I am not a religious person. I did, however, feel a bit guilty when I ‘wrote’ my last blog post for this site.
I say ‘wrote’ as that whole post was 95% written by the great AI machine in the sky. Whilst I shuddered putting my name to it, I wanted to to do a test. Today, I want to share the findings from that test.
Why use AI to write a blog post?
The world is awash with AI hype and we are constantly being bombarded with promises of how the rise of the machine will spell the death of humanity. I have always clung on to the hope that grey cells will ultimately triumph and maintained that AI is amazing, but that it is just a tool. In the right hands, it is an incredibly powerful tool but the real power lies in the human that is using the tool.
It would be irresponsible for us, as an agency, not to keep abreast of developments within the artificial intelligence space. If we can be more efficient, we should absolutely embrace modern technology. Not to cut corners, but to be able to offer our clients the best possible value and focus our time on the creativity that will ultimately dictate the success of any campaign.
I wanted to test just how good AI content has got and whether it is actually a viable means of generating web traffic with minimal effort. Ouch – I just shivered writing that as I don’t like the premise of doing as little as possible, but it is important to know what we are up against as I know that a lot of marketeers rely heavily on their AI friends to do their jobs for them.
Ultimately, it was a test to see how a lazy blog post would perform.
How hard was it to use AI to create the blog post?
There are lots of options available to you if you want to outsource content creation to the robots. Chat GPT is probably the best known, but I actually used a service called seowriting.ai as I thought that it would be good to test one of the options that promises a particular focus on creating content that is designed to perform well in the search engines. I was less interested in prosaic beauty than I was raw SEO performance.
I have to admit that the process was very easy. There are a few options to choose from but I went peak lazy and used the ‘1-click blog post’ option to create the blog post. You are walked through a fairly intuitive process where you need to clarify the focus on the post and stipulate various parameters such as keyword targets, themes, length, tone of voice and whether you want particular features such as lists / tables / FAQs / summaries.
I dutifully completed the form and hit the go button. It was probably a grand total of around 4mins work?
Impressions of the AI content
I hate to admit it, but my initial reaction was broadly along the lines of, “Uh oh. Houston, we have a problem” and I started to think about what my next career could possibly be. The post looked really good and was on topic and had some interesting insight, backed up with some nice touches such as feature tables. It was horribly easy to create something that would have taken several hours to research, consider and craft a meaningful piece of content.
Once I read it properly, my initial enthusiasm was tempered. I read it again and breathed a sigh of relief, safe in the knowledge that I wasn’t actually facing the spectre of a P45 form.
It was far from terrible, but it definitely wasn’t perfect. Some of the issues I had with it were plain errors, such as Americanisations despite stipulating that I wanted UK English and it was for a UK audience. It was therefore disappointing to see so many ‘z’s when I wanted ‘s’s. That alone took a few minutes to undo – I wasn’t prepared to publish it as it was, as I think it would have been immediately obvious that I had cheated and used a tool to create the post.
Another issue was that of repetition. The post said the same thing in several different ways. It therefore felt a bit lost and waffly. To be fair, I suspect that this is a direct result of trying to hit the word count target that I had set. Whereas a human would realise that there is only so much that you can actually say about something before you start to repeat yourself, the AI monster simply focused on the word count target and then set about churning out different ways of saying the same thing.
It is hard to explain, but the post just *felt* as though it had been written by a machine. Or perhaps a poor copywriter. I had to remove quite a few chunks of the post as they added nothing other than word count and I wanted to avoid tarnishing our blog with a verbose post that lacked any real sense of direction.
Perhaps that is a little harsh, but I spent a good 15mins tidying it up before I was prepared to publish it. The 4mins therefore turned into nearer 20mins, but that is still very quick to produce something of that length and I wanted to see how it actually performed.
How did the blog post perform?
It is obviously very unfair to use this one post as a test to determine whether AI content can perform well in terms of SEO. Not only is AI a fiercely competitive space, so it was always going to be a challenge, I know that I could have done a better job of tweaking the settings on the initial post creation and could have spent longer going through the content with the SEO tooth comb. I wanted to see how well you can do with the minimum of effort though, so didn’t really do much beyond a very quick tidy up.
A quick dip into Google Search Console shows that the post did attract some attention:
Whilst this is far from setting the world on fire, the raw facts (according to the GSC Oracle) are that the post was shown 1,690 times in the last 28 days and was returned in search results for a total of 227 different keywords. Individual rankings varied across different keywords but there were plenty of immediate page 1 rankings for the less competitive terms.
It is interesting, however, so see that there is an immediate spike but then a fairly quick drop off. Some of that will be News related visibility, but I think it is fair to say that improvements in visibility are fairly short lived. That said, it seems to have settled at a fairly steady stream following the initial jump. If that continues, the longer term value of the content will obviously grow.
Looking at GA4 data for that page, we can drill a bit deeper into actual value of that post:
Again, I don’t think this is going to get anyone wildly excited, but the fact of the matter is that 20mins of effort delivered 87 users to the site and engagement metrics with the post are actually far from terrible. Interestingly, universitybusiness.com obviously thought that it was good enough to warrant linking to the post. I cannot find the link on the site thanks to a paywall, but it should have helped with the domain authority challenge that we all face, so that alone is worth something?
I am not generally a great fan of assigning a monetary value to organic search traffic, but some of the keywords that the post was ranking for would be in the region of £3 per click, according to Google Ads data. You could therefore argue that the 100 views (since publication – this will grow) is worth around £300.
Crucially, the post did not lead to any conversions. We track form submissions and all forms of contact as conversions, but the post has so far failed to deliver any conversions. Brand exposure – yes. Potential customers – no.
Conclusion
I wasn’t expected any earth shattering revelations from this exercise and it is only really intended as a bit of fun. With more time, I would like to do a human v. machine fight to see which performs better. That would be interesting, although hard to do effectively on the same domain.
I will admit that I was impressed by the ease at which you can create passable content. As with any tool, the quality of the output is governed by the combination of the tool itself and its operator. Whilst the ‘1-click blog post’ option that I used was very easy, I am sure that I would have got much better results if I had spent more time configuring what I wanted from the post. As I mentioned, the first impressions are good but closer scrutiny leads to disappointment. The content was not terrible, but had some fatal flaws.
AI content machines are a dream come true for those who are lazy or who really struggle with the written word, but I still do not believe that they rival a good brain.
This limited test also shows that it is not an SEO elixir – whilst it did drive some traffic to our site, the AI optimised content did not perform any better than the content that we write as humans on this blog. Crucially, I have no doubt that the search engines will be increasingly adept at identifying AI generated content and will ultimately reward human crafted content (assuming that it is of exceptional content). I personally think that you would have to be very brave to use Bard for generating content as it would obviously be very easy for Google to join the dots and I would be very surprised if that performed well.
We will continue to explore what AI has to offer. I think it is very good at creating a ‘starter for ten’ as it definitely can help reduce research time and create a workable outline to any content. It is, however, just a tool and true art will require the grey cells to use that tool.
