It is every website operator’s dream. Enter a keyword into a computer and within seconds it will write as many relevant SEO texts as you need. But how advanced is artificial intelligence (AI) really when it comes to writing? Can it produce perfect, search engine optimized texts?
Before we answer the question of how well artificial intelligence can write so-called SEO texts, of course it must be clear what an SEO text really is. This is why we ought to take a quick look into the near past. Search engines could be fooled until around 2011–2012. If a person searched Google for “nice sports shoes”, the page that used this word pair most often was displayed at the top. The texts on the website in question only had to make slight references to nice sports shoes. Plastering the website with these keywords was sufficient. There were also many other ways of positioning a website in the search engine top ten, even if its content was very poor. In the past, SEO texts were therefore articles in which certain keywords appeared particularly frequently.
However, in the course of time, Google and Co. learned to distinguish these spam pages from websites with valuable content. This is why search engine operators advise the use of texts specifically created for the readers and not aimed at search engines. The texts should therefore be informative, entertaining or encourage purchases depending on the respective target group.
In short: A SEO text is a text that addresses the recipient.
It is therefore clear that there is no standard formula for SEO texts. They must always be tailored to the particular application. For instance, an article description ought to provide information but above all it ought to motivate a purchase. It must therefore not be too complex and it should not include technical terms. Ultimately, everyone must be able to understand it. On the other hand, an article aimed at scientists ought to include technical terms. Here, the readers expect an appropriate level of complexity.
Get perfect SEO texts written by humans at clickworker.
Yes, it is. If for example the AI is supposed to create product descriptions, then it must be trained with this type of text. It adopts the structure of the texts and then creates article descriptions at the push of a button. This can of course also be done with scientific articles, travel reports, product reviews or even poems and novels – the text type is irrelevant for artificial intelligence.
However, there is a huge problem: The AI does not understand what it is writing.
(More about what AI is and is not able to do can be found here: „What is AI really capable of? Dispelling eight myths about artificial intelligence“)
As previously mentioned, the AI determines the structure of the texts. The AI then builds new texts based on this analysis. But the AI does not know that one article description is about a T-shirt and the other about a toothbrush. As a result, some sentences may sound strange.
Of course, the training data can be adapted. The AI should, for instance, only analyze article descriptions of clothing to write texts about T-shirts, trousers and the like. In the end, the more precisely the target area is defined in which the artificial intelligence becomes active, the better the AI will write the texts.
That is why AIs can already write excellent weather reports or articles about sports results. Here, the structure is practically always the same. You only need to give the AI weather data or game results and it does the rest on its own.
In fact, language AIs such as Transformer, BERT, GPT-2 and Megatron build texts that look like they were written by a human being. This is achieved by training the AIs with a huge database from which they can retrieve terms, their synonyms and sentence structures.
For example, GPT-2 was provided with a paragraph written by humans. The AI then used it to build a text in the same style and with astonishing results.
Paragraph provided by a human being:
Scientists made a surprising discovery: they found a herd of unicorns living in a remote, previously unexplored valley in the Andes. The researchers were even more surprised by the fact that the unicorns spoke perfect English.
The AI continued as follows:
The scientist named the population after its distinctive horn, Ovid’s unicorn. These four-horned, silvery-white unicorns were previously unknown to science.
Now, after almost two centuries, the mystery of what caused this strange phenomenon has finally been solved.
At first glance, it seems that the AI is in fact writing an article that matches the content of the given paragraph. However, with the use of the term “four-horned” it becomes clear that the AI does not know what the word unicorn means. Even the remark regarding why the mystery was solved after “almost two centuries” would only make sense if the unicorns had been discovered two hundred years ago. But that is not the case according to the paragraph provided.
Later statements by Pérez, the researcher invented by the AI, also raise doubts. He says: “When we reached the peak of a mountain, the water looked blue, with some crystals on the peak.” This makes us wonder, how can water have a peak?
You are sure to notice that texts created with such advanced AI’s can at first glance appear to have real content relevance. Moreover, the scientific style is ideally suited to mislead readers. Papers written by actual scientists are also often difficult to understand and may in parts appear contradictory to the recipient. But we then usually assume that we do not possess the knowledge to understand everything. We seldom doubt the validity of the text. Therefore, when it comes to scientific topics, the AI can easily hide the fact that it does not understand what it is writing about. It only excels at reproducing style (i.e. certain sentence structures and choice of words) and replacing terms with synonyms and incorporating related adjectives and verbs.
Even if AIs can easily fool us on the surface, they lack the understanding of the content that is at stake. However, readers must be attentive in order to detect the deception.
Simple AI game: Do you want to create an AI text yourself? Try out Talk to Transformer. Here, the GPT-2 language model continues a paragraph you have invented yourself.
Training data is generated by humans. Humans make mistakes. These mistakes continue to spread in the AI.
The wrong evaluation of training data leads to fatal errors. Image recognition AIs, for example, were being racist because they were trained only with images of fair-skinned people. (cf. https://www.smithsonianmag.com/smart-news/art-project-exposed-racial-biases-artificial-intelligence-system-180973207/)
Training the AI requires a lot of computing time. Preparing the training data requires a lot of manpower, because the more high-quality data the AI gets as a basis, the better it becomes. The correct evaluation of the training data is an equally time-consuming step.
An AI text can work for simple texts in advertising or product descriptions. But as soon as it contains relevant information, the AI will fail.
The training data determines how the AI writes. This can make the texts boring in the long run. And it causes readers to bounce off quickly, which is a negative ranking signal for Google.
Note: Of course, an artificial intelligence can write texts with different styles. However, it will hardly be able to produce changes of style within, which can make the text exciting. Unless it has been trained with these kinds of changes of styles in the texts. But then this change of style becomes the style of the AI, and therefore predictable and boring.
Even if artificial intelligences can write novels (cf.https://winteriscoming.net/2018/01/10/artificial-intelligence-writes-the-winds-of-winter-funny-game-of-thrones/), this does not mean that they are creative. They may add surprising twists and turns, but they hardly ever make sense. Humans certainly love surprises, but these have to be somewhat logical. And only human beings can actually do this.
Google will certainly accept texts written by an AI as long as users read them. But if the number of automatically written articles increases, more and more of their weaknesses will become apparent. The readers will bounce off. AI texts can also cause a new wave of spam. Therefore, it cannot be excluded that Google will “search for AI texts” in the future. Then AI texts would no longer be SEO in the first place.
If you are not afraid of the effort these texts require, then AI texts can actually get you to the top of Google. But with good authors it is much faster and above all cheaper. You also avoid scenarios in which AI texts are used as spam and Google takes action against them.
Furthermore, good online copywriters write more vividly than any AI, which is important because readers also want to be entertained. But above all, good writers know what they are writing about. AI only imitates knowledge. It does not understand content.
No related posts.
Dieser Artikel wurde am 07.May 2020 von Thomas geschrieben.
Leave a Reply
E-Mail (erforderlich wird aber nicht veröffentlicht)
We are using cookies to give you the best experience on our website.
Find further information in our data protection policy. Change cookie settings.
Cookies are small text files that are cached when you visit a website to make the user experience more efficient.
We are allowed to store cookies on your device if they are absolutely necessary for the operation of the site. For all other cookies we need your consent.
You can at any time change or withdraw your consent from the Cookie Declaration on our website. Find the link to your settings in our footer.
Necessary cookies help make a website usable by enabling basic functions like page navigation and access to secure areas of the website. The website cannot properly without these cookies.
If you disable this cookie, we will not be able to save your preferences. This means that every time you visit this website you will need to enable or disable cookies again.
Any cookies that may not be particularly necessary for the website to function and is used specifically to collect user personal data via analytics, ads, other embedded contents are termed as additional cookies.
Please enable Strictly Necessary Cookies first so that we can save your preferences!