.
In January 2023 we heard about a tool called GPTZERO for detecting AI content. Of course, we were intrigued by this so had a good play with the tool to test its capabilities.
Therefore we decided to carry out phase three of our “using AI for SEO” experiment and put these tools to the test using our two experiment articles.
To see the full results of our experiment using the above tools and the conclusion as to whether Google can detect ChatGPT, read below:
We don’t want to give everything away here but our opinion is that both articles passed the detection tests and were believed to be written by a human.
One of the tools deemed the articles to be “98% or more real”, the other “very unlikely to be very unlikely AI-generated” and only GTPZero mentioned “your text may include parts written by AI”, which came to around 14% of the total text for the longer article and 17% of the total text for the shorter one.
Out of curiosity, we also ran a 100% human-written article through GPTZero, one that we had published on April 28th 2022 long before we even knew about ChatGPT. Guess what, that was deemed to have “sentences likely to be written by AI”, 18% of the overall text, which is more than either of the AI articles.
Final Thoughts – a review of the benefits of using AI for SEO
One of the most impressive outcomes of this experiment is the responses from those we asked to read the articles.
Everybody thought they were either written by a human. Nobody guessed that they were 100% AI-generated. That said, there were some minor points made:
Spellings – one person asked why the article was written using US spellings
Structure – the same person said the length of each section seems a bit rigid and could be varied more
It’s clear that both articles can be improved and would probably rank better with some human edits to the content and regular enhancements, but with the right input, direction and prompts, we now know that is possible to create exceptional content written entirely by the AI.
So, now let’s go back to our original theories below:
- AI content that has been carefully directed by a human will be of better quality
- AI content that has been optimised with the knowledge and expertise of an SEO specialist will rank better
We feel that we can confidently say that our first and second theories are correct. The first article wins on both fronts.
To conclude, tools like OpenAI and ChatGPT are absolute game changers for SEOs, content creators, coders, writers, academics and even creatives. Within the next year, it’s quite possible that things will change significantly as a result of this technology. However, people are normally reluctant to change, especially where that change could impact their jobs and livelihood. Our main fear is that tools like ChatGPT will be seen as a threat and end up becoming regulated or even banned.
We hope that doesn’t happen. Technology and AI will always bring about innovation and change. That gives us an opportunity to adapt and improve the way we work, to become smarter, and more efficient. It shouldn’t be seen as a threat but we know it will cause many people to panic.
Revisiting things in 2023 after undertaking our content refresh and AI detection experiments, we are even more confident that using AI for SEO is the future. Not only that, but it seems within the short period of writing this article and this latest update, outlawing or regulating tools like ChatGPT seems less likely. Instead, the emergence of tools for detecting AI content will empower readers to check the content and decide how they feel about the prospect of AI being involved.
What do you think about using AI for SEO? We would really like to hear other people’s experiences and thoughts so add your comments below. It’s not the end of this article and we will be back to add more.