A message from our MD, Jonathan Guy.
Over recent months we have received more and more enquiries from customers asking if we either use AI to write the content we produce, or if we can use it as it would allow them to generate a lot more content very quickly.
Every time we have to explain that we use real humans to write content and to date, other than an ongoing experiment, we are not using it for our monthly output.
The temptation, of course, is always there. As the fourth biggest expenditure line on our P&L the possibility of finding extraordinary savings by switching to AI is clear.
But money is not the driver here. We’re more focused on getting the right result for every customer.
So, why do we simply not throw everything into some AI model and present it back to you, our customers, as something you can put on your website?
Originality
To understand this part, you need a simple understanding of how AI models work. I’m going to oversimplify to make the point here, but when you ask an AI a question it relies on two things to answer; its training database (what it already knows) and what it can find on the internet.
If we assume that the training database contains things that are already known, as does the internet, all the AI can do is to take what already exists and reword it. In academia it would be plagiarism; taking work that already exists and presenting it as your own original thought. Of course, the AIs don’t do that as they cite their sources, but this brings me back to the original point. It only knows what already exists. Moreover, it assumes that what already exists is correct.
This is not the ‘added value’ content that Google and other search engines are looking for.
Accuracy
Recently we created an article for a customer on a deeply technical aspect of their business. The article was good and was well researched by our copywriter. They cited multiple sources and referenced the correct aspects that needed referencing, but the problem was that it was factually incorrect. It relied on examples from the American market (which make it incorrect in the U.K.) and part of what had been written was out of date.
Our copywriter had done a good job, but it transpired that the original material they had sourced from was AI generated. It was incorrect to start with. This means that not only is AI giving the wrong answer to everyone who would ever ask about this topic, but people who are researching new articles are also being given incorrect information.
Someone had used AI to create a long deep article about the topic and it was inaccurate and was published unchecked (see scale below). What comes after that are a cascade of further inaccuracies which can persist until someone who understands the topic casts a critical eye over the output.
Tone of Voice
What is AIs tone of voice?
This might be a strange question, but it cuts to the heart of the matter.
Why do people buy books from their favourite author or buy newspapers to read a particular journalists view? Why, in some cases, will they queue to get the latest book in the series on the day it is released?
In some instances, it’s because of the unique world the author has created to which the reader can be transported. But put that world in the hands of another author and the book is not the same. In the case or newspaper columnists it’s because you know that they write with a tone of voice that speaks to you.
Tone of voice is such an important part of writing and can be the difference between people reading an article or book and abandoning it after just a few seconds.
Case in point; this article is entirely written by me, and no AI was used in its construction. I’ve not used the ‘autocorrect’ suggestions that the grammar police built into this software. I’m hoping you can tell the difference.
Trust
Google has been very clear about what it wants to see on a company website before it will rank it. Apart from a decent, easy to follow structure, speed of load and valuable content, it wants to see signals that mean search engines, and by default, therefore, humans, can trust what they see.
Building trust isn’t easy. Meeting someone in person is a great way to build trust but that’s not possible for a website. Trust can be conveyed by the words that are written (trust me, I’m great, honest!) or by images that give you a good feeling, but most of the trust signals that humans use are not simple written or visual cues.
Google lists a range of elements it, and by default humans look for. These are known as the E-E-A-T factors; Experience, Expertise, Authority and Trustworthiness.
These range from something as simple and subtle as the correct company name and address at the bottom of the website to how times you as the author, or your company, have featured in the news.
Google and other search engines have a role to play in ensuring that what you read in their search results is factually correct, reliable and not misleading.
Using AI to write your content does not sit easily with these and AI can be very bad at providing this level of reassurance. Not only can it often be inaccurate (see Accuracy above), but it also lacks the authenticity of someone who knows their craft or trade inside out.
And despite the ability to prompt AI to have a specific tone of voice, it often reads more like it has been written by a machine than a person, which does not engender trust.
Focus
If you are business owner writing an article for SEO, the chances are you will try writing something that focuses on a single keyword.
This makes sense, as if you’re trying to rank for the work ‘dog food you’re going to want to write about dog food. You may even be tempted to include the words ‘dog food’ lots of times in the article as that’s what SEO is, right?
The problem is that if you ask AI to do it your prompt will ultimately deliver what you ask, and that may not be a good thing, as SEO is not about how many times you can get the keyword on the page.
How often have we seen a prompt like ‘write me a 1500-word article about dog food that will rank in SEO’? Too many, unfortunately.
Something like this will make sure it mentions your special keyword a certain number of times (probably based on a number that someone has authoritatively asserted on their website that is the golden number for ranking in search) but in reality, will be just more AI generated slop that clutters up the internet.
What it won’t do it think tangentially, think along parallels or think more broadly about what you are trying to achieve. It also, unless prompted, won’t assess the piece in the context of what already exists on your website.
In short, the more you focus it on a keyword the less likely it will be to be worthwhile and valuable.
Scale
AI can produce prodigious amounts of content. We’ve seen it generate over 60,000 words a month for a website and that is a problem.
How do you proof 60,000 words?
How do you make sure that every word in every article is correct, valuable, and doesn’t mix up facts? How can you quality check that much work?
And what if you find a sentence you don’t like? Do you just edit it, delete it or rewrite it? If the latter, where do you stop? Why not rewrite the whole thing?
Scale is genuinely a problem. Ask any writer to create at scale and you’ll find the output gets worse, not better. Currently, AI is no different.
Hallucination
AI can make things up. We’ve seen it far too frequently and despite vast improvements there’s still no guarantee that what it churns out will be right, especially in very niche areas.
From our own experience, niche websites we run, and have carefully curated every word over many years, are still used by AI to make up answers. They reference the website but incorrectly attribute words that have been written or even take a leap and assume something that is completely untrue.
In short, the more niche your area the more likely AI is to create something it thinks you want to read instead of what is factually correct.
Summary
There’s a lot more reasons why AI shouldn’t be used for this type of work, (when I asked AI it gave me a list of 20 reasons. Not 19 or 21, but 20. This feels like it is trying to please me…) and pleasing the person writing the prompt is just one of them.
As an Agency owner there are always temptations to take short cuts. These have existed since I started the agency, when the easy option as to buy backlinks from a link farm in Singapore or Ukraine (we never did this by the way). The reason we didn’t was first, it never sat easily with me as the right thing to do and secondly, I always felt it would be something that would harm a website in the long term.
One year later I was proved correct.
Today, the temptation is to ditch our seasoned and experienced copywriters but as you can tell from this article, I’m not going to do it.
I’m not asserting that AI will never be able to write content for a website. The experiment we are currently running on AI content means that we are testing and validating the output just in case it can.
As a business owner, and one of our customers, I hope you appreciate that we are trying to do the very best job we can for you and your business. We don’t do anything to put your website in harm’s way or do anything that is against Google’s Webmaster guidelines.
More importantly, we want nothing but the best for you.
So, for now, we’re sticking with our copywriters as in the world of AI slop, these people are possibly the most valuable resource we have.
Jonathan Guy
Managing Director
