What are autogenerated texts?
Autogenerated texts (also called machine-generated texts) are texts written in whole or in part by machines. This is not just a distant future scenario, but a practice that has been taking place for several years. Several tools exist for this purpose, and although they are no match for skilled copywriters, they have reached such a quality that they can no longer be ignored.
It takes a long time and costs a lot of money to write good copy - and if you have a large webshop with tens of thousands of product pages and category pages that all need unique copy, it's an almost prohibitive task. So there may be good reasons to use autogenerated texts - but unfortunately the results rarely live up to expectations.
We work with several webshops that have tens of thousands of product numbers, yet we have not made use of autogenerated texts ourselves. For the sole reason that we cannot vouch for the quality. Instead, our approach has been manual copywriting for selected pages - typically prioritised by the number of Google searches and the client's profit margin on each product.
Google's position on autogenerated texts
Google has a set guidelines on autogenerated contentwhich, not surprisingly, states that publishing content designed to manipulate Google search results can have negative consequences. Such content, according to Google, includes:
- texts that make no sense to the reader
- texts translated via an automatic tool without human editing
- texts generated by automated processes
- texts generated by copying RSS feeds or search results
- texts composed of content from different websites.
It goes without saying that texts should make sense to the reader and that it is a bad idea to copy content from RSS feeds, search results and other websites. On the other hand, points 2 and 3 are a grey area, both because it is very difficult for Google to identify this type of text and because the texts are no longer necessarily so bad that they need to be sanctioned.
A simple method for autogeneration of texts is so-called text-spinning (in English called spinning top). The method works simply by a machine automatically creating a number of variations of an original text - including by replacing and swapping words and phrases - so that the variations of the text appear as their own unique text.
Unfortunately, the result is rarely useful. The language is often poor or completely incomprehensible, and the texts are often not varied enough. If you are tempted to copy texts from your competitors, put them through a text spinner and use them on your own website, don't!
The only case where text spinning can be useful is for product variations. For example, suppose you sell a smartphone cover in 10 different colours and each colour variation has its own product page. You could then write one product text and then use a text spinner to generate 10 unique text versions in which you simply replace the name of the colour.
Natural Language Generation
A more advanced method for autogenerating texts is Natural Language Generation. The method is a combination of text spinning and a set of conditional rules that define how specific product information should automatically be used in the texts and how the texts should automatically change according to the product information.
An example of a conditional rule might be that if a product has wireless WiFi, the phrase '[product name] has wireless WiFi with a range of up to [range] metres' is added to the product description. Product name and range are dynamically replaced based on product data. Another example might be that if a product is over a certain size, the adjectives 'large', 'giant' or 'in size XL' are added at specific points in the product description.
The more conditional rules you define based on different product data, the better the texts will be. In practice, you typically create a template (i.e. a collection of rules) for each product category. For example, one template can be used to generate product descriptions for televisions, while another can be used for nails and screws, and a third can be used for razors.
Natural Language Generation works best for product descriptions because they are typically very consistent. Category texts and blog posts should always be written manually. Natural Language Generation therefore does not make copywriters redundant. It also requires a skilled copywriter to set up the templates that are the foundation for autogenerating good texts.
How to get started with Natural Language Generation
There are several tools for Natural Language Generation. Among the most widely used are:
The advantage of tools like the above is not only that new texts can be generated automatically, but also that the autogenerated texts are uniform and have a consistent word choice. At the same time, it is easy to translate the texts into other languages, because you do not have to translate each text individually, but simply translate the templates used to generate the texts.
Despite the many benefits of Natural Language Generation, you should think carefully before you embark on it. It takes a lot of work to set up the templates that can generate good texts, and while the tools have improved, they are nowhere near the level of good copywriters. In many cases, the quality of the texts is low and the texts are not varied enough.