Text Generation in SEO

November 12, 2020 - 2  min reading time - by Rebecca Berbel
Home > Technical SEO > Text Generation in SEO

The idea behind using machines to write text is to free up your resources — human beings! — whose time can be better spent on more sensitive or strategic tasks that require a human touch.

Why generate content for SEO short texts?

In 2020, Artefact found that their SEOs

spend in average 10 hrs per month creating meta descriptions, this can be translated into cost:
30 SEOs x 10 hrs x 80 USD = 24,000 USD/month

Machine learning techniques can use your niche’s or industry’s content to learn the right types of phrasing and vocabulary.

For short texts — like meta descriptions or page titles — machine learning scripts produce extremely high-quality content. This means you can automate the creation of custom, accurate, and quality meta descriptions, titles, product descriptions, and many other types of short texts for your website’s pages.

Once the model is prepared, the cost is negligible, even for creating tens or hundreds of thousands of meta descriptions based on a product name or any other available prompt.

Why generate text for content ideas, briefs, and drafts?

For longer texts, the some models can produce coherent and developed ideas. The content produced by our model will still require human editing work, just like any other rough draft.

This can be a great way to save time and money brainstorming unique content ideas within your specific field or churning out first drafts and content briefs.

The Guardian published an example of what GPT-3 models can do — but the version you can read here was also retouched by a human copyeditor, the same as for articles written by any journalist at The Guardian.

What you need to get text generation up and running

The text generation model you use needs to be re-trained so that the texts it produces are perfectly adapted:

  • to your language
  • to your brand voice and
  • to your industry

You’ll need a large corpus of general texts in your language, as well as specific texts in your industry and on your website–the more the better.

Training a machine learning model can take time, depending on the computer hardware you’re trying to run it on, and the size of your corpus.

You’ll also need a data scientist or a software engineer to run the operations following the steps.

Why use these models for text generation in SEO?

There are many text generation models out there, and there have been for years. Google has repeatedly advised against using low-quality models to create main content.

What’s changed?

First, machine-generated text and NLP (natural language processing, or how machines understand and produce human language) have undergone a revolution in the past few years. Thanks to advances such as those we see in BERT (here’s the technical version) and other, similar models like GPT-2 and GPT-3, the quality we are capable of producing has skyrocketed.

Furthermore, as Vincent Terrasi, the Product Manager at Oncrawl explains in his 2019 TechSEO Boost presentation, you will need an accessible model to produce quality content in any language it is trained on, not just English. Vincent’s research for this model was recognized as one of the top three axes of research in technical SEO.

As with crawl technology, we are convinced that this type of technology is something that everyone should have, not just the biggest players.

Rebecca Berbel See all their articles
Rebecca is the Product Marketing Manager at Oncrawl. Fascinated by NLP and machine models of language in particular, and by systems and how they work in general, Rebecca is never at a loss for technical SEO subjects to get excited about. She believes in evangelizing tech and using data to understand website performance on search engines. She regularly writes articles for the Oncrawl blog.
Related subjects: